Pinned post
Grok 4 is live: x.com/i/events/19427…
War Room squad locked inView quoted post on X
Latest posts 5
Here I simply recommend every chip or perf engineer to chat with @nilscmr to have some more clarity of what future (workload) will look like with hardware abundance
congrats @TendrilsCompute for the launch! (photos from sitsit while visiting @shipfr8 where the company was born. insane energy)
congrats @TendrilsCompute for the launch! (photos from sitsit while visiting @shipfr8 where the company was born. insane energy)
View quoted post on X
have people tried these new languages?
agentlanguages.dev
agentlanguages.dev
one less surf, one more ablation
I had an incredible time training this model. tldr on some nonobvious things I learned:View quoted post on X
- Dense models can be surprisingly competitive with much large MoEs in downstream evals! Matching perplexity doesn't necessarily mean matching downstream capability; our dense model often performs better, due to a variety of factors (attention density, less memorization, etc.)
- Once you solve your infrastructure and get your synthetic environment pipelines working well, mixed RL is actually extremely efficient at improving your model. Can't wait to continue scaling posttraining!
- You can use swarms of small agentic models to turn qualitative questions about your datasets into programmable measurements. This creates a powerful feedback loop for systematically improving data quality and climbing tasks
THE 65B BABY IS BORN
it was a wild ride for a team of X to create this model and the memory at @RicursiveAI still last. This February, I wrote my first line of code for optimizer and data-mixing experiments for pretraining and @DZhang50 is a living one-person pretraining legend. it was fun to bootstrap our posttraining infra from scratch by combining best open-source practices, restructuring to be agent-friendly, and being more token budget and cost-aware across multiple stages of SFT and RL. I didn’t get to finish the model training before I left, but the benchmark numbers are impressive @esonghori2 @ethanboneh @talaei_shayan are the great minds behind synth data, big run infra, and countless recipe studies
Can’t wait for the team to continue the mission on open-source model and AI for chip design 💙 Rooting for Ripple
it was a wild ride for a team of X to create this model and the memory at @RicursiveAI still last. This February, I wrote my first line of code for optimizer and data-mixing experiments for pretraining and @DZhang50 is a living one-person pretraining legend. it was fun to bootstrap our posttraining infra from scratch by combining best open-source practices, restructuring to be agent-friendly, and being more token budget and cost-aware across multiple stages of SFT and RL. I didn’t get to finish the model training before I left, but the benchmark numbers are impressive @esonghori2 @ethanboneh @talaei_shayan are the great minds behind synth data, big run infra, and countless recipe studies
Can’t wait for the team to continue the mission on open-source model and AI for chip design 💙 Rooting for Ripple
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
