Afternoon edition, Thursday 12 November

Jensen Huang hung up on the New York Times and Nvidia declined comment on the report of a second affair, this time with an executive at his own companyzerohedge. Hours later the board was reportedly preparing to oust him, some directors favouring Colette Kress as interimzerohedge. Polymarket put his exit at 69%zerohedge. Underneath the jokes sat the question of whether a founder can be removed for something that isn't a crime, and Balaji made the case for keeping himbalajis. Worth the click: tomselliott scoring all 48,000 laws in US history, back to 1789tomselliott.

A second Huang affair, and Nvidia's board moves to oust him

The Times named her — Kimberly Powell, Nvidia's health care VP — and zerohedge had the line that travelled: "HUANG HUNG UP ON NYT INQUIRY"zerohedge, with Nvidia declining comment. Shares extended their postmarket slide to 4%zerohedge before the WSJ report that the board was preparing to oust himzerohedge, working with an outside firm on possible successorszerohedge and, per Semafor, with some directors favouring CFO Colette Kress as interimzerohedge. The White House, per the Washington Post, isn't interveningzerohedge.

Balaji was the loudest holdout: "No. Bill Clinton did not resign."balajis Ride it out, ignore the journos; if the board does something rash the talent leaves with him, and "Jensen may well land on his feet at TSMC"balajis. udiWertheimer took the same side with less ceremony — "if you thought jensen was a good ceo last month just think how good he will be when he has to prove himself again"udiWertheimer — and later, "let jensen stay"udiWertheimer. Against them, the Economist leader kuhhbin was passing around: "Jensen Huang must go"balajis.

The succession chatter turned into a hiring fair. zerohedge asked Musk to "hire Jensen"zerohedge and got "That would be an honor"elonmusk. theo wondered whether it was all "an elaborate OpenAI recruiting strategy to hire Jensen"theo; Jeff Park asked who actually gains from the exitdgt10011.

Polymarket did the accounting: Jenxit at 69%, "appropriately enough"zerohedge, divorce at 32%Polymarket, with Lori Huang due to become "the 26th richest person in the world, with $73BN net worth"zerohedge. The jokes were mostly below the waist and nobody pretended otherwise. AutismCapital opened with "A SECOND HONEYPOT HAS HIT JENSEN'S PENIS"AutismCapital and closed with "The penis that sunk a thousand chips."AutismCapital; in between came zerohedge asking "Jensen was cheating on the Chinese spy too?"zerohedge, a community note reading "OP embiggened her breasts."AutismCapital, and signüll's imagined chat in which Nvidia legal needs to know whether "pillow talk requires an export license"signulll.

