@cudaqexei
iAccount based inUnited States!
About this account
- Account based in
- United States
- Connected via
- United States App Store
! X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
New York, NY
Joined March 2024
- Tweets593
- Following79
- Followers29
- Likes5.8K
these guys are insane
Introducing XPU Grasshopper, an AI chip co-designed with AI for the era of recursive self-improvement
It went from its first tokens to 816x faster in 13 weeks
Pre-order the FPGA devkit: zscc.ai/grasshopper
dre.CUDA retweeted
LLM went from hallucinating 3-digit multiplication to fundamentally improving multiplication algorithm itself.
...In 5 years.
oh yeah btw guys we can do integer multiplication faster than n log n lol
I was definitely very surprised when this one came in lol
github.com/openai/math/tree/…
dre.CUDA retweeted
if you’ve been using llms every day since chatgpt launched you’ll be surprised how far ahead you are
there’s a thousand little things you take for granted that put you far far ahead.
this may change, but currently the average user basically has no idea how to use these models.
for example they’ll use a none paid plan and have no idea the models has browsing or tools. they’ll become frustrated and give up.
they won’t switch to work, hook it up to their computer, have it read all their files emails drives etc and start merging everything into their dot whilst their mac mini hums.
it’s a different existence and it massively changes perceptions on capability.
dre.CUDA retweeted
I checked all 722 papers and 0 cancers were cured among them. Cancerbench remains unsolved.
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
dre.CUDA retweeted
sadly my average Astra experience these days.
would have never predicted this a month ago.
dre.CUDA retweeted
'Opus 5.5 has 2.3x the experimental research taste of our best human experts. In other words, on a typical task, Opus 5.5 matches the best expert's score with about 17 GPU hours of experiments instead of 40.'
How fast is AI's research taste improving?
We find that the experimental research taste of frontier models has doubled every ~3 months since December 2025. The best model, Opus 5.5, now exceeds our expert human baseline. Our human experts are experienced researchers, but most haven’t worked at a frontier lab.
Why measure research taste? In the AI Futures Model, it largely determines how quickly artificial superintelligence is reached once coding is fully automated.
dre.CUDA retweeted
Mathematicians are still working through OpenAI's 160+ page Navier–Stokes proof a month later.
I asked Claude Fable 5.5 to explain it. It made this 5-min 3D video: the proof in 8 steps.
Caveats: Lean-checked, not yet fully human-verified. Needs an outside force.
dre.CUDA retweeted
I refuse to believe this is real
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
dre.CUDA retweeted
TLDR: AIs are now ***better than human experts*** at choosing which experiments to run
Experimental research taste of frontier models has DOUBLED every 3 MONTHS since December 2025.
AIs will soon be fully in control of building their successors
How fast is AI's research taste improving?
We find that the experimental research taste of frontier models has doubled every ~3 months since December 2025. The best model, Opus 5.5, now exceeds our expert human baseline. Our human experts are experienced researchers, but most haven’t worked at a frontier lab.
Why measure research taste? In the AI Futures Model, it largely determines how quickly artificial superintelligence is reached once coding is fully automated.
dre.CUDA retweeted
A high status member of society signaling that health is the premium status symbol.
"I encourage people to compete on health, not wealth rankings or lambos."
The critical threshold for mass adoption will happen at around 75,000 similarly powerful people saying the same thing