morgan —
@morqon
deep yearning
For humans, mathematical symbols (and formal systems like lean) are *tools* we learn how to use, not a structure that wraps around us. I think that's the right role for formal still manipulation: a tool that can be employed by an intelligent system if/when it supports a goal.
new kinda guy who finds out about A.I. data center water consumption and stops drinking water so the data centers can have more
the american predilection for ai pessimism
A whole lot of people are telling you that AI is going to kill jobs and increase inequality. They have no idea what they're talking about. noahpinion.blog/p/stop-pretend…
My team at RAND is hiring! Technical analysis for AI policy is desperately needed. Particularly keen on ML engineers and semiconductor experts eager to shape AI policy. Also seeking excellent generalists excited to join our fast-paced, impact-oriented team. Links below.
New Anthropic Research: “Inverse Scaling in Test-Time Compute” We found cases where longer reasoning leads to lower accuracy. Our findings suggest that naïve scaling of test-time compute may inadvertently reinforce problematic reasoning patterns. 🧵
of course he did
a day after Google celebrated its gold medal achievement in the International Math Olympiad, three of the researchers who worked on that team are already leaving for Meta scoop via @KalleyHuang and me theinformation.com/articles/meta-…
New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵
we’re continuing to expand Stargate further, and expect to exceed the $500B commitment announced at the White House in January thanks to strong momentum with partners including Oracle and SoftBank.
Congrats to GDM on the parallel gold medal result! After hearing from the IMO board member, I made the call to wait until after the closing ceremony since "it'll look like bad form with respect to the IMO and other labs to front run." There wasn't ill intent.
On IMO P6 (without going into too much detail about our setup), the model "knew" it didn't have a correct solution. The model knowing when it didn't know was one of the early signs of life that made us excited about the underlying research direction!
One piece of info that seems important to me in terms of forecasting usefulness of new AI models for mathematics: did the gold-medal-winning models, which did not solve IMO problem 6, submit incorrect answers for it? 🧵
do not build Infinite Jest (V), do not build the infinite AI TikTok slop machine, do not build the P-zombie AI boy/girlfriend, do not build the child-eating short-form video blackhole, do not build the human-feedback-optimized diffusion transformer porn generator. save yourselves
do not build Infinite Jest (V), do not build the infinite AI TikTok slop machine, do not build the P-zombie AI boy/girlfriend, do not build the child-eating short-form video blackhole, do not build the human-feedback-optimized diffusion transformer porn generator. save yourselves
the two cultures
Fascinating to compare the solutions of OpenAI vs Deepmind to the IMO 2025. Both won Gold for answering P1 to P5 correctly. OpenAI (left) vs Gemini (right)
I cannot emphasize this enough: the system use no tools, no lean — text in, text out. And the more we scale inference compute, the more accurate the proofs get, while still reading like natural text.
as an aside
Btw as an aside, we didn’t announce on Friday because we respected the IMO Board's original request that all AI labs share their results only after the official results had been verified by independent experts & the students had rightly received the acclamation they deserved
fast followers fast following
The xAI office just got a Grok-powered vending machine, thanks to our friends at Andon Labs! How much dough do you think Grok is gonna rake in in the next month?
According to reporting by the WSJ, there are at least ten employees at OpenAI who have turned down $300 million offers from Mark Zuckerberg.
systems that create new knowledge must do a few hard things well > they must generate lots of varied ideas, expose them to reality, discard most of them, and remember what worked > they must be able to question themselves - not just their outputs, but their methods > they must…