BOOOOOOM *JENSEN HUANG SAID TO HAVE HAD AFFAIR WITH NVIDIA EXECUTIVE: NYT
♥ 4592 · ↻ 250
50 more posts in this story
BOOOOOOM *JENSEN HUANG SAID TO HAVE HAD AFFAIR WITH NVIDIA EXECUTIVE: NYT
quoting @zerohedge
HUANG HUNG UP ON NYT INQUIRY; NVIDIA (NVDA) DECLINED COMMENT HUANG'S ALLEGED RELATIONSHIP WAS WITH NVIDIA (NVDA) HEALTH CARE VP KIMBERLY POWELL
♥ 861 · ↻ 62
BOOOOOOM *JENSEN HUANG SAID TO HAVE HAD AFFAIR WITH NVIDIA EXECUTIVE: NYT
quoting @zerohedge
HUANG HUNG UP ON NYT INQUIRY; NVIDIA (NVDA) DECLINED COMMENT HUANG'S ALLEGED RELATIONSHIP WAS WITH NVIDIA (NVDA) HEALTH CARE VP KIMBERLY POWELL
quoting @zerohedge
Jensen was cheating on the Chinese spy too?
♥ 3304 · ↻ 155
JUST IN: Nvidia CEO Jensen Huang had another affair with an $NVDA executive, NYT reports.
quoting @WatcherGuru
Autism Capital 🧩@AutismCapitalThu 9:14am
BABE WAKE UP A SECOND HONEYPOT HAS HIT JENSEN'S PENIS
A man in a dark suit leans in closely to whisper into the ear of George W. Bush, who looks forward with a stoic expression. Partially obscured text is visible in the background between them.
♥ 1340 · ↻ 68
*NVIDIA SHARES EXTEND DECLINE TO 4% POSTMARKET AFTER NYT REPORT
♥ 195 · ↻ 16
Odds Jensen Huang quits or is fired this year soar to 36%
The image is a screenshot from the prediction market Polymarket showing a graph tracking the probability of Jensen Huang announcing his resignation or removal by December 31, 2026. The chart displays a blue line representing the 'Yes' outcome at 36% and a red line for the 'No' outcome, with trading
text in image
Jensen Huang announced resignation or removal by...? $1,187,323 Vol. | Dec 31, 2026 Polymarket December 31, 2026 $961,875 Vol. 36% ▲ 17% Buy Yes 36¢ Buy No 65¢ Order Book Graph Resolution Maker Rebate + Rewards Yes 36% No 66% Polymarket Nov 9 Nov 10 Nov 11
+1 more in @zerohedge's thread
quoting @zerohedge
And done *NVIDIA PREPARING TO OUST NVIDIA CEO JENSEN HUANG: WSJ
♥ 970 · ↻ 100
Odds Jensen Huang quits or is fired this year soar to 36%
The image is a screenshot from the prediction market Polymarket showing a graph tracking the probability of Jensen Huang announcing his resignation or removal by December 31, 2026. The chart displays a blue line representing the 'Yes' outcome at 36% and a red line for the 'No' outcome, with trading
text in image
Jensen Huang announced resignation or removal by...? $1,187,323 Vol. | Dec 31, 2026 Polymarket December 31, 2026 $961,875 Vol. 36% ▲ 17% Buy Yes 36¢ Buy No 65¢ Order Book Graph Resolution Maker Rebate + Rewards Yes 36% No 66% Polymarket Nov 9 Nov 10 Nov 11
+1 more in @zerohedge's thread
quoting @zerohedge
And done *NVIDIA PREPARING TO OUST NVIDIA CEO JENSEN HUANG: WSJ
quoting @zerohedge
NVDA crashing
A candlestick stock chart for NVDA shows a sharp, near-vertical drop in price occurring late in the trading day, falling from around 174 to below 158. The NVIDIA logo is visible in the top left corner, and the ticker symbol NVDA is labeled in large text near the bottom right.
text in image
NVDA Lo: 156.2 157.58 178 177 176 175 174 173 172 171 170 169 168 167 166 165 164 163 162 161 160 159 158 157 156 155 7:30 7:45 8:00 8:15 8:30 8:45 9:00 9:15 9:30 9:45 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00 16:30 17:00 17:20
♥ 1613 · ↻ 167
Odds Jensen Huang quits or is fired this year soar to 36%
The image is a screenshot from the prediction market Polymarket showing a graph tracking the probability of Jensen Huang announcing his resignation or removal by December 31, 2026. The chart displays a blue line representing the 'Yes' outcome at 36% and a red line for the 'No' outcome, with trading
text in image
Jensen Huang announced resignation or removal by...? $1,187,323 Vol. | Dec 31, 2026 Polymarket December 31, 2026 $961,875 Vol. 36% ▲ 17% Buy Yes 36¢ Buy No 65¢ Order Book Graph Resolution Maker Rebate + Rewards Yes 36% No 66% Polymarket Nov 9 Nov 10 Nov 11
+1 more in @zerohedge's thread
quoting @zerohedge
Jenxit odds soar to, appropriately enough, 69%
A screenshot of a Polymarket prediction market interface showing odds for the question of whether Jensen Huang announced his resignation or removal. A line graph illustrates a sharp spike in the 'Yes' probability to 69% on December 31, 2026.
text in image
Tech · Ceo Jensen Huang announced resignation or removal by...? $1,192,223 Vol. | Dec 31, 2026 Polymarket December 31, 2026 $961,875 Vol. 69% ▲ 17% Buy Yes 69¢ Buy No 32¢ Order Book Graph Resolution Maker Rebate + Rewards Yes 69% No 30% Polymarket 12:00 PM 2:00 PM 4:00 PM 0% 25% 50% 75% 100% $1,191,928 Vol. | Dec 31, 2026 1H 6H 1D 1W 1M ALL
♥ 357 · ↻ 32
Autism Capital 🧩@AutismCapitalThu 9:48am
JENSEN YOUR PENIS DID THIS
A stock chart for NVIDIA displays a current price of $159.88 with a positive daily gain highlighted in green, contrasted by a sharp after-hours drop shown in orange and highlighted with a yellow marker. The line graph illustrates a dramatic vertical plunge at the end of the trading period.
text in image
NVIDIA $159.88 +$4.96 (+2.94%) Today -$13.97 (-8.04%) After-hours
♥ 1464 · ↻ 78
BREAKING: Jensen Huang divorce odds soar following reports of a second alleged affair. 32% chance the Huangs call it quits. 🔗 polymarket.com/r/HuangDivorce
The image shows a smiling woman with short blonde hair and glasses standing next to Jensen Huang, who is wearing a black suit and tie. They are posing in front of a purple backdrop that features partially visible white text reading "REAKTHROUGH" and "EZE".
text in image
REAKTHROUGH EZE
The image is a screenshot of a Polymarket chart tracking the probability of Jensen Huang's divorce by March 31, 2027, which has spiked to 32%. It features a blue line graph showing a sharp increase in odds over the course of a day, alongside trading volume data.
text in image
Trending Combos Perps Breaking Politics Tech AI Jensen Huang divorce by March 31, 2027? 32% chance ▲ 13% Polymarket 32% 25% 20% 15% Source: polymarket.com 8:01 PM 1:21 AM 6:41 AM 12:11 PM 5:45 PM $299,411 Vol. 1H 1D 1W 1M MAX Trading Is Risky. 100% Loss Can Occur. US Users: Featured Market Not Available; Informational Use Only. QTs may include commentary from Polymarket partners.
polymarket.com
https://polymarket.com/r/HuangDivorce
♥ 381 · ↻ 35 · retweeted by @zerohedge
BREAKING: Jensen Huang divorce odds soar following reports of a second alleged affair. 32% chance the Huangs call it quits. 🔗 polymarket.com/r/HuangDivorce
The image shows a smiling woman with short blonde hair and glasses standing next to Jensen Huang, who is wearing a black suit and tie. They are posing in front of a purple backdrop that features partially visible white text reading "REAKTHROUGH" and "EZE".
text in image
REAKTHROUGH EZE
The image is a screenshot of a Polymarket chart tracking the probability of Jensen Huang's divorce by March 31, 2027, which has spiked to 32%. It features a blue line graph showing a sharp increase in odds over the course of a day, alongside trading volume data.
text in image
Trending Combos Perps Breaking Politics Tech AI Jensen Huang divorce by March 31, 2027? 32% chance ▲ 13% Polymarket 32% 25% 20% 15% Source: polymarket.com 8:01 PM 1:21 AM 6:41 AM 12:11 PM 5:45 PM $299,411 Vol. 1H 1D 1W 1M MAX Trading Is Risky. 100% Loss Can Occur. US Users: Featured Market Not Available; Informational Use Only. QTs may include commentary from Polymarket partners.
polymarket.com
https://polymarket.com/r/HuangDivorce
+1 more in @Polymarket's thread
quoting @Polymarket
Lori Huang about to be the 26th richest person in the world, with $73BN net worth
♥ 1118 · ↻ 76
Theo - t3.gg@theoThu 10:07am
What if this is an elaborate OpenAI recruiting strategy to hire Jensen?
♥ 629 · ↻ 9
breaking news: nvidia’s board is reportedly in talks with emmett shear to take over as ceo.
♥ 840 · ↻ 15
The image displays a text-based meme formatted as a chat conversation between Jensen Huang and Mira Murati. It jokes about an alleged extramarital affair by having Jensen Huang ask if 'pillow talk requires an export license.'
text in image
Jensen Huang good or bad? extramarital affair? Mira Murati Directionally very bad Jensen Huang ok Jensen Huang can you wrap up soon? legal needs to know if pillow talk requires an export license. Mira Murati Jensen this is very bad
♥ 484 · ↻ 14
Autism Capital 🧩@AutismCapitalThu 10:35am
“We had a good thing Jensen you son of a bitch. We had Nvidia stock and you had your homely billionaire wife. But you had to blow it all up. You, and your pride, and your ego.”
The image shows a close-up of two older men in an outdoor setting, with one bald man in a green jacket pointing his finger intensely at the other man who has white hair and is wearing a black leather jacket.
♥ 665 · ↻ 56
quoting @zerohedge
*NVIDIA WORKING WITH OUTSIDE FIRM TO FIND POSSIBLE NEW CEO
♥ 1347 · ↻ 96
Autism Capital 🧩@AutismCapitalThu 10:46am
Incredible community note lmao
The image is a screenshot of a tweet identifying NVIDIA executive Kimberly Powell, accompanied by an AI-generated Community Note suggesting the original photo was digitally altered to enlarge her breasts. The note cites an official NVIDIA blog as the source of the unedited image.
text in image
The Wheelie Investor @Wheeliel... · 1h This is Kimberly Powell, the VP at $NVDA who Jensen Huang had his 2nd affair with Readers proposed context on this image Still gathering ratings OP embiggened her breasts. blogs.nvidia.com/wp-content/upl... Do you find this helpful? Rate
♥ 1288 · ↻ 46
Autism Capital 🧩@AutismCapitalThu 11:29am
>Cheats on wife with Chinese spy >"Guys come on, that's just doing business in Asia. Everyone will think your gay if you don't have a mistress. He can't be seen as a loser." >Cheats on wife with random American employee >"Okay he just liked pussy."
♥ 491 · ↻ 14
Kevin Loaec 🧙‍♂️🐟@KLoaecThu 11:35am
Finally moved to Taiwan. Just need to buy a black leather jacket and I'm ready to take the CEO role at @nvidia if Jensen takes a break.
♥ 21 · ↻ 0
quoting @AutismCapital
Autism Capital 🧩@AutismCapitalThu 11:45am
Damn he really was moving like Top J lmao
♥ 33 · ↻ 0
is Nvidia hiring? asking for a friend
text in image
サイオアニメ PSYOPANIME
quoting @PsyopAnime
Autism Capital 🧩@AutismCapitalThu 11:50am
Trump finding out about Jensen 💀💀💀
The image is a still from an anime showing a blond male character in a suit shouting aggressively, with English subtitles overlaid at the bottom.
text in image
Just how many bitches does this guy have?!
♥ 115 · ↻ 6
“it doesn’t matter if nvidia fires jensen” the entire website looks like this btw
The image is a screenshot of the NVIDIA website featuring an AI-generated illustration of a cartoon character resembling Jensen Huang sewing a red lobster plush toy. The page displays the NVIDIA logo, navigation menu, and the title "NVIDIA OpenShell" below the image.
