@garymarcus.bsky.social
AI and cognitive science, Founder and CEO (Geometric Intelligence, acquired by Uber). 8 books including Guitar Zero, Rebooting AI and Taming Silicon Valley. Newsletter (100k subscribers): garymarcus.substack.com
Stick it up *your* ass, chatbot.
www.404media.co/rfk-jrs-nutr...
Great reference from @garymarcus.bsky.social to this comprehensive paper.
arxiv.org/pdf/2602.06176 "Large Language Model Reasoning Failures"
Besides many other important finding, this was pretty clear:
@garymarcus.bsky.social:
I have a hard time seeing how this doesn’t end badly. A recession, maybe worse.
Wow! @garymarcus.bsky.social has made the most important German daily newspaper! Congratulations on getting the message out there!
www.sueddeutsche.de/kultur/kuens...
some things never change
11.02.2026 02:04 — 👍 46 🔁 9 💬 4 📌 1
Der Neurowissenschaftler und #KI- Unternehmer @garymarcus.bsky.social warnt seit Jahren, dass der Boom der »Künstlichen Intelligenz« auf einem Missverständnis beruht. Jetzt häufen sich auffällig die Stimmen, die auf seine Linie einschwenken: archive.is/3PFlk #Interview
08.02.2026 21:31 — 👍 117 🔁 35 💬 3 📌 1they also make money from Krugman, Reich, Vance, Snyder Cadwalladr etc.
Free speech is a complicated beast.
Anthropic vs OpenAI Superbowl smackdown!
open.substack.com/pub/garymarc...
Four theories about the SpaceX merger: garymarcus.substack.com/p/four-theor...
05.02.2026 20:13 — 👍 24 🔁 4 💬 5 📌 0best not miss!
05.02.2026 19:38 — 👍 1 🔁 0 💬 0 📌 0Is the Great AI meltdown imminent? A huge shift from Nvidia doesn’t bode well for OpenAI or Generative AI maximalism:
05.02.2026 19:37 — 👍 31 🔁 8 💬 1 📌 1
What will A.I's impact be on scientific research? Melanie Mitchell Computer scientist “I believe this impact will not be as rapid as many think. A.I. systems still can't do things humans are essential for, like asking the right questions, planning experiments and entire scientific programs, and understanding data in different contexts.”
Gary Marcus Cognitive scientist “People are greatly confused about large language models, attributing a humanlike intelligence to mimicry machines that turn out to be superficial and unreliable. Intelligence is about reasoning flexibly in face of the unknown, and large language models continue to struggle mightily with that.”
Helen Toner A.I. policy researcher “I expect we'll have A.I. systems that can clearly contribute at the cutting edge of multiple scientific fields, but that you still wouldn't trust with planning summer camps for your kid.”
@melaniemitchell.bsky.social @garymarcus.bsky.social @hlntnr.bsky.social on where AI won’t be taking science (“asking the right Qs, planning experiments &entire sci programs,&understanding data in diff contexts”), intelligence (“reasoning flexibility in the face of the unknown”) &summercamp madness🧪
02.02.2026 14:20 — 👍 16 🔁 8 💬 2 📌 0Hehe, quoting @garymarcus.bsky.social "if you care about the security of your device or the privacy of your data, don’t use OpenClaw. Period."
(Bonus : if your friend has OpenClaw installed, don’t use their machine. A. Don’t catch a CTD — chatbot transmitted disease)
OpenClaw (aka Moltbot) may have gone completely viral, but it is also a disaster waiting to happen.
A PSA for why you should steer clear.
open.substack.com/pub/garymarc...
but i don’t really miss X, tbh
01.02.2026 05:54 — 👍 3 🔁 0 💬 0 📌 0they cannot be trusted and it’s a security nightmare and in the making
31.01.2026 18:36 — 👍 2 🔁 0 💬 1 📌 0my own take is explained in detail here: bsky.app/profile/kost...
31.01.2026 18:35 — 👍 18 🔁 6 💬 3 📌 0strongly strongly urge you to watch this, for a different perspective:
bsky.app/profile/kost...
The fundamental mistake of people like @chrislhayes.bsky.social is to ignore and fail to engage with the principled skepticism of LLMs by people like Ilya Sutskever, Rich Sutton, Yann LeCun, Demis Hassabis, Melanie Mitchell and myself that has been increasingly vindicated.
31.01.2026 18:32 — 👍 41 🔁 9 💬 6 📌 0
This conversation between Steve Eisman (who predicted the last big financial bubble) and AI researcher @garymarcus.bsky.social is a great resource for those trying to understand the distinctions between LLMs and other more defensible AI endeavors. www.youtube.com/watch?v=aI7X...
30.01.2026 17:58 — 👍 22 🔁 3 💬 1 📌 2Even the conservative Wall St Journal sees that Trump and
Miller’s goons have gone far too far.
They recruited the comments section and gave them weapons and sovereign immunity.
24.01.2026 18:19 — 👍 305 🔁 71 💬 5 📌 4firing and incarcerating the officers who did this is not enough.
ICE is rotten from the top down.
yesterday’s murder wasn’t an accident. it was a culture.
Murder in broad daylight, by the U.S. government.
www.nytimes.com/interactive/...
will Minnesota be Trump’s undoing?
25.01.2026 02:48 — 👍 35 🔁 2 💬 4 📌 0*I* am not actually the one who is confused.
24.01.2026 20:51 — 👍 2 🔁 0 💬 1 📌 0“the federal government cannot be trusted to lead this investigation.”
- Tim Walz with the quote of the year, though honestly at this point you could just cut the last four words.
Finally somebody who understands me!
24.01.2026 16:03 — 👍 53 🔁 4 💬 5 📌 0World models are necessary but not sufficient.
Pure LLMs don’t have them at all.
i of course have been pushing for them for many years.
"With the help of Meta’s media lobby he has succeeded in fooling most of the press and some fraction of the public. LeCun has lapped it up, and done absolutely nothing to set the record straight.
But the myths about the originality of his thought simply aren’t true." - @garymarcus.bsky.social
