Search: “safety”

Fields Medalist Jacob Tsimerman knows MUCH MUCH more about math than I do, or ever will, but I have been studying natural and artificial intelligence for 40 years, and I think his prediction here (“AI

interesting (and completely opposed to @haider1’s take on the same graph) *the labs themselves* have only modestly changed their predictions on AGI timelines over the last decade. note also that (on a

The post contains two distinct snippets. First, user @GaryMarcus asks whether there is a more polite way to refer to “circle jerk.” Second, it shares a (likely satirical) claim that Microsoft’s AI rev

so do i have this right? the market is up on optimism about Microsoft’s revenue but MSFT’S biggest customer by far is burning billions a month with no obvious way to meet all of its future obligations

🚨The endgame of commoditization and falling prices and lack of a moat that I warned was inevitable in August 2023 took three years. But that moment has come. TBD whether OpenAI and Anthropic can survi

Welcome to the thunderdome of commoditization 🚨The endgame of commoditization and falling prices and lack of a moat that I warned was inevitable in August 2023 took three years. But that moment has co

A pioneer of AI-in-biology just published a sharp rebuttal to the 'AI will cure cancer' narrative. The core argument: It's a measurement problem. Over 90% of drugs that enter clinical trials still fai

can anyone with sufficient training in mathematics refuse this provocation? no ad hominem please; I have seen enough of that and am looking for real counterarguments, only. 🙏 OpenAI and Anthropic each

can anyone with sufficient training in mathematics refute this provocation? no ad hominem please; I have seen enough of that and am looking for real counterarguments, only. 🙏 OpenAI and Anthropic each

Gary Marcus tweeted on August 4, 2026 at 3:53 PM that a forthcoming weekend presentation will explore how chess illustrates the limitations of generative AI systems—highlighting an image produced by C

fascinating comments on possible slowdown – but probably a red herring. - Anthropic and OpenAI presumably aren’t colluding to slowdown, they are asking governments to build a framework such that a slo

Gary Marcus warns that, with increasingly sophisticated generative‑AI tools, future generations may struggle to tell what is real from what is fabricated. Matt Stoller adds that generative AI will tra

Gary Marcus commented on Twitter that OpenAI “distilled” Google’s brand name, prompting a humorous response from Tony Carden. Carden’s reply referenced Google DeepMind’s Project Astra and included a s

Gary Marcus tweeted “OpenAI: No Anthropic: Unlikely,” suggesting that the company does not expect Anthropic to satisfy particular requirements. The tweet references a discussion about the “RPO obligat

Part of the challenge with the AI math advances is the current air of secrecy around them. We don't know how Astra/Fable/Mythos etc are trained. We know many mathematicians have been paid to create tr

“poorly evidenced sensationalism makes for bad policymaking, politics, philanthropy and much else.” beautiful coda for a weekend in which people went completely nuts without even asking for a control

There should be a public investigation into the AI hacking incidents by OpenAI and Anthropic. We deserve to know whether these labs are genuinely world-class security organizations facing a novel thre

this is the real reason people from OpenAI etc are desperate to shut me up. Astra (which didn’t even have a control group and is maybe not that much better than Fable and certainly not ASI) was perhap

Checkmate: you can’t take the harness (which is typically in large part symbolic) away from the neural model without giving up performance. HUGE victory for neurosymbolic AI, straight from @AnthropicA

exactly. math isn’t done. not at all. I don’t think being critical of the amazing work AI is doing in pure math is fair to @OpenAI until I can start to say why I feel it’s not yet at the level of our

LLMs can know a task is impossible and still optimize it anyway. Ask whether to walk or drive to a car wash 50 meters away, and some models focus on distance while missing that the car itself must rea

OpenAI guy lies about my intent. I would absolutely love to see progress in AI for science and medicine. I have said that here, in my books, on countless podcasts, in multiple NYT opeds, in the US Sen

This wins the prize for sleazy misrepresentation. @mattShumer took my 2023 argument for hybridizing LLMs with symbolic tools – which is *exactly* what everyone does nowadays – and made it sound like I