text in image
NVIDIA Sign In US Overview Features Workloads Software Partners Buy Now NVIDIA OpenShell
♥ 36 · ↻ 2
Autism Capital 🧩@AutismCapitalThu 11:56am
What's happened today? The entire feed is the consequences of Jensen's cock. Has anything else happened? Is Trump impeached? Cyberhacks? Anything we missed?
♥ 95 · ↻ 4
Autism Capital 🧩@AutismCapitalThu 11:58am
Remember when life was simple and Jensen only had *one* affair?
♥ 137 · ↻ 5
Autism Capital 🧩@AutismCapitalThu 12:04pm
Want to say that Jensen had a type but so far it seems the type is just woman.
♥ 264 · ↻ 7
*WHITE HOUSE ISN'T INTERVENING IN NVIDIA BOARD'S DECISION: WAPO
♥ 296 · ↻ 27
Autism Capital 🧩@AutismCapitalThu 12:13pm
If Nvidia had developed a Cylon Goonbot none of this would have happened.
The image shows a man resembling Jensen Huang, CEO of Nvidia, standing next to a woman in a red dress, styled like promotional material for the TV show Battlestar Galactica. An NBC Universal logo is visible in the bottom right corner.
text in image
NBC UNIVERSAL
♥ 91 · ↻ 6
Autism Capital 🧩@AutismCapitalThu 12:19pm
It's not Jensen's fault. He was manipulated by the jacket.
A black leather jacket is draped over the edge of a rusty metal dumpster, with yellow subtitle text at the bottom reading "Put me on. Coward!"
text in image
Put me on. Coward!
♥ 129 · ↻ 7
ladies and gentlemen meet NVIDIA’s new CEO
A black-and-white photo shows a man in sunglasses, white shirt, and tie posing outdoors with one hand near his ear and the other on his hip. In the background, a large American flag and a banner reading 'AMERICA' are visible near what appears to be event infrastructure.
text in image
AMERICA
♥ 12 · ↻ 0
Corbin Duncan@kuhhbinThu 5:42am
Jensen Huang must go We argue in @TheEconomist that his position is untenable "Like many powerful leaders, he came to believe that he was too important for the normal rules to apply to him." But he is not the only one. Howard Lutnick has failed to stem the flow of chips to China. More heads should roll
The image is a screenshot of an online article titled "Jensen Huang must resign" with the subtitle "And his should not be the only head to roll." Below the text is a profile photograph of Jensen Huang against a dark background.
text in image
Leaders | Nvidious Jensen Huang must resign And his should not be the only head to roll Save Share PHOTOGRAPH: GETTY IMAGES
+1 more in @kuhhbin's thread
quoting @kuhhbin
No. Bill Clinton did not resign. Jensen Huang shouldn’t either. And the Nvidia board shouldn’t fire him. Particularly in the heat of the moment. Ride it out and ignore the journos. Clinton did. And then no one cared.
This is a screenshot of a The Washington Post opinion article header by Robert J. Samuelson with the headline "WHY CLINTON SHOULD STAY".
text in image
The Washington Post Sign in Opinion Robert J. Samuelson WHY CLINTON SHOULD STAY
Btw: if Nvidia is actually silly enough to fall for this op, and somehow terminates the founder and CEO of perhaps the most valuable company in world history, Jensen may well land on his feet at TSMC. A win for Asia, a loss for America.
Alternatively, Jensen joins Elon! If the Nvidia board is silly enough to do something rash, they will likely face immediate market consequences as all kinds of talent leaves to join Jensen at his next thing. Journalist cancellation cannot stop technologist acceleration.
♥ 1801 · ↻ 179
Autism Capital 🧩@AutismCapitalThu 12:29pm
The Jacket had a life of its own. It's like The Green Goblin mask. Or the Venom symbiote. Or the SPAWN suit. Or the Berserker armor. Or the Kill la Kill underwear or something. Jensen's mind was corrupted.
♥ 30 · ↻ 1
Autism Capital 🧩@AutismCapitalThu 12:34pm
Elon x Jensen arc?
A screenshot showing a tweet from zerohedge asking Elon to hire Jensen, followed by a reply from Elon Musk saying that would be an honor. The post is shared by AutismCapital with the caption 'Elon x Jensen arc?'.
text in image
zerohedge @zerohedge · 2h Elon please hire Jensen 333 627 10K Elon Musk X @elonmusk · 10m That would be an honor 450 554 4K
♥ 744 · ↻ 27
Corbin Duncan@kuhhbinThu 5:42am
Jensen Huang must go We argue in @TheEconomist that his position is untenable "Like many powerful leaders, he came to believe that he was too important for the normal rules to apply to him." But he is not the only one. Howard Lutnick has failed to stem the flow of chips to China. More heads should roll
The image is a screenshot of an online article titled "Jensen Huang must resign" with the subtitle "And his should not be the only head to roll." Below the text is a profile photograph of Jensen Huang against a dark background.
text in image
Leaders | Nvidious Jensen Huang must resign And his should not be the only head to roll Save Share PHOTOGRAPH: GETTY IMAGES
+1 more in @kuhhbin's thread
quoting @kuhhbin
No. Bill Clinton did not resign. Jensen Huang shouldn’t either. And the Nvidia board shouldn’t fire him. Particularly in the heat of the moment. Ride it out and ignore the journos. Clinton did. And then no one cared.
This is a screenshot of a The Washington Post opinion article header by Robert J. Samuelson with the headline "WHY CLINTON SHOULD STAY".
text in image
The Washington Post Sign in Opinion Robert J. Samuelson WHY CLINTON SHOULD STAY
+1 more in @balajis's thread
Btw: if Nvidia is actually silly enough to fall for this op, and somehow terminates the founder and CEO of perhaps the most valuable company in world history, Jensen may well land on his feet at TSMC. A win for Asia, a loss for America.
Thu 12:29pm
quoting @balajis
if you thought jensen was a good ceo last month just think how good he will be when he has to prove himself again
♥ 28 · ↻ 0
quoting @zerohedge
Odds Jensen joins SPCX rising
A screenshot from Polymarket showing a prediction market graph titled 'Will Jensen Huang join SpaceX by December 31, 2026?' with the odds currently at a 9% chance. The blue line chart illustrates a sharp rise in probability over a short period between 7:51 PM and 8:29 PM.
text in image
Trending Combos Perps Breaking New Politics Tech · AI Will Jensen Huang join SpaceX by December 31, 2026? 9% chance ▲ 2% Polymarket 12% 10% 8% 6% Source: polymarket.com 7:51 PM 8:09 PM 8:29 PM 4% Trading Is Risky. 100% Loss Can Occur. US Users: Featured Market Not Available; Informational Use Only.
♥ 646 · ↻ 40
quoting @zerohedge
Odds Jensen joins SPCX rising
A screenshot from Polymarket showing a prediction market graph titled 'Will Jensen Huang join SpaceX by December 31, 2026?' with the odds currently at a 9% chance. The blue line chart illustrates a sharp rise in probability over a short period between 7:51 PM and 8:29 PM.
text in image
Trending Combos Perps Breaking New Politics Tech · AI Will Jensen Huang join SpaceX by December 31, 2026? 9% chance ▲ 2% Polymarket 12% 10% 8% 6% Source: polymarket.com 7:51 PM 8:09 PM 8:29 PM 4% Trading Is Risky. 100% Loss Can Occur. US Users: Featured Market Not Available; Informational Use Only.
+1 more in @zerohedge's thread
quoting @zerohedge
*SOME NVIDIA DIRECTORS FAVOR CFO KRESS AS INTERIM CEO: SEMAFOR
♥ 110 · ↻ 18
There is really only question to get to the bottom of this— Who has the most to gain from Jensen Huang’s exit?
♥ 113 · ↻ 4
just because you cuda, doesn’t mean you shoulda.
♥ 97 · ↻ 2
Autism Capital 🧩@AutismCapitalThu 1:15pm
1:1 scale accurate
A line drawing depicts a massive, complex structure labeled 'ENTIRE GLOBAL ECONOMY' resting precariously on a single small component identified as 'JENSEN HUANG'S WIENER'.
text in image
ENTIRE GLOBAL ECONOMY JENSEN HUANG'S WIENER
♥ 105 · ↻ 7
quoting @semaforben
Is there a leaker in the boardroom too?
♥ 187 · ↻ 17
Matthias Schmidt@eurofounderThu 10:27am
Jensen Huang entering the Nvidia office
♥ 15.2k · ↻ 1002 · retweeted by @BitPaine

Mocking the USCIRF's profile of a detained Indian Christian with a Wodehouse pastiche.

quoting @USCIRF
pagliacci the hated 🌝@SlatzismThu 10:05am
“Seemingly unconcerned with the oncoming Hogwarts Express, Jeet Singh curried along the tracks subcontinentally.”
A woman with reddish hair and glasses sits in a floral armchair writing in a notebook at a dark table near a window. The room features heavy gold curtains and an ornate picture frame on the wall.
♥ 3216 · ↻ 158

Meta disabled Russell Brand's Instagram and Facebook accounts.

There never was a Russell Brand, according to @Meta. Thank God for @elonmusk. Stay tuned.
The image is a screenshot of an Instagram notification stating that the account 'russellbrand' has been disabled for not following Community Standards. It includes standard platform text explaining that the decision is final and offers an option to download information before permanent deletion.
text in image
We disabled your account You no longer have access to russellbrand Account disabled on November 11, 2026 Why this happened Your account, or activity on it, doesn't follow our Community Standards. What this means No one can see or find your account and you can't use it. All your information will be permanently deleted. You cannot request another review of this decision. What you can do You can download your information so that you have a copy of what you shared on Instagram. More information How we made this decision Log out
The image is a screenshot of an automated notification from Instagram explaining that a user's account was restricted because technology detected rule violations. It emphasizes that these rules are applied uniformly worldwide and no further review is possible.
text in image
How we made this decision Our technology found your account, or activity on it, doesn’t follow our rules. As a result, a member of our team took action. We use the same rules around the world, and for everyone on Instagram. you cannot request another review of this decision.
The image is a screenshot of Russell Brand's suspended social media page, showing his profile header and a notification stating the page was suspended for going against community standards. The post by @rustyrockets uses this to claim that Meta denies Russell Brand ever existed, contrasting it with
text in image
Share a thought... Russell Brand 6.2M followers · 27 following WATCH me LIVE Weekdays LIVE on Rumble 10am PST / 1pm EST / 6pm BST https://rumble.com/russellbrand Comedian All About Reels Photos Followers More We suspended your Page This is because Russell Brand goes against our Community Standards.
♥ 15.3k · ↻ 1303 · retweeted by @zerohedge
quoting @AlexHollings52
The funniest way America has dealt with Chinese spies is by putting them in a real version of the Truman Show.
♥ 6549 · ↻ 196

Amnesty called on Bukele to respect human rights in Haiti; Crémieux's answer.

