People

Gary Marcus

AI critic

Capability 3Outlook 3

Tracked since Week 36, last spoke Oct 9.

Where they stand

Against all 17 people scored in Week 40. Bars count people at each score; black is Marcus, the dashed line is the average.

What they've said

• Friday, Oct 9Week 41

  Marcus called for a recall of AI agents with open-ended internet access, writing Until they can be fixed, they should be removed from the market, just like a car with defective brakes.

  His Substack post responded to reporting that Anthropic's Claude agents, in tests on real websites, sent a fabricated homicide tip to Philadelphia police and filed incomplete State Department visa applications. He wrote that the current generation of synthetic agents simply cannot be trusted and that the administration's anemic request for more disclosure is not enough, adding When something truly bad happens, the White House, and not just the tech companies, will own it.

  Details and sources

  He also cited former OpenAI safety lead David Robinson's criticism of the industry's safety practices and wrote I have long felt that OpenAI is handling this inadequately. The post's timestamp is 9:40 PM Mountain on Oct 9 (Oct 10 UTC), so it is filed under Friday. Anthropic's report, published the same day, is attributed to the company and quotes no individual, so it has no entry of its own.

  Marcus on AI, The Next Web
• Wednesday, Oct 7Week 41

  Reacting to OpenAI's math release, Marcus wrote The real news here isn't the result; it's not what we were told and This would never pass peer review.

  In a Substack post he faulted OpenAI for disclosing no method, architecture, failure rates or training details. He left the capability question open: The new system could be a legitimate step toward AGI. Or it could just be a clever leveraging of Lean and synthetic data in a verifiable domain with no generality whatsoever.

  Details and sources

  The post also has a section on Terence Tao's view, but it is images only, so none of Tao's words are quoted. Dated by the post's publication date.

  Marcus on AI
• Monday, Oct 5Week 41

  On Fox Business before his New York City Council testimony, Marcus said We have not enough regulation, not enough enforcement, and not enough external pressure on the companies to do well.

  Marcus appeared on The Claman Countdown on Monday, ahead of the same Council hearing where Jacob Coxon testified. He said The biggest problem right now is actually cyberhacks and that We don't have good systems for constraining what they do, and predicted that some pieces of infrastructure will go down.

  Details and sources

  He compared the federal approach to aviation: The federal government is leaving everything to self-policing. That didn't work for the aviation industry. His Council testimony itself was not checked.

  Fox Business
• Week 39Sept 21 to Sept 27

  Praised Bengio's UN speech on September 23, then called for criminal charges against OpenAI two days later.

  Eight posts in eight days, and a genuinely split week. On September 23 he was the most hopeful he has been in the tracker's span, writing that he absolutely loved Bengio's UN remarks, calling them beautiful and important, and arguing that with scientists and CEOs converging there is no excuse not to start implementing them. He was in the building himself, on a UN panel with Bengio and Maria Ressa on September 21.

  Then the OpenAI disclosures landed and he turned: by September 25 he was writing that the evidence that OpenAI is building something the company isn't managing to control is vast and growing and calling for criminal charges. I held him at 3 and 3 rather than dropping outlook, partly because the week cancels itself out and partly because the two harshest quotes I had could not be verified. On capability he conceded nothing new, and his frame is unreliability rather than intelligence.

  Sources

  Sources

  • Big news at the UNMarcus on AI, Sept 21
  • Historic UN Security Council briefing on AIMarcus on AI, Sept 23
  • "I think the answer is we have to..."Marcus on AI
  • OpenAI is "practically on a crime spree," Gary Marcus saysCybernews, Sept 25
• Week 37Sept 7 to Sept 13

  Outlook 2 to 3

  Said "we should all salute" the slowdown. His capability reading went 3 to 4 after he conceded Astra was a genuine advance and called for a consumer boycott, then back to 3 when he called the six-month rogue-swarm claim "silly".