Yet another paper argues that LLMs aren’t close to doing real discovery. MIT and Harvard argue LLMs are nowhere near doing real scientific discovery. They published a paper called “Evaluating Large La

If Leopold had read this on June 26 and trimmed his bets accordingly, SALP would not have melted down. I laid everything out. https://open.substack.com/pub/garymarcus/p/the-month-generative-ai-lost-it

The “AGI-is-near” community keeps committing the same logical fallacy over and over; I have seen it at least half a dozen times today alone. Every time there’s an advance, I see the same error. Here’s

Anthropic employee vouches for fiancee of Dario’s chief of staff who holds a lot of Anthropic stock… Where do the Anthropic employees get their media training, exactly? Prediction: SALP will be bigger

Everyone’s going on about how smart Leopold Aschennbrenner is (or was). But 1. He obviously didn’t know anything at all about risk management. (Or arrogantly chose to disregard whatever he might have

this is the whole problem in a nutshell. LLM-centered systems just can’t be trusted to follow hard constraints, we absolutely must find alternatives that can, or we are screwed. Anybody remember this

Situational Awareness, the $20bn hedge fund founded by former OpenAI employee Leopold Aschenbrenner, has sought to raise fresh capital from investors after suffering heavy losses during the recent rou

// The agent is its own best speculator // Agents spend a large share of wall-clock time waiting on tool results. Speculation hides that latency by predicting and pre-executing the next call, but exte

Gary Marcus has argued that both OpenAI and Anthropic must increase their pricing, citing the rapid decline in token costs and competition from open‑source models. He presents a binary choice: either

Gary Marcus shared a “handy dandy AI crisis flowchart” created by Kate Klonick (@Klonick) on Saturday, July 28. The tweet also notes that he had planned to write an article about Hugging Face and Open

The problem: agent workflows alternate between GPU-heavy reasoning and GPU-idle waiting on tools. Run hundreds concurrently and their KV caches fight for memory. Engines evict on a dumb LRU policy, ev

We ran a large-scale distillation attack on the Kimi K3 technical report by reading it in parallel at the Hugging Face Journal Club :) https://youtu.be/MW8-kqd2SD8?si=jSKDogcUWbJ8N2k7 Our main takeawa

AGI smarter than the smartest humans, my ass. @Kasparov63 (peak rating 2851) probably could’ve crushed the best commercial large language models in chess when he was 7 years old. graph courtesy @chess

The administration is very consistent: All aspects of foreign involvement in AI are banned: people (students and employees), hardware and models. And funding is down to a drip and capital allocation i

Why is this breaking news? And what the heck does it actually mean? Has anyone actually defined it? Is it anything more than a ruse to distract from a disturbing hack and falling confidence in AI? BRE

Great technical paper from Harvard and MIT. It's on role drift in compound LLM systems. (bookmark it) End-to-end RL improves the accuracy of a multi-module LLM pipeline without constraining how the mo

If multiple Mythos level models are out there in the hands of attackers, now or in a few years, then US builders and software developers need strong, open defenses. They cannot have gated, crippled de

New research from NVIDIA. They just dropped a PyTorch-native training framework for agentic RL. (bookmark it) Paper summary: Molt is a PyTorch-native agentic RL framework with an unusual design target

Stocks should be ripping today given the move in oil. But they're not thanks to NVDA and it's unbelievable 250 Billion roundtrip with OpenAI. There's no hiding anymore that the whole AI bubble is real

Very cool paper from Microsoft. The idea is to train agents on replayed teacher trajectories instead of live environment rollouts. On-policy distillation for agentic tasks is expensive because every u

important thread; don’t read just the first tweet (which has a caveat in the second). In case you think the problem of science slop is hypothetical: Here's the president of OpenAI retweeting wrong sci

training to benchmarks ≠ getting to AGI This suggests that Opus 5 's gain on ARC-AGI-3 was the result of specific training to improve on that eval, and not a generalized increase in abstract reasoning

All anyone has to do is go back 15 years and look at all the ridiculous predictions Elmo has made about Mars, self driving cars, his ridiculous Boring Company. No one should take anything he says seri