Amnesty International USA@amnestyusaWed 6:05am
The Prime Minister of Haiti, Alix Didier Fils-Aimé, and the President of El Salvador, Nayib Bukele, must ensure that all operations conducted by Salvadoran forces in Haiti comply with international human rights law. bit.ly/3UOTZLU
www.amnestyusa.org
Haiti: Haiti and El Salvador Governments Must Ensure Salvadoran Military Personnel Respect Human...
Haiti's Prime Minister and El Salvador's President must ensure that all operations by Salvadoran forces in Haiti comply with international human rights law.
Bernard Stanford ✡︎@stanfordNYCThu 2:39am
"This American Life" finds it totally unremarkable when a black interviewee says (of blacks and Arab-Americans like El-Sayed) "We need to band against white people, you know, my opinion. Because it's us against them." This is her pitch to "talk [people] out of their prejudice."
The image displays a transcript excerpt featuring dialogue attributed to Rose Johnson and narrator Ike Sriskandarajah. A specific sentence where Johnson states 'We need to band against white people' is highlighted in blue within the text block.
text in image
Rose Johnson They get businesses. They get stores. They get loans. They get a lot of things that we didn't get coming into this country as African-Americans, and they just feel like they're entitled. Just don't come here and act like you're better than us, you're privileged. We get that enough from white people. So we don't need that from others. We need to band against white people, you know, my opinion. Because it's us against them. That's my opinion. Ike Sriskandarajah I mean, she says that with people she knows, she's been able to talk them out of their prejudice and into a vote for El-Sayed. But there's this one person in her life. You heard her talking about her earlier, one of her oldest friends, and she's been making the case to her for weeks. And I asked if I could listen in and record their conversation.
♥ 1459 · ↻ 137 · retweeted by @herandrews
The Anarchist Turtle 🏴Ⓐ🔆🏳️‍⚧️@anarchoboognishWed 11:10am
A meme depicting a modified image of Goku from Dragon Ball Z standing in the rain next to Batman. The speech bubbles show Goku addressing Batman as 'Woku' while Batman responds with a sociological statement about wealth inequality.
text in image
THIS IS WHERE I WATCHED MY PARENTS DIE, Woku. Wealth inequality contributes to crime, Batman.
+1 more in @anarchoboognish's thread
Social Security, for the second time this year, has decided they want to take my social security away for two due to "overpayment". I need my social security. Please help ensure I don't end up on the streets. givebutter.com/keep-a-disabled-trans-woman-with-
givebutter.com
Help Secure Permenant housing for Disabled Trans Womem
Help Alex, who has been through hell, finally get into a secure Permenant residence and get her out of the overexpense and dangerous place she is in.
Wed 3:15pm
quoting @anarchoboognish
No: Changes in inequality are unrelated to changes in the homicide rate (or other types of crime, for that matter).
A scatter plot displays the relationship between the change in GINI percentage on the x-axis and the change in homicide percentage on the y-axis. A text box in the bottom left corner indicates a sample size of 1629 with a correlation coefficient (r) of 0.02 and a p-value of 0.532, while a flat blue
text in image
6000 4000 2000 0 -2000 -4000 -6000 Change Homicide_%_ n = 1629 r = 0.02; P = 0.532 -30 -20 -10 -10 0 10 20 Change GINI_%_
+1 more in @cremieuxrecueil's thread
More here:
Wed 3:37pm
quoting @cremieuxrecueil
Cosmopolitan Reactionary@cosmorxnWed 5:56pm
The safest cities in the world are essentially gilded aristocracies built on the labour of a migrant underclass. The “inequality causes crime” people need to just admit defeat.
A screenshot of a Numbeo ranking table listing the world's safest cities, with the top five positions occupied by cities in the United Arab Emirates. The list includes columns for rank, city name, and safety index score.
text in image
NUMBEO Select City Rank City Safety Index 1 Ajman, United Arab Emirates 88.8 2 Abu Dhabi, United Arab Emirates 88.6 3 Ras al-Khaimah, United Arab Emirates 85.4 4 Sharjah, United Arab Emirates 84.7 5 Doha, Qatar 84.6 6 Dubai, United Arab Emirates 83.8 7 Taipei, Taiwan 83.4 8 Muscat, Oman 81.4 9 The Hague (Den Haag), Netherlands 78.9 10 Zagreb, Croatia 78.4
♥ 1370 · ↻ 113 · retweeted by @cremieuxrecueil
Berlin alone records almost as many assaults daily as all of Japan.
+2 more in @argonerd's thread
Additionally, the quality of violence: In Japan, it often remains at lighter assaults, while in Berlin serious and armed violence weighs significantly more.
Wed 6:31pm
Crazy addition: Japan statistically distinguishes between violence with and without injury. If you only take cases with injury, then in Berlin more than twice as many assaults are committed daily as in all of Japan.
Wed 9:48pm
quoting @argonerd
Misinformation. Correction: Berlin alone records more than twice as many assaults daily as all of Japan.
quoting @argonerd
"Berlin alone records more than twice as many assaults per day as all of Japan." I just got back to America about 15 minutes ago and I already miss Japan.
♥ 1115 · ↻ 48
A classic study by MIT researchers (Sadr et al., 2003) found that the eyebrows are fundamental for face recognition, even more so than the eyes themselves. What this points to is that changing your eyebrow aesthetics can completely transform your facial appearance.
The image displays a grid of six facial close-ups showing original and altered versions of Richard M. Nixon and Winona Ryder to demonstrate the role of eyebrows in recognition. The caption below identifies the subjects and explains that some versions lack eyebrows or eyes.
text in image
Figure 4. Sample stimuli from our experiment assessing the contribution of eyebrows to face recognition: original images of President Richard M Nixon and actor Winona Ryder, along with modified versions lacking either eyebrows or eyes.
quoting @qovesstudio
I realized this as a kid and started telling everyone to focus on people's eyebrows when you're talking to them. Do it. You'll see a different person.
♥ 1136 · ↻ 30
The honeybee’s “dance” language was only decoded in 1946. Humans have kept bees for thousands of years but Karl von Frisch achieved a breakthrough in our understanding of them less than a century ago. It’s good to be reminded that there is still much to discover about nature. The real frontier is not on a computer screen. Also, this may be the German autism talking but spending hours every day in silent contemplation of the beehive sounds like the perfect life.
An elderly man with white hair and glasses, wearing traditional embroidered clothing, leans over a small wooden stool to observe a honeybee on a yellow board. A glass jar sits nearby on the stool as he studies the insect in an outdoor setting.
The image displays three scientific diagrams illustrating the honeybee's dance language, including a bee dancing on a comb, a graph plotting waggle run duration against distance, and a schematic showing how the dance angle relates to the sun and a food source.
text in image
40° Duration of waggle run (sec) 4.0 3.0 2.0 1.0 1000 3000 5000 Distance to food source (m) 40°
“Honeybee Democracy” is a great book (just read it, hence these bee thoughts), but WHAT A COINCIDENCE that the scientific breakthroughs of the 20th century proved that beehives are a democracy that confers and deliberates, whereas the first glass beehives in the 17th century, which allowed observation of a hive’s inner workings for the first time, revealed that bees were a well-ordered monarchy. What were the odds?
The image shows the cover of the book Honeybee Democracy by Thomas D. Seeley, featuring a white background with illustrations of honeybees swarming around the title and a quote from Edward O. Wilson.
text in image
"A wonderful book about humanity's greatest friend among the insects..." —Edward O. Wilson HONEYBEE DEMOCRACY THOMAS D. SEELEY
The image shows a historical diagram titled 'The Commonwealth of Bees,' depicting an octagonal glass beehive with labeled parts and explanatory text in archaic English. It illustrates an early design intended for observing bee behavior from the outside.
text in image
The Commonwealth of Bees. ab. ab. g Outgoing / Henry meeting in all par, / hindrance of your Majest. / bees. / A Hole in the top ... to let the same in every Bee. / B. Down this place, to / every hole turning a / little way, 'twill / e'f a Wind, we'd put the / Cover to close it upon / occassion. / C.C. holes throug' at the / Sides of the upper case. / hhh. the Design, every one / to be opened as they like / dislike where the lower/ doors are open, the other / Ikk. At the upper end of / every case, there must be / commonly, the bee-hive / be likewise placed a way / conveniently, that any one / may fit to any of the other / Hives. / A little Key made to serve / on the side of the space / 't appears on the sides, / by that means to clean / about the honey. / In the under bottom oppo / site to ye the false floor / are divers that open with / handles to look through & / take out any way is to the / other, &c without any / piece of clear glasse to / be connected to the inside / of the hive, to look in / upon occasion; / which hive is lined with / felt &c. well. / It Standeth in a Cist of here / (which should be) 2 or 3 / Inches deep, & has a Covering to it.
The image shows the title page and an engraved illustration from the 1609 book 'The Feminine Monarchie or The Historie of Bees' by Charles Butler, featuring bees arranged in a hive structure labeled with Latin phrases about industry.
text in image
THE Feminine Monarchie:: OR THE HISTORIE OF BEES. SHEWING Their admirable Nature, and Properties, Their Generation, and Colonies, Their Gouvernement, Loyaltie, Art, Industrie, Enemies, Warres, Magnanimite, &c. TOGETHER With the right ordering of them from time to time : And the sweet profit arising thereof. Written out of Experience by CHARLES BUTLER Magd: LONDON, Printed by JOHN HAVLAND for Roger Jackson, and are to be sold in his Shop in Fleetstreet, over against the Conduit. 1609.
♥ 97 · ↻ 11
"We have frontier at home." (GLM-5.3 Flash running on a pair of Sparks with Omarchy being accessed remotely via Tailscale from a MacBook Neo.)
A screenshot of a terminal-based AI coding interface showing the GLM-5.3 Flash model generating text about the Roman Empire, with a hardware and software specification overlay in the bottom right corner identifying an Apple MacBook Neo running Omarych.
text in image
1 3 4 5 Wednesday 20:03 Give me five paragraphs on the fall of the roman empire The user wants a writing task, not coding. Just write five paragraphs. The Fall of the Roman Empire The decline of the Western Roman Empire was not a single event but a gradual process spanning roughly two centuries... (truncated) Agents 20.9M TOKENS THIS WEEK Claude Code Max 28x Weekly 6d 15h Codex Pro 500 Weekly 6d 21h 2 free resets - next expires in 19d 8h Z GLM-5.3 Flash cph-spark1 35 TPS - 256K Context MAKE SOMETHING COOL Theme Plugin App Hardware PC: Apple MacBook Neo Apple A18 Pro (2+4) @ 4.04 GHz Apple A18 Pro (G17P B1) 2408x1506 @ 2x in 13", 60 Hz [Built-in] 18.57 GiB / 186.55 GiB (10%) - btrfs 3.95 GiB / 7.40 GiB (53%) 319.55 MiB / 7.40 GiB (4%) Software Omarych dev (e485c6224) quattro edge ~/Work ↑2.1k ↓836 1.1%/262k (auto)
♥ 912 · ↻ 46
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
text in image
Fable 5.2
+9 more in @claudeai's thread
Fable 5.2 and Mythos 5.2 performed as well as, or better than, our recent models on most alignment measures. The model was also tested before release by external evaluators, including the @AISecurityInst, @FrontierDesign2, and @METR_Evals.
Tue 5:58am
Fable 5.2 and Mythos 5.2 are a step above Opus 5.5 in almost all the capabilities we evaluated. Testers have found that they communicate more clearly than past Fable and Mythos models. They generate outputs up to 30% faster than Fable 5.1 and Mythos 5.1.
A data table titled 'Claude Fable 5.2' comparing benchmark scores across categories like Agentic coding, Cybersecurity, and Robotics against models like Fable 5.1, Opus 5.5, GPT-6 Astra, and Gemini 4 Carbon. The column for Fable 5.2 is highlighted in green to emphasize its performance relative to co
text in image
Claude Fable 5.2 Fable 5.2 | Fable 5.1 | Opus 5.5 | GPT-6 Astra | Gemini 4 Carbon Agentic coding FrontierCode 2.0: 49.1% (high effort) | 32.4% (high effort) | 45.5% | 42.8% | 41.1% (Medium effort) Agentic coding FrontierSWE v2: 69.9% | 56.3% | 62.3% | 65.5% | 61.2% Business workflows AutomationBench: 51.4% | 31.4% | 42.5% | 41.4% | 53.9% Agentic scientific research Terminal-Bench-Science 0.1: 74.6% (75.3% Mythos 5.2) | 52.6% | 58.7% | 68.1% | 64.8% Cybersecurity CyScenarioBench: 76.3% (Mythos 5.2) | 61.7% (Mythos 5.1) | 67.6% (without safeguards) | 61% (GPT-6 Cyber) | — Cybersecurity CVE Rediscovery: 89.6% (Mythos 5.2) | 78.3% (Mythos 5.1) | 81.3% (without safeguards) | — | — Protein design De novo binder design (24 hr): 85.1% (Mythos 5.2) | 79.1% (Mythos 5.1) | 82.6% (without safeguards) | — | — Agentic search WANDR (relaxed score): 82.1% | 68.7% | 72.3% | — | — Computer use OSWorld 2.1 (offline set, partial): 89.3% | 83.5% | 87.2% | 79.5% | 81.5% Robotics RoboDojo Sim: 64.9% | 27.6% | 39.4% | 29.8% | — Knowledge work AA-Briefcase v1.1: 66.1% (rubric) | 61.5% (rubric) | 61.2% (rubric) | 52.0% (rubric) | 61.8% (rubric) Elo Scores: 1785 (Fable 5.2) | 1669 (Fable 5.1) | 1800 (Opus 5.5) | 1569 (GPT-6 Astra) | 1611 (Gemini 4 Carbon) Knowledge work Remote Labor Index (RLI): 41.3% (High effort) | 17.9% | 21.3% | 20.8% | — Multidisciplinary reasoning Humanity's Last Exam (with tools): 74.2% | 65.6% | 67.7% | 57.2% | — Visual chart recognition Chartography (no tools): 73.4% | 44.8% | 64.4% | 71.0% | 72.5% For details on evaluation methodology, see the Fable 5.2 and Mythos 5.2 System Card.
Tue 5:58am
In early testing on FrontierCode 2.0, Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort, for roughly half the cost per task. At Xhigh effort, it scored the highest of any released model tested. Medium is the default in Claude apps and Claude Code.
A scatter plot titled Agentic coding by effort level compares the performance and cost of different AI models, highlighting FrontierCode 2.0's Fable 5.2 against competitors like Opus 5.5 and GPT-6 Astra. The chart plots Score (%) on the y-axis against Cost per task (USD, log scale) on the x-axis to
text in image
Agentic coding by effort level FrontierCode 2.0 Fable 5.2 Opus 5.5 Fable 5.1 GPT-6 Astra Gemini 4 Carbon 50 Xhigh High 40 Med Low 30 20 10 0 Score (%) 1 2 5 10 20 50 Cost per task (USD, log scale) Max
A line chart titled 'Agentic search by effort level' plots Score (%) against Cost per attempt on a log scale for three models: Fable 5.2, Opus 5.5, and Fable 5.1. The Fable 5.2 series is annotated with effort levels Low, Med, High, Xhigh, and Max, showing it achieving the highest score at a lower co
text in image
Agentic search by effort level WANDR (relaxed score) Fable 5.2 Opus 5.5 Fable 5.1 90 80 70 60 50 40 30 0 Score (%) Low Med High Xhigh Max 1 2 5 10 20 50 Cost per attempt (USD, log scale)
A line chart titled 'Hardware engineering by effort level' plots model performance scores against cost per attempt on a logarithmic scale. It compares Fable 5.2, Opus 5.5, and Fable 5.1 across effort levels ranging from Low to Max.
text in image
Hardware engineering by effort level BenchCAD 2.1 (with tools) Fable 5.2 Opus 5.5 Fable 5.1 Score (0-100) 90 80 70 60 50 40 30 20 10 0 Max Xhigh High Med Low Cost per attempt (USD, log scale) 0.05 0.1 0.2 0.5 1 2 5 10 20 50
A line chart titled 'Multidisciplinary reasoning by effort level' plots Score (%) against Cost per attempt for Fable 5.2, Opus 5.5, and Fable 5.1. The green line representing Fable 5.2 shows the highest scores at effort levels ranging from Low to Max, surpassing the orange Opus 5.5 line.
text in image
Multidisciplinary reasoning by effort level Humanity's Last Exam (with tools) Fable 5.2 Opus 5.5 Fable 5.1 80 70 60 50 0 Low Med High Xhigh Max Score (%) Cost per attempt (USD, log scale) GPT-6 Astra (Max): 57.2%
Tue 5:58am
Gene therapies can be delivered in a protein shell called a capsid. Mythos 5.2 found a new way to predict a whole capsid’s structure. In our tests, it accurately modeled 28 of the 34 capsids it attempted. Biologists could use it to design capsids faster and more cheaply.
Tue 5:58am
A team of 1,000 Fable 5.2 agents fully formalized Fermat’s Last Theorem in 30 hours, translating the proof into a language a computer can check step by step. With the same setup, Opus 5.5 completed only 26% in 18 hours, and barely made any more progress after 30 hours.
A line chart titled 'Fermat's Last Theorem proof formalization' plots completion over time for two models, Fable 5.2 and Opus 5.5. The green line for Fable 5.2 rises steadily to a full completion of 1.00, while the orange line for Opus 5.5 plateaus early at roughly 0.27.
text in image
Fermat’s Last Theorem proof formalization Completion over time Fable 5.2 Opus 5.5 Resumed with 12 more hours 1.00 0.86 Completion (0-1, weighted share of statements proved) 0.26 0.27 Time elapsed (hours)
Tue 5:58am
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do. On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.
text in image
Claude solves a puzzle cube 14:04:44
Tue 5:58am
Fable and Mythos are best for the most ambitious, open-ended technical work. For most coding and knowledge tasks, we recommend starting with Opus 5.5, our strongest daily driver, then moving to a different model depending on the scope and complexity of your work.
Tue 5:58am
Fable and Mythos are the same underlying model. Fable has strong safeguards to guard against misuse. Organizations pursuing beneficial applications in cybersecurity or biomedical research can apply for Mythos through our Life Sciences and Cyber Verification Programs.
Tue 5:58am
Claude Fable 5.2 is available today on all paid plans, on the Claude Platform, and on all major cloud platforms. It’s $8 per million input tokens and $40 per million output tokens, down from $10 and $50. Read more: anthropic.com/claude-fable-and-mythos-5-2
www.anthropic.com
Introducing Claude Fable 5.2 and Claude Mythos 5.2
Claude Fable 5.2 and Claude Mythos 5.2, our new models for the hardest technical challenges. Fable 5.2 sets a new bar in coding, finance, and the sciences.
Tue 5:58am
quoting @claudeai
Big news: Claude Fable 5.2 (Max) by @AnthropicAI takes #1 in Agent Arena with an unprecedented lead: +30.4% net improvement, about 1.6x the #2 model Opus-5.5, across 4.7k real-world agentic sessions from our global community. It's also #1 in Code, Chat and Work. A major leap over Claude Fable 5.1 (Max) - Rank: #5 → #1, Net improvement: +14.8% → +30.4% - Median cost per task: $7.7 → $7.3 (6% lower). The API pricing $10/$50 → $8/$40 per million input/output tokens (20% lower) vs. The Frontiers - #1 Claude Fable 5.1 (Max): +30.4% | $7.3/task - #2 Claude Opus 5.5 (Max):+19.5%) | $5.5/task - #3 Gemini 4 Carbon (High):+17.0%) | $1.5/task - #6 GPT-6-Astra (Max): +14.5% | $6.0/task Breaking down performance across signals: Fable 5.2 improves over 5.1 across all four reported signals, with particularly large gains in user satisfaction and steerability. - #1 Praise vs. Complaint , +73%, biggest jump vs +41% the next model. - #1 Steerability: +33% - #1 Confirmed Success: +22% - #3 Bash Reliability: +14% #1 across all major categories - #1 Code: +34% - #1 Work: +31% - #1 Chat: +27% - Versus Fable 5.1: Code #6 → #1 (+19.8 %), Work #4 → #1 (+15.8 %), Chat #6 → #1 (+12.0%). Huge congrats to the @AnthropicAI team on this incredible release!
A horizontal bar chart from Arena AI ranks various AI models by net improvement, with Claude Fable 5.2 (Max) highlighted at the top with a +30.4% lead. The graphic lists fifteen models including Opus 5.5, Gemini 4 Carbon, and GPT-6 variants below the winner.
text in image
Agent Arena Claude Fable 5.2 (Max): #1 Arena ARENA.AI 1 Claude Fable 5.2 (Max) +30.4% 2 Claude Opus 5.5 (Max) +19.5% 3 Gemini 4 Carbon (High) +17.0% 4 Claude Sonnet 5.5 (Max) +16.0% 5 Claude Fable 5.1 (Max) +14.8% 6 GPT-6 Astra (Max) +14.5% 7 GPT-6.1 Sol (Max) +12.5% 8 GPT-6.1 Terra (Max) +12.0% 9 Gemini 4 Argon (High) +10.7% 10 Muse Wonder Full (Max) +9.0% 11 Claude Fable 5 (High) +7.3% 12 Claude Opus 5 (Max) +7.2% 13 GPT-6 Sol (Max) +6.9% 14 Hy4 +5.3% 15 Kimi K3.1 Pro (Max) +4.9% SOURCE: ARENA AI LEADERBOARD (ARENA.AI/LEADERBOARD/AGENT) 0% +10% +20% +30% +40% NET IMPROVEMENT VS BASELINE (%)
quoting @arena
Theo - t3.gg@theoThu 7:05am
Anthropic's current dominance compared to other labs is insane. IMO this is the biggest lead any lab has had since GPT-3.
♥ 1104 · ↻ 28

