Top all-time · #220
Gary Marcus
Personal blog · garymarcus.substack.comaiscience
Gary Marcus writes Marcus on AI, a personal blog at garymarcus.substack.com about ai and science. Since 2022, 214 different Hacker News readers have linked it 428 times, most often “The whole thing was a scam”. It is still writing: the last post was on 2026-10-31, and you can follow it by feed.
Sometimes recommended · 8 of 40 sampled comments recommend it?
Why HN links it
HN readers link Gary Marcus's Substack as a skeptical critique of LLM capabilities, scaling claims and hype, covering reasoning limits, hallucinations, predictions and copyright.
- Recommended for · 8 comments
- Called a good job exposing LLM limits, interesting on neurosymbolic AI, right on symbolic AI, worth reading on AI harms.
Summarised by Claude from a sample of 40 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
"Marcus has become one of our few indispensable public intellectuals. The more people read him, the better our actions in shaping Al will be." - Kim Stanley Robinson, author of Ministry for the Future. Click to read Marcus on AI, by Gary Marcus, a Substack publication with hundreds of thousands of s
Most-linked posts?
- The whole thing was a scam6 people · 6 mentions · first linked 2026-02-28
- Horse Rides Astronaut (On Imagen/DALL-E)5 people · 5 mentions · first linked 2022-06-19
- o3 and Grok 4 accidentally vindicate neurosymbolic AI5 people · 5 mentions · first linked 2025-07-13
- Alt Intelligence vs. AI (Gary Marcus Critique of DeepMind’s de Freitas)5 people · 5 mentions · first linked 2022-05-15
- Did GoogleAI just snooker one of Silicon Valley’s sharpest minds?4 people · 5 mentions · first linked 2022-09-15
- /4 people · 5 mentions · first linked 2022-11-20
- Dear Elon Musk, here are five things you might want to consider about AGI4 people · 4 mentions · first linked 2022-05-31
- Are large language models a good model of *human* language?4 people · 4 mentions · first linked 2022-05-24
- OpenAI's Waterloo?4 people · 4 mentions · first linked 2025-08-14
- LLMs don't do formal reasoning4 people · 4 mentions · first linked 2024-10-11
What Hacker News says?
You might like Gary's blog on potential AI harms: https://garymarcus.substack.com/
Comment linking /
on “GPT-4o”Submitted ChatGPT went berserk as a story · 432 points
HN discussionTrue. At a minimum, as long as LLMs don't include some kind of more strict representation of the world, they will fail in a lot of tasks. Hallucinations -- responding with a prediction that doesn't make any sense in the context of the response -- are still a big problem. Because LLMs never really develop rules about the world. For example, while you can get it to predict good chess moves if you train it on enough ch…
Comment linking /p/generative-ais-crippling-and-widespread
on “GPT-5”Submitted Things are about to get worse for generative AI as a story · 403 points
HN discussionI know that this seems likely from a theoretical perspective (in other words, I would way underestimate it at the sprint planning meeting!), but A) checking each output against a regex representing a hundred years of literature would be expensive AF no matter how streamlined you make it, and B) latent space allows for small deviations that would still get you in trouble but are very hard to catch without a truly lat…
Comment linking /p/things-are-about-to-get-a-lot-worse
on “Meta torrented & seeded 81.7 TB dataset containing copyrighted data”Submitted LaMDA is not sentient as a story · 366 points
HN discussionPersonally I would say the disparity between what was in their founding agreement "be open-source—freely and publicly available for anyone to use, modify and distribute" https://archive.ph/R0LBL to the current state of affairs. But I guess the list of grievances could be longer: https://garymarcus.substack.com/p/what-should-we-learn-from-...
