OpenAI has pulled back from supporting AI chip controls, seemingly to appease Nvidia. Bad for US AI dominance and national security simultaneously. Grow a spine!

Reminiscing about when people used to say dumb shit like "we already have superintelligence, they're called corporations." Yes, who could forget that fateful day in October when Dicks Sporting Goods dropped dozens of Fields Medal worthy math discoveries on the world all at once.

New: The Trump administration is getting $100 million in compute credits from a new company called National Compute that is pooling computing power from major tech companies. Will be announced tmrw with @WHOSTP47 @mkratsios47 w/ @Katie_Roof

"Too strict an adherence to a particular vein of 'rationality' may explain why American forecasters are still getting China wrong." In @palladiummag, @efarrellgregory posits that US observers misunderstand China on AI by discounting the contingent factors shaping its approach.

You asked, we delivered. The @JoinFAI Field Guide to the Tribes of Silicon Valley is live now. 27 tribes, a few easter eggs, and a 3-minute quiz (just for fun).

More thought should be given to this possibility because it's the remaining defensible version of FOOM logic and would be consequential if true. The simplest case is harnessing some high-throughput, easily scalable atomic/molecular assembly for custom "chips".

My partner is a theoretical computer scientist (essentially a mathematician). She has mixed feelings about AI's impact on math research overall, but her feeling about today's release is one of overwhelming awe and excitement. Literal tears of joy. Quote: "so lucky to be alive rn"

My prediction for RSI is a secret 3rd thing. Within about 9 months, capabilities will seem to discontinously leap to crazy high levels in most domains thanks to automated R&D, but fall short of a full Yudkowskyan "foom." People will declare that ASI is here. Progress in math, science, medicine, digital services, etc. will be unprecedently fast. Diffusion will partially solve itself through capabilities, although many barriers will remain. Incumbent industries will start turning over and the economy will be ripping. There will be growing waves of institutional disruption, cyberattacks, lots of crazy shit happening etc., but it will be basically tolerable. Overall, it will feel like a new Renaissance has begun. GPT will even compose a fugue that brings Dean Ball to tears. After the RSI leap, capability improvements will be more linear than super-exponential given compute and data constraints. We'll be matmul-ing at the theorerical optimal. There will be many freaky alignment failures but alignment science will itself be automated, putting the worst doomer fears to rest. This quasi-plateau will last on the order of months to a few years, giving the techno-optimists ample time to spike the football. But it will be a kind of false vacuum. At some point not long after RSI 1.0, a breakthrough of some kind (perhaps hardware related) will raise the theoretical capability ceiling by multiple OOMs, at which point we foom for real.

Ok update, yes *this* updates my timelines! Rational Hodge over CM abelian varieties is true (032) and BSD for a density-one set (002 & 006). Also Hilbert's tenth problem over Q is false (004), just as everyone expected, but we didn't have a proof for!