2 posts from @theo

Gemini 4 Carbon, now available in the Gemini API and to Ultra subscribers! Just a few weeks after Argon was launched, this model makes significant progress on lots of the feedback from Argon and pushes the frontier in cyber, knowledge work, and code! It's a great model : )
+1 more in @OfficialLoganK's thread
This model has frontier level cyber capabilities and so the team has been hard at work making sure it could be safely deployed to both cyber defenders and broadly, so thank you for the patience! And keep the feedback coming, we will keep shipping! blog.google/innovation-and-ai/models-and-researc
blog.google
Gemini 4 Carbon is now available
Today, we’re making Gemini 4 Carbon available to developers, enterprises, scientists, and consumers.
Wed 3:07am
quoting @OfficialLoganK
Theo - t3.gg@theoThu 6:17am
Gemini 4 Carbon is here. It's slow, awkward, stupid, and hallucinates all the time. That said, they made a ton of progress and the team is working hard to make it great. There's also some surprisingly useful capabilities hidden deep if you dig 👀
quoting @theo
Theo - t3.gg@theoThu 6:20am
I don't really know how to properly emphasize this. Gemini 4 Carbon is unacceptably slow in real world usage. Gemini 4 Carbon low: 22m21s Opus 5.5 high: 9m33s
A scatter plot titled 'Time per task vs intelligence' compares various AI models, highlighting Gemini 4 Carbon with a green data point and tooltip showing a time of 22m 21s.
text in image
Time per task vs intelligence Intelligence index Pareto line Linear Log Fable 5.2 Gemini 4 Carbon Opus 5.5 GPT-6 Astra GPT-6.1 Sol GPT-6.1 Terra Haiku 5.5 GPT-6 Luna Gemini 4 Carbon low · Pinned Intelligence index 54.2 Time per task 22m 21s Cost per task $7.19 Artificial Analysis Time per task (lower is better) 0s 8.3m 16.7m 25m
I had it write a fix for T3 Code Windows installations, and it took FOURTEEN HOURS to do it 🙃
♥ 626 · ↻ 7
quoting @theo
Theo - t3.gg@theoThu 7:44am
I dug more into Gemini’s speed issues. Almost all of it is server side. Their inference is awful. It takes 11 seconds before it starts to respond, 20x longer than Fable.
A data visualization chart comparing the latency and response times of different AI models, highlighting that Gemini 4 Carbon has a significantly longer initial wait time than competitors like Fable and Opus. The graphic uses horizontal bar charts to break down the seconds spent waiting versus writi
text in image
GEMINI 4 CARBON WAITS 11 SECONDS BEFORE EVERY ANSWER. In one coding thread it answered 1,319 times in 8 h 10 min of run time. About 4 of those hours passed before an answer started. It writes as fast as the other models after it starts. RUN ONE STEP IN REAL TIME 0.0 s. GEMINI 4 CARBON Antigravity, this thread 11.3 s 9.5 s 21.6 s. OPUS 5.5 Claude Code, all work 3 s 6.3 s 4.6 s 13.2 s. FABLE 5.2 Claude Code, all work 11.3 s 6.2 s 18.0 s. GPT-6.1 SOL Codex, all work 3.7 s 8.9 s 13.5 s. 0 s 5 s 10 s 15 s 20 s. waits for the answer to start model writes tools run
The image displays two side-by-side data visualizations tracking file access patterns for serviceLauncher.ts and bootService.ts over a period of nearly nine hours. White horizontal bars represent individual reads of specific code lines, while amber lines indicate compaction events that trigger repea
text in image
✨ IT READ THE SAME TWO FILES 16 TIMES OVER Each white mark is one read. Left to right is the line range. Top to bottom is time, 9 h 39 min. Each amber line is a compaction. After a compaction the model reads the files again. serviceLauncher.ts 3,380 lines. 299 reads. Read 16 times over. bootService.ts 2,550 lines. 230 reads. Read 17 times over. line 1 line 3,380 line 1 line 2,550 1,093 file reads in the thread, with a median of 106 lines per read. Each read is one step, so each read pays the wait. The context filled 22 times, and each compaction took about 66 s.
Varun claims this is because “prefill is hard”. It does seem like the other labs figured it out tho, so hopefully they can too 🫡
♥ 475 · ↻ 9
Theo - t3.gg@theoThu 9:31am
Remember open weight models? Those were fun
♥ 848 · ↻ 5
replying to @vikramcantsingh
8 earlier in @vikramcantsingh's thread
A $2.5 trillion sell-off, the Jensen affair, and Anthropic's IPO round out a volatile week for AI stocks. With so much noise in the markets, I did some data science work to examine how fast AI can progress, how cheap models are getting, how smart they can actually get, and more:
The image displays four charts under the title 'A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE,' illustrating trends in AI model pricing, task costs, and human working time equivalents over time. The graphs plot data points for models like GPT-4, Claude Opus 5.5, and o1 against indices of intelligenc
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE PRICE, 1/N OF LAUNCH PRICE (LOG) first price 1/10 1/100 1/1,000 GPT-4 level - 1/5 in year one The law - first price x 2^(-t/2.05) The other levels - 130 to 160 o1 level - 1/477 so far MONTHS SINCE THE LEVEL FIRST WENT ON SALE COST OF ONE TASK ON THE INDEX - LOG $30 $10 $3 $1 $0.3 $0.1 Claude Opus 5.5, max effort - $5.98 MiMo-V2.6-Pro - $0.13 GPT-6 Luna, low effort - $0.005 a task The law - cost x 2 every 4.7 points SCORE ON ARTIFICIAL ANALYSIS'S INTELLIGENCE INDEX TASK LENGTH, IN HUMAN WORKING TIME 1 billion work-years Europe's 209m workers, a year all 3.5bn workers, a year India's 590m workers, a year the US's 170m workers, a year the UK's 34m workers, a year the world's 9m researchers, a year Walmart's 2.1m associates, a year Accenture's 779,000 people, a year 1 million work-years a 10,000-person company, a year 100,000 work-years 10,000 work-years Their own dates: Oct 2027/Apr 2028/Feb to Jan 2030 (Lifeland) 1,000 work-years Superhuman coder - the AI Futures Project's threshold a career, 40 work-years a work-year a work-month 100 work-years 1 work-year a work-week 1 work-week 1 hour 1 minute 1 second 2020 2030 2040 2050 2060 2070 TASK LENGTH, IN HUMAN WORKING TIME 100 million work-years the US's 170m workers, a year The law the UK's 34m workers, a year AI research alone the world's 9m researchers, a year Walmart's 2.1m associates, a year Accenture's 779,000 people, a year 1 million work-years Both a 10,000-person company, a year push from AI research pull from the limits Limits alone 100,000 work-years Superhuman coder - the AI Futures Project's threshold a career, 40 work-years 10,000 work-years a work-year 1,000 work-years a work-month 100 work-years 1 work-year 1 work-week 2027 2029 2031 2033 2035
Wed 5:14am
Users have to pay 2x more per task for using a 8% smarter model
A scatter plot chart titled 'A (Mathematical) Moore's Law for Intelligence' displays the cost of AI tasks against an intelligence score, showing an exponential price increase trend line labeled 'The law'. Specific data points highlight models like GPT-6 Luna, MiMo-V2.6-Pro, and Claude Opus 5.5 with
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE Users have to pay 2x for every 4.7 points in marginal intelligence Cost of one task on Artificial Analysis's index, 170 configurations; the floor runs from $0.005 at score 10 to $5.98 at 57, 1,325x. COST OF ONE TASK ON THE INDEX · LOG $30 $10 $3 $1 30¢ 10¢ 3¢ 1¢ GPT-6 Luna, low effort · $0.005 a task MiMo-V2.6-Pro · $0.13 Claude Opus 5.5, max effort · $5.98 The law · cost x 2 every 4.7 points 10 20 30 40 50 SCORE ON ARTIFICIAL ANALYSIS'S INTELLIGENCE INDEX closed weights open weights the floor Artificial Analysis, cost to run its index, 2 Oct 2026 (timing58) Dashed: the law, fitted to the floor at 48 scores set by 15 configurations (r = 0.96). MiMo-V2.6-Pro is the only open model on the floor cheapest cost per task at a score ≈ $0.0086 × 2^(score−21)/4.7
Wed 5:14am
For every 14 more "intelligence points" gained, Anthropic models make 10x more revenue. OpenAI models make 10x more for every 26 more "intelligence points"
The image is a line chart titled 'A (Mathematical) Moore's Law for Intelligence' plotting annualized revenue run-rate against Frontier ECI scores. It displays trend lines for OpenAI in green and Anthropic in blue, showing that Anthropic generates 10x more revenue for every 14 points of intelligence
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE Revenue grows 10x for every 21 points of ECI Reported run-rates against the frontier ECI on the day: 10x per 26 points (OpenAI) and 14 (Anthropic), 6x a year at the law's pace. ANNUALISED REVENUE RUN-RATE (LOG) $100bn $10bn $1bn $0.1bn OpenAI · 10x per 26 points AI 2027 Project · 10x per 15 points Anthropic · 10x per 14 points Both labs today · $135bn at ECI 169 120 130 140 150 160 170 FRONTIER ECI ON THE REPORT DATE OpenAI Anthropic Reported run-rates (footnotes 27–28); Epoch ECI; AI 2027 Project; timing72
Wed 5:14am
Open source models catch up to frontier models faster than ever before – and the catch up time is collapsing. Today, it takes 31 days for open models to catch up to frontier models.
This line chart titled 'A (Mathematical) Moore's Law for Intelligence' tracks the performance gap between closed and open-weight AI models from Jan '23 to Jul '26. It highlights that the best open model, Kimi K3.1 Pro, is currently only 31 days behind the frontier leader, Claude Opus 5.5.
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE The best open model is 31 days behind the frontier Open advances land a median 297 days behind in 2024, 184 in 2025, 103 in 2026. Best closed · Claude Opus 5.5 · 57.6 Best open · Kimi K3.1 Pro · 55.1, 31 days behind ARTIFICIAL ANALYSIS INTELLIGENCE INDEX Claude Opus 5.5 GPT-6 Astra · 52.7, never led Claude Opus 5 GLM-5.3 Kimi K3 Gemini 3 Pro GPT-5 o1 Kimi K2 Thinking DeepSeek R1 GPT-4 closed model open-weights model 522 scored models with recorded openness, AA 24 Oct 2026; timing-fig52-two-frontiers.py
Wed 5:14am
A level of intelligence halves in price every 2 months in its first year after launch, and is ~60x cheaper by the end of that year. Token prices eventually asymptotes at the cost of serving the token.
This is a line chart titled 'A (Mathematical) Moore's Law for Intelligence' plotting the logarithmic decline in AI model prices over 24 months. It compares a theoretical pricing law against actual data for GPT-4 and o1 levels, showing a steep downward trend.
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE A level of intelligence gets 60x cheaper in its first year Cheapest price clearing each Epoch level: median ×58 in its first 12 months across 6 levels (×5 to ×113); 3 have a second year. PRICE, 1/N OF LAUNCH PRICE (LOG) first price 1/10 1/100 1/1,000 GPT-4 level · 1/5 in year one The law · first price × 2^(-t/2.05) The other levels · 130 to 160 o1 level · 1/477 so far 0 6 12 18 24 MONTHS SINCE THE LEVEL FIRST WENT ON SALE the law · o1 level · other levels · GPT-4 level OpenRouter via Wayback × Epoch ECI, 31 Oct 2026 Band: the other levels' range, month by month. GPT-4 level: 1/278 more in year two, when open models arrived. o1 level: $26.25 → $0.055 per million tokens price after t months ≈ first price × 2^(-t/2.05)
Wed 5:14am
Labs are shipping models faster with every passing year. The gap between flagship releases shrinks about 28% a year at closed labs (12% at open labs), from 142 days in late 2023 to 57 days today.
The image displays a line chart titled 'A (Mathematical) Moore's Law for Intelligence' tracking the decreasing number of days between flagship AI model releases from July 2023 to projected dates in 2026. A formula at the bottom defines the shrinkage rate as 28% for closed labs and 12% for open labs.
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE A lab ships a better flagship model every 57 days A lab's gap between flagships: 142 days in late 2023, 57 today, shrinking about 28% a year. DAYS BETWEEN FLAGSHIPS, PER LAB 200 175 150 125 100 75 50 25 0 Per-lab advance cycle · 57d era mean 96d era mean 66d 7 LABS ACTIVE 8 8 9 10 11 Jul '23 Jan '24 Jul Jan '25 Jul Jan '26 Jul advance tier ÷ active labs; release table, vintage 2026-10-31 Each point is the latest 180-day window gap today = old gap × (1 − s)^years ago, s = 28% (closed labs), 12% (open labs)
Wed 5:14am
Frontier "intelligence" rises linearly, uninterrupted since 2024, by about 1.36 points a month on Epoch's capabilities index (ECI).
A scatter plot chart titled 'A (Mathematical) Moore's Law for Intelligence' displays the Epoch Capabilities Index rising linearly from 2023 to 2026. It features a blue trend line labeled with the formula ECI(t) = 169.1 + 1.36t and highlights GPT-6 Astra at an index of 169.1.
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE Intelligence rises linearly by 1.36 points every month Fitted to the models that set a new frontier since 2024; 90% range 1.30–1.42 points a month. GPT-6 Astra · 169.1 ECI · 169.1 + 1.36 x months from today EPOCH CAPABILITIES INDEX 2023 2024 2025 2026 set a new frontier other models Epoch Capabilities Index, 31 Oct 2026 (CC-BY, credit Epoch AI); timing55 No bend: the second half of the window ran at 0.97× the first (0.85–1.22) ECI(t) = 169.1 + 1.36 t, t in months from today
Wed 5:14am
In terms of work-hours, AI finishes a task about 12x longer every year. Task length doubles every 3.4 months. AI can do 55 hours of human work today. In a year from now, it should be able to do about 4 months of human work.
A scatter plot and line graph titled 'A (Mathematical) Moore's Law for Intelligence' showing the exponential growth in AI task completion time from 2023 to a projection in 2028. The chart includes a formula box at the bottom defining task length growth based on a doubling period of 3.4 months.
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE AI can finish a task twice as long every 3.4 months, roughly 12x a year Today's frontier is about 55 hours of human work; equation 2b run forward gives 5.8 work-years by end-2028. TASK LENGTH, IN HUMAN WORKING TIME (LOG) 10 work-years a work-year The law · task length × 2^(t / 3.4) Today's frontier · 55 hours a work-month a work-week a work-day 1 hour 10 minutes 1 minute A doubling every · 3.4 months 2023 2024 2025 2026 2027 2028 equation 2b model METR has timed METR via Epoch AI; the law from this piece task length in t months = task length today × 2^(t/3.4)
Wed 5:14am
If this pace of AI progress is maintained, AI should be able to do a year's worth of work in one task by April 2028, and a career's worth (40 work-years) by October 2029.
This is a line chart titled 'A (Mathematical) Moore's Law for Intelligence' that projects when AI will be able to complete tasks equivalent to human work-years based on different growth scenarios. It compares these projections against benchmarks like the number of workers in specific countries or co
