gemini's version of big bang - argon, carbon, hydrogen....
Divy Thakkar
Verified account This account is verified. Learn more@divy93t
co-lead for science and reasoning capabilities for Gemini (post-training) @GoogleDeepMind, advancing human-centered llms. Ph.D @CityStGeorges . Personal views.
2.2K Following 12K Followers 2.3K Posts
Latest posts 5
it is - try it!
Ok. Carbon is *really* good at code.View quoted post on X
Congratulations to the talent-*dense* team at Ricursive for shipping a great open model!
@DZhang50 @ethanboneh @Azaliamirh @annadgoldie
@DZhang50 @ethanboneh @Azaliamirh @annadgoldie
Today we release Ripple-65B Preview, our 65B dense LLM we pretrained from scratch.
Ripple sets a new state-of-the-art for U.S. open-weight models in agentic coding tasks, surpassing significantly larger models including Muse Spark Open-816B, Beam-501B, Inkling-975B, and Nemotron 3 Ultra-550B.
Learn more: ricursive.com/rippleView quoted post on X
Take a moment to appreciate that this finding (and its reproduction) came from academic labs - we need to continue to invest in curiosity-driven research. The shape / form of research will change - as in this case, where researchers probe large models - but developing research taste in the process is what it was all about!
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weight-sp…View quoted post on X
standouts
> slightly higher output token length for Gemini 4 Carbon at a third of the cost per task compared to Opus and Astra
> #1 for biology #2 for Materials
> the overall 1.1% difference in performance is highly subjective across fields -- see what works for yours and make a judgment call
> evals are guiding signals, not the final state
> slightly higher output token length for Gemini 4 Carbon at a third of the cost per task compared to Opus and Astra
> #1 for biology #2 for Materials
> the overall 1.1% difference in performance is highly subjective across fields -- see what works for yours and make a judgment call
> evals are guiding signals, not the final state
Today we’re releasing GDPx, a benchmark for measuring agentic capabilities across the most valuable domains.
GDPx is a true frontier-level benchmark: even the best model scores just 12% (1/7)🧵View quoted post on X