Comment linking /p/what-should-we-learn-from-openais
on “Apple Intelligence for iPhone, iPad, and Mac”Submitted Seven replies to the viral Apple reasoning paper and why they fall short as a story · 343 points
HN discussionActually, I think the industry is finally realizing that there is no intelligence there, especially with gpt3 which can figure out with great precision what statistically comes next, but there is zero understanding of the space-time conceptual meaning for gpt3 in that answer - its not designed to do anything but figure out statistically what is most likely to come next. Gary Marcus has been doing a good job of expos…
Comment linking /
on “Kite is saying farewell and open-sourcing its code”Submitted The OpenAI board was right as a story · 340 points
HN discussionHey, I'm definitely on your side of the Great AI Wars--and definitely share your thoughts on the overall framing--but I think you're missing the serious nature of this contribution: 1. Small correction, it's actually a whole book AFAIK, and potentially someday soon, a class! So there's a lot more thought put in then the typical hot-take blog post. I also pop into one of these guy's replies on Bluesky to disagree on…
Comment linking /p/llms-dont-do-formal-reasoning-and
on “Modern-Day Oracles or Bullshit Machines? How to thrive in a ChatGPT world”Submitted GPT-5: Overdue, overhyped and underwhelming. And that's not the worst of it as a story · 304 points
HN discussionhttps://garymarcus.substack.com/p/the-last-few-months-have-b... nope
Comment linking /p/the-last-few-months-have-been-devastating
on “ChatGPT Atlas”Submitted Did GoogleAI just snooker one of Silicon Valley’s sharpest minds? as a story · 264 points
HN discussionhttps://garymarcus.substack.com/p/elon-musks-terrifying-visi... I'm not sure if this was a very bad joke by Elon, or if Grok 3 is really biased like that.
Comment linking /p/elon-musks-terrifying-vision-for
on “Grok3 Launch [video]”Submitted Hot take: GPT 4.5 is a nothing burger as a story · 256 points
HN discussionBased on recent rumblings about AI scaling hitting a wall, of which this article is perhaps the most visible - and in a high-reach financial publication, I'm considering increasing my estimated probability we might see a major market correction next year (and possibly even a bubble collapse). (example: "CONFIRMED: LLMs have indeed reached a point of diminishing returns" https://garymarcus.substack.com/p/confirmed-ll…
Comment linking /p/confirmed-llms-have-indeed-reached
on “OpenAI, Google and Anthropic are struggling to build more advanced AI”Submitted Sam Altman's pants are on fire as a story · 210 points
HN discussionThey pirate movies as well: https://garymarcus.substack.com/p/an-artist-fights-back-and-...
Comment linking /p/an-artist-fights-back-and-midjourney
on “NY Times copyright suit wants OpenAI to delete all GPT instances”Submitted Bill to block OpenAI's for-profit conversion gets mysteriously gutted as a story · 201 points
HN discussionDiscussion of this article by Gary Marcus: https://garymarcus.substack.com/p/a-knockout-blow-for-llms
Comment linking /p/a-knockout-blow-for-llms
on “The Illusion of Thinking: Strengths and limitations of reasoning models [pdf]”Submitted LLMs and coding agents are a security nightmare as a story · 194 points
HN discussion> Do you honestly think that once we're able to build AI that _fully_ mimics humans by every measurement we have, What made you think that is measurable and if it is then we can build something like that ever ? I already linked https://garymarcus.substack.com/p/this-one-important-fact-ab... did you read it?
Comment linking /p/this-one-important-fact-about-current
on “TSMC execs allegedly dismissed OpenAI CEO Sam Altman as 'podcasting bro'”Submitted OpenAI probably can't make ends meet. That's where you come in as a story · 170 points
HN discussion> If humans can be fooled by artificial intelligence in pretty much all areas, This is just preposterous. You can be fooled if you have no knowledge in the area but that's about it. With current tech there is, there can not be anything novel. Guernica was novel. No matter how you train any probabilistic model on every piece of art produced before Guernica it'll never ever create it. There are novel novels (sorry for…
Comment linking /p/this-one-important-fact-about-current
on “TSMC execs allegedly dismissed OpenAI CEO Sam Altman as 'podcasting bro'”Submitted LLMs don't do formal reasoning as a story · 144 points
HN discussionImpressive prediction, especially pre-ChatGPT. Compare to Gary Marcus 3 months ago: https://garymarcus.substack.com/p/reports-of-llms-mastering-... We may certainly hope Eliezer's other predictions don't prove so well-calibrated.
Comment linking /p/reports-of-llms-mastering-math-have
on “OpenAI claims gold-medal performance at IMO 2025”Submitted LLMs have reached a point of diminishing returns as a story · 131 points
HN discussionIt's not about it being theoretical, it's moreso that the language model is still far more simplistic than our own, and struggles with anything but the most basic relations between nouns. The "horse riding an astronaut" post is a good example of this.[0] [0] https://garymarcus.substack.com/p/horse-rides-astronaut
Comment linking /p/horse-rides-astronaut
on “DALL·E 2 vs. $10 Fiverr Commissions”Submitted The "AI 2027" Scenario: How realistic is it? as a story · 129 points
HN discussion