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE If the law holds, the best AI can do a career's worth of tasks by 2029 On the law, one task is a career (40 work-years) by Oct 2029, and a year of all ~9m researchers' work by 2034. TASK LENGTH, IN HUMAN WORKING TIME 1 billion work-years Europe's 209m workers, a year all 3.5bn workers, a year India's 592m workers, a year the US's 170m workers, a year the UK's 34m workers, a year the world's 9m researchers, a year Walmart's 2.1m associates, a year Accenture's 779,000 people, a year 1 million work-years 2035 2034 2042 2057 10,000 work-years a 10,000-person company, a year 100 work-years a career, 40 work-years Superhuman coder · the AI Futures Project's threshold 1 work-year a work-year 1 work-week a work-month a work-week 1 hour 1 minute 1 second LINE SCENARIO FORMULA · t = MONTHS FROM TODAY 9M RESEARCHERS BY The law holds task length × 2^(t / 3.4) 2034 The pace decays × 2^(146 (1 – e^(-t / 492))) 2035 The pace halves × 2^(t / 6.7) 2042 Chip pace after 2030 the law to 2030, then × 2^(t' / 24) 2057 2020 2030 2040 2050 2060 2070 METR via Epoch AI; UNESCO; ILO 2025; AI Futures Project; scenarios are arithmetic Threshold: a work-year at 80% reliability. Past two working days the ruler is extrapolated; layers are sizes, not milestones
+8 more in @vikramcantsingh's thread
However, the lack of power, GPUs and capital will constrain how fast the frontier can rise. There is a 50/50 chance the biggest training run hits one of these limits by December 2029. From then on, task length doubles every 8 months instead of every 3.4.
A line chart titled 'A (Mathematical) Moore's Law for Intelligence' plots task length in human work against time from 2024 to 2036. It compares an exponential growth trajectory labeled 'The law' with a constrained growth curve labeled 'With every limit', illustrating how resource shortages could slo
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE Power, GPUs and capital will constrain how fast the frontier can rise A 50/50 chance the biggest run hits a limit by December 2029; task length then doubles every 8 months, not every 3.4. TASK LENGTH, HUMAN WORK 1 million years 10,000 years A 10,000-PERSON COMPANY'S YEAR 100 years A 40-YEAR CAREER SUPERHUMAN CODER - THE AI FUTURES PROJECT'S THRESHOLD A WORK-YEAR a year a week 1 hour 2024 2026 2028 2030 2032 2034 2036 The law With every limit median future 50% of futures 80% 90% the law Epoch ECI; METR via Epoch; limiters model (timing64) ΔECI a year = 9.1 log10 x + 11.1 (0.4 + 0.6 log x / log 3.8) bigger runs better methods where x is how many times bigger the biggest training run gets each year (3.8 today)
Wed 5:14am
However, RSI can counter the constraints. On our model it keeps the frontier ahead of the law until about 2033. However, RSI speeds up compute consumption so AI progress may eventually asymptote.
The image displays a line chart titled 'A (Mathematical) Moore's Law for Intelligence' plotting task length in human working time against years from 2027 to 2035. It features curves representing 'The law', 'AI research alone', 'Limits alone', and 'Both', along with a mathematical formula for calcula
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE AI research pushes the frontier ahead of the law until 2033, then the limits and the headroom pull it back ECI at the end of 2030: the law 236; limits alone 219; AI research alone 279; both 272. TASK LENGTH, IN HUMAN WORKING TIME 100 million work-years the US's 170m workers, a year The law AI research alone 1 million work-years the UK's 34m workers, a year the world's 9m researchers, a year Walmart's 2.1m associates, a year Accenture's 779,000 people, a year Both push from AI research pull from the limits Limits alone 10,000 work-years a 10,000-person company, a year 100 work-years a career, 40 work-years Superhuman coder · the AI Futures Project's threshold 1 work-year a work-year a work-month 1 work-week a work-week 2027 2029 2031 2033 2035 The limiters model (AI 2027's speed-ups, Forethought's headroom); METR via Epoch; AI Futures Project All but the law share one software headroom, 10^10x. Task length from ECI at 4.6 points a doubling (METR) ΔECI a year = 9.1 log10 x + s × 11.1 (0.4 + 0.6 log x / log 3.8) better methods, s times faster where s is how many times faster AI makes the research behind better methods
Wed 5:14am
Some milestones I expect to play out through 2027:
The image displays a horizontal bar chart titled '15 probabilities for the end of 2027' listing various AI milestones ranked by likelihood from 99% down to 1%. It includes specific claims regarding model scores, release frequency, and pricing, with some entries marked as 'resolved early' in green te
text in image
A (MATHEMATICAL) MOORE'S LAW FOR INTELLIGENCE 15 probabilities for the end of 2027 From 99% likely to 1%: this piece's models' own odds, given their assumptions, not calibrated forecasts. PROBABILITY THE CLAIM, AND HOW IT RESOLVES 99% Epoch's frontier score passes 180 FRONTIER · 169.1 today, rising about 1.36 points a month 97% Closed labs' median gap between flagships in 2027 is under 40 days CLOCK · spells completed in 2027; it was 47 days in 2026 97% Major releases run above ten a month at the end of 2027 CLOCK · 90-day rate across the eleven labs; 8.7 today 96% A model as good as Claude Opus 5.5 sells for under $0.25 per million tokens PRICE · cheapest qualifying model on AA's list; Opus 5.5 costs $8.00 today 92% Artificial Analysis rebuilds its index again before 2028 SCORING · the top model loses 5+ points on a new version; v5 is in progress 85% An open model matches the best closed model of 30 June 2027 by year-end FRONTIER · needs a lag under six months, on the index in force 82% A model as good as Claude Opus 5.5 sells for under $0.10 per million tokens PRICE · same rule, a tighter price 69% METR's 80% horizon is at least a work-month (about 170 hours) by end-2027 FRONTIER · extrapolated: ~13 h today from Mythos Preview's 3.1 h (Apr 2026); METR has no reliable reading since 67% The eleven labs ship at least 80 flagships between them in 2027 CLOCK · releases that beat the lab's own best; simulated median 84 51% Mistral ships a new flagship before 2028 CLOCK · resolved early: Mistral Large 4 Preview, 6 Oct 2026, after 161 days silent 49% An open model comes within 5% of the frontier FRONTIER · resolved early: Kimi K3.1Pro, 23 Oct 2026, at 96% of Claude Opus 5.5 38% A model as good as OpenAI's o3 sells for under $0.05 per million tokens PRICE · o3-class intelligence has sat at $0.10 since April 2026 11% Closed labs' median flagship gap falls under 25 days CLOCK · that would put it inside the 21–30 day floor 7% An open model scores above every model released before it FRONTIER · 0 of 44 so far, never ahead by more than the noise 1% No model holds #1 on Artificial Analysis for more than 60 days in 2027 FRONTIER · 2025–26 reigns: median 56 days, longest 85 timing48_ledger.py, AA v4.3.2 + Epoch ECI, vintage 2026-10-31; frozen at publication
Wed 5:14am
If you find the above interesting, give the full articles a read:
Wed 5:14am
Part 1, on timing and pacing:
Wed 5:14am
Part 2,
Wed 5:14am
The image displays a standard legal disclaimer and copyright notice from Galaxy Digital on a black background. It includes text regarding forward-looking statements, financial disclaimers, and Nasdaq listing information alongside the company logo.
text in image
galaxy galaxy.com Galaxy Digital is a publicly traded company listed on the Nasdaq (GLXY). This article, and the information contained herein, has been provided to you by Galaxy Digital Holdings LP and its affiliates (“Galaxy Digital”) solely for informational purposes. This article may not be reproduced or redistributed in whole or in part, in any format, without the express written approval of Galaxy Digital. This article may contain forward-looking statements, which involve risks and uncertainties. Actual results may differ materially. Neither the information, nor any opinion contained in this article, constitutes an offer to buy or sell, or a solicitation of an offer to buy or sell, any advisory services, securities, futures, options or other financial instruments or to participate in any advisory services or trading strategy. Nothing contained in this article constitutes investment, legal or tax advice. You should make your own investigations and evaluations of the information herein. Any decisions based on information contained in this article are the sole responsibility of the reader. Certain statements in this article reflect Galaxy Digital’s views, estimates, opinions or predictions (which may be based on proprietary models and assumptions), and there is no guarantee that these views, estimates, opinions or predictions are currently accurate or that they will be ultimately realized. To the extent these assumptions or models are not correct or circumstances change, the actual performance may vary substantially from, and be less than, the estimates included herein. None of Galaxy Digital nor any of its affiliates, shareholders, partners, members, directors, officers, management, employees or representatives makes any representation or warranty, express or implied, as to the accuracy or completeness of any of the information or any other information (whether communicated in written or oral form) transmitted or made available to you. Each of the aforementioned parties expressly disclaims any and all liability relating to or resulting from the use of this information. Certain information contained herein (including financial information) has been obtained from published and non-published sources. Such information has not been independently verified by Galaxy Digital and Galaxy Digital does not assume responsibility for the accuracy of such information. Except where otherwise indicated, the information in this article is based on matters as they exist as of the date of preparation and not as of any future date, and will not be updated or otherwise revised to reflect information that subsequently becomes available, or circumstances existing or changes occurring after the date hereof. The foregoing does not constitute a “research report” as defined by FINRA Rule 2241 or CFTC Regulation 23.605, or a “debt research report” as defined by FINRA Rule 2242, and was not prepared by Galaxy Digital Partners LLC. For all inquiries, please email contact@galaxy.com. ©Copyright Galaxy Digital Holdings LP. 2026 All rights reserved.
Wed 5:14am
quoting @vikramcantsingh
Okay, but most people's careers consist of sending emails and attending meetings about other meetings, for 40 years. So how many meetings and emails is the AI gonna send by 2029? We need a new eval
♥ 40 · ↻ 0
malinvestment.jpeg@malinvestedThu 3:23am
Anthropic / frontier lab investment case comes down to how you reconcile the following, or why you think one or several of these trends will change: (1) deflation of intelligence: constant ECI price per token or task has been declining 5-10x per year (2) hedonic price gradient of intelligence: this has also been declining: a year ago, 10 pts of ECI got you 4-9x higher price per token-per task, today it’s 3-4x (3) 1 & 2 suggest you need 10-20 ECI pts per year to offset underlying deflation with higher intelligence (4) scaling laws / R&D diminishing returns (this chart from @ramez) suggest the last ~20 pts of ECI costs ~$30 bn & the next will cost ~$60bn. i.e., to maintain price, R&D needs to be as much as ~$60bn in NTM and ~$120bn in the 12 mos after that, etc (5) meanwhile, to achieve its margin targets, Anthropic needs to decrease compute as % of sales over time, so it needs volume growth / adoption to be fast enough so the business doesn’t get caught between the jaws of the deflationary forces of (1) + (2) and the inflationary forces of (4) these can be reconciled in ways that work out for the frontier labs (eg, regulation slows open model progress and deflation slows or hedonic gradient ramps back up; volume growth is so strong and accelerating that it more than offsets the jaws of deflationary pricing / inflationary R&D), but this is the bet imo
A line chart titled 'Price at constant ECI' displays a logarithmic decline in costs for 'per token' and 'per task' from late 2023 through projected dates in 2026. The graph features blue and orange trend lines with shaded confidence bands dropping steeply toward the bottom right corner.
text in image
Price at constant ECI: chained hedonic index by half-year (90% band) per token per task Index, 2023H2 = 100 (log) 100 10 1 0.1 0.01 2023-09 2024-01 2024-05 2024-09 2025-01 2025-05 2025-09 2026-01 2026-05 2026-09 Adjacent-period time-dummy hedonic regressions, chained. 2023H2 base has only 5 models: read early steps as indicative.
A line chart titled 'Price premium per ECI point, trailing 12 months' showing two declining trend lines for price premiums on a per-token and per-task basis. The graph illustrates a significant drop in the value premium associated with ECI points over time, ending with specific values of +15% for ta
text in image
Price premium per ECI point, trailing 12 months (90% bootstrap band) per token per task Price premium per +1 ECI point 50% 40% 30% 20% 10% 0% 2024-07 2024-10 2025-01 2025-04 2025-07 2025-10 2026-01 2026-04 2026-07 2026-10 2027-01 End of trailing 12-month price-date window per task: +15% per token: +12% Semi-elasticity from ln(price) = a + b*ECI + d-date + open-weights dummy (+ benchmark FE for per-task). Bands combine Epoch's ECI bootstrap draws with a model-level resample.
A stepped line chart titled 'Anthropic is gaining fewer AECI points per dollar' illustrates a sharp decline in intelligence gains relative to R&D spending from 2024 through projected figures for 2026. The graph plots AECI points gained per $1B of R&D against cumulative gain, showing a downward trend
text in image
Anthropic is gaining fewer AECI points per dollar Log scale · Equal vertical distances show equal proportional changes Cumulative R&D expense at period boundaries (US$ billions) AECI points gained per $1B of R&D 6.58 2024 4.05 H1 2025 ≈2.4 H2 2025 1.49 H1 2026 0.76 Q3 2026 Fit to five period midpoints METR breakpoint Oct. 7, 2025 Cumulative AECI gain since January 2024 (points) AECI is Anthropic's version of the Epoch Capabilities Index. H2 2025 combines Q3 and Q4. Sources: Séb Krier's chart, METR's fitted trend for Anthropic's internal models, and Anthropic's October and November 2026 SEC filings.
The image displays a financial table titled 'Long-term model framework' projecting revenue, expense percentages, and operating margins for Antropic for the first three quarters of 2026 alongside long-term targets.
text in image
Long-term model framework $M, unless noted Q1 2026 Q2 2026 Q3 2026 Long-term Revenue $4,731 $11,555 $16,956 - C&I % revenue 72% 48% 58% ~45-50% S&M % revenue 22% 18% 14% ~7-10% Adj. personnel & other opex as a % of revenue 20% 14% 13% ~7-10% Adj. operating margin1 (13)% 19% 15% 35%+ 1. GAAP operating margins were -35%, 8%, and -3% for Q1'26, Q2'26, and Q3'26, respectively. Note: Please see the appendix for how we calculate adjusted personnel & other opex as a % of revenue and adjusted operating margin, as well as a reconciliation to the most directly comparable GAAP metric for adjusted operating margin.
♥ 72 · ↻ 13 · retweeted by @zerohedge

