Foreword
AI news has returned to a torrent, but I wish more of it were about the screaming pace of AI capabilities instead of Jensen Huang’s affair and Anthropic’s impending IPO.
You know what else is impending? 2027.
AI 2027 authors, after previously pushing back their timelines, now say, “Looks like the intelligence explosion will happen in 2027 after all.”
Daniel Kokotajlo’s estimate for the arrival of a superhuman AI researcher was December 2028 just a few months ago; now it’s May 2027.
In their lengthy new Q3 update, the writers now estimate that reality is running about 5 months ahead of their original AI 2027 timeline. The recent assessment from METR that AI may be speeding up Anthropic’s R&D by 2x or more is cited as a big reason for the shift.
Eli Lifland asks himself, “Surely this crazy stuff isn’t actually happening while most of the world is barely paying any attention, right??”
At this pace, they write, “an industry-wide pause on AI R&D should be the top priority for the AI industry, the US government, and the entire world.”
AI Futures Project - Q3.5 Timelines Update: Looks Like the Intelligence Explosion Will Happen in 2027 After All
MIRI launches Do Not Accept campaign
I was one of many to play some small part in the development and rollout of donotaccept.ai, our latest effort to raise awareness of humanity’s precarious situation.
This project was pretty far along when Jacob Coxon’s whistleblowing resignation from Anthropic did more of that work than our project’s video and website ever could. But we think the video may hit harder for some people.
We hope you’ll throw it a like, share it, and add your name to the growing list of those who do not accept the terms and conditions under which smarter-than-human AI is being developed.
Claude Fable 5.2 is an artist, among other things
Anthropic advanced the publicly available frontier yesterday with the release of the new model. As with all new releases, posted benchmarks may be impressive, but real-world achievements matter more and can take time to emerge.
For quick impressions of Claude models in particular, I advise looking at their artistic outputs and going with your gut. That’s because Claude models, lacking the native image generation features of their competitors, can’t cheat your senses — that is, they can’t generate a complete image in one go that passes your initial scans but on closer inspection has bent geometry, text matching no alphabet, and other subtle tells of an image built out of vibes rather than understanding.
Claude Fable’s art is all made through code, programmatically placing each shape, color, or brush stroke. This requires intention and taste, as the model looks at its work and iterates to dial in a vision.
One of the most impressive demos I’ve seen so far is a set of paintings released by an Anthropic employee (check out the whole thread, which includes a photorealistic image made of ~700,000 brush strokes).
But you can also spend a long time gazing into Ethan Mollick’s showcase, in which Fable floats your perspective through a series of portals to hundreds of virtual worlds. This was created by over 25,000 agents taking inspiration from books, art, and real-world places.
Similar rules apply to music. Claude isn’t a native audio generator like Suno that creates full tracks out of whole cloth. But it has shown something approaching virtuosity for composing music note-by-note. Try giving a listen to this Sonata from Fable 5.2, for example.
The takeaway I want people to get from these is not, “Wow, artists are cooked,” but rather, “Wow, models not trained specifically to output images and music are rivaling or even surpassing the specialists.” This is what general intelligence looks like, and it’s a sign that we could be very close to AI that outperforms humans at just about anything it applies itself to.
Anthropic has set a date (Thursday 11/19) and a target valuation ($1.33 trillion) for its IPO
That’s down from the $1.5T analysts expected before the big AI stock plunge last week, but still above OpenAI’s most recent valuation of $1.25T.
Everyone knows that everyone else knows about Jensen Huang’s affair.
That’s probably the most important takeaway from the many updates we got today about the AI chip magnate’s “close relationship” with Alice Huang, a probable Chinese spy. Making this common knowledge forces Nvidia and the White House to respond, and increases the political cost of standing by Jensen.
The story went full viral after the dropping of any pretense that Huang and Huang’s “close relationship” was anything other than an extra-marital affair.
The Washington Post published the full text of an apology letter Jensen sent employees in which he says “I was unfaithful to my wife.” It makes no mention of Alice, the company she ran that likely funneled his export-restricted chips to China, or the national security concerns.
Per the New York Times, Secretary of State Marco Rubio oversaw the memo that circulated among top Trump administration officials about the affair. Some in the administration were worried about Huang’s influence on AI policy. Trump is still standing by his AI whisperer: “I think he’s an honorable guy,” he said of Jensen on Sunday. “I think he probably has an excuse for it.”
Separately in the NYT: Beijing officially denied that Alice was a Chinese spy. Chinese social media was more excited to learn that Jensen’s “41-Year Loyal Husband Persona” was a lie.
Also in NYT: An email late last year from Nividia’s lawyer to the Commerce Department halted a probe into export-control violations of Megaspeed, the company Alice was running. The email seems to have hinted that a top official at the the department who was close to Trump would be displeased by continued investigation. Shortly after, the Justice Department had Commerce pause it.
WaPo reports: Nvidia’s board is investigating the relationship, but Huang will keep running the company and is probably seen by the board as irreplaceable.
Per Politico: In the wake of the scandal, congressional Democrats are now expected to attach chip export controls to a must-pass defense bill, betting that Republicans will accept it rather than wait for the new Democratic majority to take office in a couple of months and pass something even stronger.
The Washington Post - Read Jensen Huang’s letter to Nvidia about his affair with suspected Chinese spy
NYT - Rubio Warned Officials About Nvidia Chief’s Entanglement With Chinese Spy
NYT - China Rejects Espionage Claims Involving Nvidia C.E.O. Jensen Huang
NYT - U.S. Commerce Dept. Inquiry Into Megaspeed Was Halted After Nvidia Intervention
The Washington Post - Nvidia board to investigate CEO over relationship with suspected Chinese spy
Politico - Nvidia’s spy scandal hands Hill hawks fresh ammo on China chips
“There’s material risk of AI takeover in the next 6-12 months,” says Paul Christiano
The veteran alignment researcher and former head of the US government’s AI safety agency made these remarks yesterday to attendees of The Hinton Lectures 2026.
“We have seen roughly the things we might expect to see shortly before something really bad were to happen,” he continued.
Christiano is currently on OpenAI’s safety committee.
X (Twitter) - Thread by @robertwiblin (2 tweets)
The analyses and opinions expressed on AI StopWatch reflect the views of the individual contributors and the sources they cover, and should not be taken as official positions of the Machine Intelligence Research Institute.