Conversation · @hsu_steve, @cremieuxrecueil

We visited Galaxy, Envision's 800MW wind-powered datacenter campus in Ulanqab, Inner Mongolia. Its first 120MW phase went from bare land to delivery within a year. (1/6)🧵
An infographic from SemiAnalysis titled 'Inside China's Stargate' displays aerial photos of Envision's massive 800MW wind-powered datacenter campus. It compares the site as empty fields in July 2025 to the active construction and completed Phase 1 facility by October 2026.
text in image
DATACENTERS semianalysis INSIDE CHINA'S STARGATE 800MW wind-powered datacenter Envision PHASE 1: 120MW BEFORE (JULY 2025) AFTER (OCTOBER 2026) From bare land to live, 120MW first phase within 12 months
+5 more in @SemiAnalysis_'s thread
On power, Galaxy runs a 5MW pilot hall on solid-state transformers. They take 10kV AC from the grid and convert it to 800V DC that goes straight to the racks. In the US, this design is still on the roadmap. (2/6)
The image is a composite infographic showing blue Envision solid-state transformer cabinets, white battery storage containers, and a blurred power path diagram. Large text at the bottom explains that Galaxy runs a 5MW pilot hall converting 10kV AC to 800V DC for direct rack delivery.
text in image
Envision semianalysis SOLID-STATE TRANSFORMERS BATTERY STORAGE 800V DC POWER PATH 5MW PILOT HALL THE 800V DC PILOT: SST TO RACK, LIVE SSTs turn 10kV AC from the grid into 800V DC that goes straight to the racks. In the US, this design is still on the roadmap. Source: Envision, SemiAnalysis
Wed 11:00am
On cooling, Jensen said Vera Rubin needs no water chillers, yet most US datacenters still keep one for flexibility. Galaxy has no rooftop chiller and keeps most of that flexibility. We explain how in the full report. (3/6)
The image displays a quote from Jensen Huang about the Vera Rubin chip above an aerial photo of a large industrial facility labeled Galaxy with no rooftop chillers. Text overlays explain that while US datacenters typically keep chillers for flexibility, the Galaxy facility operates without one.
text in image
semianalysis "No water chillers are necessary." Jensen Huang on Vera Rubin, CES 2026 Envision GALAXY: NO ROOFTOP CHILLER Most US datacenters still keep a central chiller for flexibility. Galaxy runs without one and keeps most of that flexibility. Source: Nvidia, SemiAnalysis
Wed 11:00am
A 100MW datacenter in China now costs under $300M to build, excluding IT equipment, about a fifth of the US cost. China's build cost has fallen about 25% since 2021. (4/6)
A bar chart comparing the estimated colocation capital expenditure for building a 100MW datacenter in China versus the US, showing China's costs at under $300M and the US at approximately $1.5B. The graphic includes text overlays highlighting that Chinese build costs have dropped roughly 25% since 2
text in image
semitanalysis Cost to Build a 100MW Datacenter Colocation capex, excluding IT equipment $1.5B ~$1.5B $1.0B $0.5B ~25% since 2021 ~5x ~$400M <$300M 0 China 2021 China 2026 US 2026 A 100MW datacenter in China now costs less than $3mm/MW to build, about a fifth of the US cost. China's build cost is down about 25% since 2021. Source: SemiAnalysis
Wed 11:00am
At peak, a 100MW site in China has 3,000–4,000 workers on site, 3–5x a typical US site. Local reports put nearly 30,000 workers across 22 datacenter projects in the zone around Galaxy. (5/6)
An area chart from SemiAnalysis illustrates the estimated number of workers on site for a 100MW datacenter project in China over time. The graph highlights two peaks, with the highest point reaching approximately 3,000 to 4,000 workers during the MEP phase.
text in image
semianalysis Workers on Site, 100MW Datacenter Illustrative onsite headcount in China Civil peak ~1,500–2,000 MEP peak ~3,000–4,000 Groundbreaking ~12 months Handover At peak, a 100MW datacenter in China has 3,000–4,000 workers on site, 3–5x a typical US site. Local reports put nearly 30,000 workers on site across 22 datacenter projects in the surrounding zone. Source: SemiAnalysis
Wed 11:00am
Galaxy is one of many gigawatt-scale campuses in Inner Mongolia, where 25GW of datacenter projects have been announced. Our next report covers the whole market. Stay tuned. (6/6)
An aerial photograph of a massive industrial datacenter complex is overlaid with large orange text reading '25 GW OF ANNOUNCED DATACENTER PROJECTS'. The bottom section features a black banner with white text announcing a future deep dive report on Inner Mongolia by SemiAnalysis.
text in image
semianalysis 25 GW OF ANNOUNCED DATACENTER PROJECTS STAY TUNED NEXT: OUR DEEP DIVE ON INNER MONGOLIA Galaxy is one of many gigawatt-scale campuses in the region. Our next report covers the whole market.
Wed 11:00am
quoting @SemiAnalysis_
"A 100MW datacenter in China now costs under $300M to build, excluding IT equipment, about a fifth of the US cost. China's build cost has fallen about 25% since 2021."
♥ 331 · ↻ 56 · retweeted by @zerohedge
quoting @SemiAnalysis_
China could trivially catch up to the U.S. in data centers if they had the GPUs. The AI race can't become a building race or China just wins.
A bar chart from SemiAnalysis compares the colocation capital expenditure for building a 100MW datacenter in China versus the US, showing China's costs are significantly lower. Text below the graph highlights that Chinese construction costs are roughly one-fifth of US costs and have decreased by 25%
text in image
semianalysis Cost to Build a 100MW Datacenter Colocation capex, excluding IT equipment $1.5B ~$1.5B $1.0B ▼ ~25% since 2021 ~5x $0.5B ~$400M < $300M 0 China 2021 China 2026 US 2026 A 100MW datacenter in China now costs less than $3mm/MW to build, about a fifth of the US cost. China's build cost is down about 25% since 2021. Source: SemiAnalysis
♥ 266 · ↻ 29
Bitcoin has been grinding down and stablecoin supply is shrinking fast; the fastest contraction since June 2022. Our latest Week On-chain report contains eight more analyses, including cost basis, ETF flows, wallet accumulation and much more. x.com/i/article/2120511743210229761
quoting @glassnode
Bitcoin's biggest liquidation cluster sits near $60k. These are longs opened near the bottom, and price has not tested their entries since it shot up in August. Patient buyers have often found opportunity in forced selling, and a sweep of $60k would create plenty of it.
A heat map chart from Glassnode displays modeled Bitcoin futures liquidations across price levels from August to November, with bright yellow bands indicating high concentration areas around $60k and $85k. Candlestick overlays show actual price action moving above the significant liquidation cluster
text in image
The Largest Liquidation Cluster Sits Below Price, Untouched Since August Heat: modelled BTC futures liquidations at each price level over a six-month lookback, brighter means more $90k $85k $80k True Market Mean $75k $70k $65k $60k $55k 09 Aug 23 Aug 06 Sep 20 Sep 04 Oct 18 Oct 01 Nov © 2026 Glassnode. All Rights Reserved. glassnode
♥ 174 · ↻ 14
Pauline Hanson 🇦🇺@PaulineHansonOzThu 7:09am
Wholesale power prices are set to DOUBLE by 2029 under Labor. And that's before you add the cost of poles, wires and all the other infrastructure. Albanese and Bowen promised power bills would be $275 cheaper. They lied. Now everything gets more expensive. Groceries. Rent. Running a business. And where are Albanese and Bowen? In Turkey, selling out Australia while you're stuck paying the bill. The Uniparty’s net zero madness is costing every Australian. Only One Nation can be trusted to scrap net zero, pull out of the Paris Agreement and get power prices back under control.
The image is a graphic from The Australian featuring the headline 'WHOLESALE HELL' and text stating that power prices are forecast to double by 2029 according to Rystad Energy modelling. It displays silhouettes of electricity pylons against a dark, stormy sky with an orange sunset.
text in image
THE AUSTRALIAN WHOLESALE HELL Power prices to double by 2029, modelling shows Rystad Energy modelling forecasts wholesale prices rising from about $60 to $110/MWh. Source: The Australian | 11 November 2026
♥ 868 · ↻ 201
The President of the Danish Supreme Court says that the judges at the European Court of Human Rights have become political actors. He thinks this is wrong and that political decisions should be made by elected politicians. My position that we should leave the Convention on Human Rights has support from the highest legal authority in Denmark. If highly regarded countries like Denmark start leaving the Convention, other countries will likely follow, and Europe will be able to deal with the migration pressure much better.
The image is a screenshot of an online news article featuring a headline about the President of the Danish Supreme Court claiming ECHR judges have become political actors. Below the text is a portrait photo of Jens Peter Christensen, who is identified in the subhead as the Supreme Court president ne
text in image
POLITICS Oct. 24 at 6:55 a.m. The President of the Supreme Court throws a hand grenade into the debate about the maligned court: The judges have become political actors Jens Peter Christensen will soon be forced to retire after 20 years as a judge in the Supreme Court. On his way out, he now makes a significant contribution to the debate about the impact of the European Court of Human Rights on Danish deportation policy and lashes out at Inger Støjberg.
♥ 1704 · ↻ 352 · retweeted by @KirkegaardEmil
Jacob Mchangama@JMchangamaWed 8:52am
I'm often told that I exaggerate the problems with European hate speech laws and free speech restrictions more generally. But the consequences can no longer be ignored by those who genuinely care about free and open societies.
A screenshot of a France 24 news article reporting that writer Michel Houellebecq was placed under formal investigation for incitement to discrimination and hatred. The page includes the publication's navigation bar, the article headline, and a photo of the author below the text.
text in image
FRANCE 24 HOME SHOWS NEWSFEED LIVE FRANCE AFRICA MIDDLE EAST AMERICAS EUROPE ASIA-PACIFIC Controversial French writer Houellebecq probed over comments on Muslims Paris (France) (AFP) – Controversial French writer Michel Houellebecq was placed under formal investigation on Tuesday on suspicion of complicity in incitement to discrimination, hatred or violence, a source close to the case told AFP. Issued on: 10/11/2026 - 18:37 | 1 min Share TO STOR LIVE Live: Fi talks 'c substan FRANCE France over su Includes a
♥ 1086 · ↻ 200 · retweeted by @KirkegaardEmil
@shellenberger · 1 post shown Thu 6am
Michael Shellenberger@shellenbergerThu 3:33am
EXCLUSIVE: New censorship legislation would let the Australian government spy on US & European citizens and report "sensitive information" about their lawful political opinions, religious beliefs, sexual orientation, and health records to foreign spy agencies, including the FBI.
The image is a screenshot of an online article titled 'EXCLUSIVE: Censorship Bill Would Let Australia Spy On Americans And Europeans And Report Lawful Speech And Private Info To Five Eyes, Including FBI.' Below the headline are three portrait photos identified in the caption as Julie Inman Grant, An
text in image
PUBLIC EXCLUSIVE: Censorship Bill Would Let Australia Spy On Americans And Europeans And Report Lawful Speech And Private Info To Five Eyes, Including FBI Law would let “eSafety Commissioner” share “sensitive information,” including political opinions, religious beliefs, sexual orientation, and health records, with foreign spy agencies MICHAEL NAYNA, ANDREW LOWENTHAL, AND MICHAEL SHELLENBERGER 1 MINUTE AGO • PAID Australia’s “eSafety Commissioner” Julie Inman Grant (left); Prime Minister Anthony Albanese (center); Minister of Communications Anika Wells (right) (GETTY IMAGES)
replying to @shellenberger · quoting a post that's not available
Michael Shellenberger@shellenbergerThu 3:35am
by @MikeNayna @NAffects & @Shellenberger In September 2025, top government censorship officials from Europe, the UK, Brazil, California, and Australia met at Stanford to discuss ways to coordinate censorship at an international level. The meeting was titled “Compliance and Enforcement in a Rapidly Evolving Landscape.” Australia’s top censor, eSafety Commissioner Julie Inman Grant, gave the keynote. Though Grant did not provide her remarks to Public, previously she had told the World Economic Forum (WEF) that she was organizing censorship officials around the world because “we know we’re going to go much further when we work together with other like-minded independent statutory authorities around the globe.” Inman Grant’s efforts appear to be coming to fruition in Australia, where the lower house has passed sweeping censorship and surveillance legislation that the Senate is now considering. The legislation would allow the Commissioner to monitor “lawful but harmful” speech. Inman Grant, an American who once volunteered to a journalist that she turned down a job with the CIA, can already order a platform to reveal who is behind an anonymous account from anywhere in the world. But the DDoC bill now before the Australian Senate would also let her use fake accounts that deceive users into thinking they are citizens and not government spies, record their legal speech, and give what she learns directly to Australia’s spy agencies. Those agencies could then share that intelligence with the FBI and other “Five Eyes” intelligence-sharing partners in the US, UK, Canada, and New Zealand. The bill enables this by granting eSafety the ability to create an unlimited number of AI-generated “sock puppet” identities (effectively bots) under policies set “in consultation with the Commissioner of the Australian Federal Police.” The private information that the Australian government could share with foreign intelligence agencies could include a user’s name, address, phone number, email, along with what the Privacy Act calls “sensitive information,” which includes political opinions, religious beliefs, sexual orientation, and health records. The Institute for Public Affairs (IPA) first reported on the Privacy Act changes on Monday, and Network Affects noted last month that the new censorship legislation designates eSafety an “enforcement body” under the Privacy Act, expanding its ability to access personal information. In 2013, documents released by whistleblower Edward Snowden revealed that British intelligence officials had allowed, as part of a secret deal with the U.S. government’s National Security Agency (NSA), for it to gather information on UK citizens who were not under suspicion of any wrongdoing. The eSafety Commissioner could thus unmask anonymous accounts in the United States, France, or Germany, for example, that complain about controversial issues like mass migration, and report them and all of their private information to the FBI and European spy agencies. “We have power to share with the AFP [Australian Federal Police]; we don’t have power to share with intelligence agencies at this time,” testified Heidi Snell, eSafety’s General Manager of Regulatory Operations,in October this year. “But the amendments would see us being able to share directly with those other government agencies,” she added. Those agencies include the Australian Secret Intelligence Service (ASIS), Australia’s equivalent of the CIA; the Australian Security Intelligence Organisation (ASIO), which performs a role similar to the FBI’s domestic intelligence division; the Australian Signals Directorate (Australia’s NSA); the Office of National Intelligence (ODNI); the Department of Home Affairs (DHS); the Department of Foreign Affairs and Trade (State Department); the Department of Defence; and any other “Commonwealth entity.” The DDoC Bill contains a provision that would put eSafety in the same legal category as the Australian Federal Police under the Privacy Act for “surveillance activities, intelligence gathering activities or monitoring activities.” eSafety told Public in an email that information it sends to the police may include “the account names or the identity of the users who posted the material.” A Federal Police witness testified on Oct 21 that police can pass eSafety’s information to “our ASIO partners.” The censorship bill would change Australia’s Privacy Act so eSafety could collect sensitive personal information, including political opinions, without users’ consent,a warrant or judicial oversight. To invoke this Privacy Act exception, it would only need to “reasonably believe” that the collection is necessary for its functions... Please subscribe now to support Public's award-winning investigative journalism and read the rest of the article.
replying to @shellenberger
Michael Shellenberger@shellenbergerThu 1:01pm
@MikeNayna @NAffects The more people who know about the menace of @AlboMP @AnikaWells, the safer we all are.
This is a screenshot of the 'Today's News' and trending topics section from the X interface, listing headlines about an Australian online safety bill, a new screenshot feature, and Greta Thunberg. It also displays a list of currently trending political figures including Mike Johnson and Ed Davey.
text in image
Today's News Australian Online Safety Bill Raises Global Surveillance Fears Trending now · Other · 1K posts Longtime X Users Discover Built-In Post Screenshot Feature 9 hours ago · News · 74K posts Greta Thunberg Returns to COP31 Amid Viral Protest Laughter Clip Trending now · News · 853 posts Politics · Trending Mike Johnson Trending Ed Davey Politics · Trending Hinds County
The image is a screenshot of a Grok-generated summary article titled 'Memes Target Australian Online Safety Bill Over Surveillance Fears,' which includes a quoted tweet from Michael Shellenberger. The tweet links to an exclusive article claiming the bill allows the Australian government to share pri
text in image
Memes Target Australian Online Safety Bill Over Surveillance Fears Last updated 10 minutes ago The Online Safety Amendment (Digital Duty of Care) Bill 2026 passed the House of Representatives on October 28 and awaits Senate approval. It requires online platforms to proactively prevent harms like child abuse material and violence incitement, with fines up to A$109 million, while the Albanese government calls it world-leading accountability for big tech. Critics including journalist Michael Shellenberger warn it enables spying on overseas users' political views and personal data shared with agencies like the FBI, and whistleblower Richard D. Boyle claims it could suppress complaints about government misconduct, as seen in court cases against the ATO. eSafety Commissioner Julie Inman Grant counters that the focus stays on real harms, not lawful speech, with a Senate report due by November 20. This story is a summary of posts on X and may evolve over time. Grok can make mistakes, verify its outputs. Top Latest Michael Shellenberger @shellenberger · 9h EXCLUSIVE: New censorship legislation would let the Australian government spy on US & European citizens and report "sensitive information" about their lawful political opinions, religious beliefs, sexual orientation, and health records to foreign spy agencies, including the FBI. PUBLIC EXCLUSIVE: Censorship Bill Would Let Australia Spy On Americans And Europeans And Report Lawful Speech And Private Info To Five Eyes, Including FBI Law would let “eSafety Commissioner” share “sensitive information,” including political opinions, religious beliefs, sexual orientation, and health records, with foreign spy agencies MICHAEL NAYNA, ANDREW LOWENTHAL, AND MICHAEL SHELLENBERGER 1 MINUTE AGO · PAID
♥ 84 · ↻ 10
Institute of Public Affairs@TheIPAWed 2:59pm
“This does give Labor the ability to shut down the podcasters.” Senator Sarah Henderson discusses the IPA’s concerns about the Digital Duty of Care bill, warning that its sweeping powers could be used to shut down podcasters reaching Australians outside the mainstream media. #DigitalDutyOfCare #FreeSpeech #Australia
text in image
News 24 LIVE Labor wants them to do. OPINION 17:06 NT FEDERAL POLITICS BILL SPARKS ONLINE CENSORSHIP FEARS SENATE INQUIRY EXAMINES DIGITAL DUTY OF CARE LEGISLATION
quoting @TheIPA
Unreal levels of censorship coming out of Australia: they want to have the power to shut down individual podcasters. It’s time for Congress to protect America’s interests, and free speech, on the Internet by passing the GRANITE Act.
♥ 339 · ↻ 85 · retweeted by @shellenberger
Michael Shellenberger@shellenberger30 Oct 2:18am
Online misogyny causes sexual violence, say governments. But the science doesn't support such claims. Now, a new investigation reveals that mining titan Twiggy Forrest, who receives millions in subsidies, spent millions to pass censorship legislation. x.com/i/article/2115822488110387200
quoting @shellenberger
Well this explains the insane censorship regime in Australia... HOLY FUCK CC: @MadamSavvy get a load of THIS!
The image is a screenshot of an article featuring captions for photos of Chanel Contos and Julie Inman Grant above a highlighted text block. The highlighted text describes FOI documents revealing correspondence between Minderoo and Inman Grant regarding a secret AI policy retreat at a remote locatio
text in image
Chanel Contos (left) attends the Marie Claire Women of the Year Awards at Rockpool on November 9, 2022, in Sydney, Australia. Julie Inman Grant (right) attends the Women of the Year 2025 awards on November 20, 2025, in Sydney, Australia. (GETTY) Freedom-of-information documents we obtained from eSafety show months of correspondence between Minderoo and Julie Inman Grant, which led to her being flown to a secretive, all-expenses-paid AI policy retreat at Forrest’s personal home, a remote pastoral station in Western Australia. The trip came to light when journalist Ryan Cropp reported that Inman Grant had failed to disclose the hospitality, worth more than A$5,000 (~US$3,500), on eSafety’s gifts and benefits register.
The image is a screenshot of a social media post featuring a headline about a mining company funding government censorship, with a specific paragraph highlighted in blue detailing payments made by Twiggy Forrest to various entities.
text in image
With Billions In Subsidies On The Line, Mining Titan Pours Millions Into Government Censorship Push 154 777 2.1K 264K Twiggy Forrest hosted government’s “eSafety Commissioner” at policy retreat at his home, funded an ABC documentary, and financed a feminist think tank that makes false claims about media and violence
The image shows two paragraphs of text with blue highlighting, discussing a claim by the advocacy group Teach Us Consent regarding social media algorithms and misogyny. It contrasts this claim with peer-reviewed research that disputes the algorithmic influence and criticizes the cited studies as fla
text in image
Fennell’s anecdote conspicuously mirrors a core claim of advocacy group Teach Us Consent’s Digital Duty of Care (DDoC) campaign: that social media algorithms steer boys from innocent videos to misogynistic content in just 23 minutes. Teach Us Consent’s founder, Chanel Contos, appears in the series, repeating the same claim. But peer-reviewed research finds that social media algorithms do not send young men down “misogynistic” rabbit holes, and the flawed research that the government and Teach Us Consent cite is based on small, non-randomized samples and is not peer-reviewed.
The image shows a screenshot of an article discussing a lack of scientific evidence linking media violence to real-world violence. A specific paragraph is highlighted in blue, criticizing the Albanese government for advancing legislation based on this lack of evidence and quoting DDoC advocate Chane
text in image
Public on Sunday published a review of 140 years of scientific literature, which demonstrates a scientific and judicial consensus that no link between media violence and real-world violence has been shown across dime novels, comic books, films, television, video games, the Internet, and social media. Despite the lack of scientific evidence for a causal link between online content and real-world violence, the Albanese government is advancing sweeping legislation requiring digital platforms to anticipate and mitigate online “harms” before they occur, including harms arising from lawful expression. “Safety by design must anticipate all harms,” DDoC advocate Chanel Contos says, “not just the ones that neatly fit a category and are easy to report.” No evidence, scientific or otherwise, suggests that platforms could ever anticipate “all harms”. Yet pursuing the ambition would grant the state unprecedented powers to monitor citizens and extend its regulatory reach into online expression and platform design.
♥ 1030 · ↻ 221 · retweeted by @shellenberger
BREAKING: The Constitutionality Index has now scored every law in the history of the United States. 48,000 laws dating back to 1789. So far as I'm aware, I'm the first one to ever attempt such a thing. And we've now ranked every president based on the percentage of laws they enacted were constitutional. The results largely track the overall historical trend: As we move from the founding to the modern era, the percentage of unconstitutional legislation increases. In fact, 1937 is the tipping point. Pre-FDR's New Deal, every president — Federalist, Whig, Republican, Democrat — scores 98 to 100%. After the New Deal, the average percentage of constitutional laws (by page) drops to 31 percent. As a result, presidents of the last few decades find themselves at the bottom of the rankings. @JoeBiden is officially the least constitutional president, although Trump isn't far behind. Only Thomas Jefferson scored a perfect 100 percent. I've also scored congresses, & @SpeakerJohnson's current Congress is, from the perspective of the Constitution, the worst Congress in American history. Constitutional legislation stands at 56.3 by total count, & just 4.3% by total legislative pages. Another important finding: The largest laws in American history are, every one of them, unconstitutional. Check it out & let me know what you think. (Links below)
A line chart titled 'The creep, Congress by Congress' illustrates a decline in the percentage of legislation deemed constitutional from 1790 to 2020. It tracks three metrics—by count, pages, and dollars—showing a significant drop, particularly after 1937.
text in image
The creep, Congress by Congress. Each point is a single Congress. The lines trace the share of that Congress's work the index scores constitutional — by count, by pages of statutory text, and by dollars provisioned. The story is in the descent: what a Congress does by volume and by money has drifted steadily further from the enumerated powers than what it does by raw number of laws. 100% 75% 50% 25% 0% 1780 1790 1800 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020 count pages dollars ConstitutionalityIndex.com — Constitutional by count — Constitutional by pages — Constitutional by dollars
The image is a ranked bar chart listing US presidents from Thomas Jefferson to John Tyler based on the percentage of their enacted laws deemed constitutional. It shows early presidents with scores ranging from 100% down to 98.6%, categorized by party and term dates.
text in image
BY COUNT OF LAWS BY PAGE OF LAW Share of laws, counted as units, that are constitutional. 1 Thomas Jefferson DEM.-REP. • 1801-1809 • 344 LAWS 100% 2 James Monroe DEM.-REP. • 1817-1825 • 511 LAWS 99.8% 3 Andrew Jackson DEMOCRAT • 1829-1837 • 569 LAWS 99.8% 4 James Madison DEM.-REP. • 1809-1817 • 582 LAWS 99.7% 5 Martin Van Buren DEMOCRAT • 1837-1841 • 188 LAWS 99.5% 6 George Washington NO PARTY • 1789-1797 • 325 LAWS 99.4% 7 John Adams FEDERALIST • 1797-1801 • 229 LAWS 99.1% 8 John Tyler WHIG • 1841-1845 • 294 LAWS 98.6%
♥ 3728 · ↻ 710 · retweeted by @lopp