Dev Patel

ALS has gradually taken away Kenneth’s ability to speak. Through Neuralink’s VOICE clinical trial, he’s exploring how a brain-computer interface designed to translate thought to speech could help restore autonomy in his daily life. Watch to learn more:

The RL training infra for Composer 2 was built with Ray. Very impressive work.

@_hansonw @arjunmgurjar the goat @arjunmgurjar!

Crazy win man

@krupaad @nvidiaomniverse awesome project!

Had a fun time building SimSafe for the @nvidiaomniverse Cosmos Cookoff!! AV companies generate millions of synthetic training clips, but bad data with broken physics and unrealistic footage, degrades model performance. SimSafe uses Cosmos Reason 2 to automatically detect physically implausible synthetic AV training data from dashcam footage (by looking at shadow consistency, vehicle dynamics, road texture realism).

Luckin Coffee (Chinese Starbucks) acquired Blue Bottle retail operations for $400M from Nestle, who had bought a majority at a ~$700M valuation back in 2017. Blue Bottle had raised from True, Index and GV. Nestle will keep the FMCG brand. Luckin was delisted from the Nasdaq in 2020 when it came out that they had fabricated $310M of sales, and paid a $180M fine. They seem to have turned it around and might get relisted after 5 years in the penalty box. It’s trading at ~$10B on OTC markets en.sedaily.com/international/…

We’ve been consistently surprised lately by how capable frontier models are at handling complex kernel implementation and system optimization. Check out this work as a step toward automating AI infrastructure building!

how do u fix openclaw internal reasoning leaking

@willccbb thank u for ur good taste in music will brown

now listening

@NovaSkyAI Personally, SkyRL is the best code repo for agentic RL. We have tested lots of other RL codebase and their examples can not even run correctly. Respect!

🔥Modifying 2 lines of code and get your agentic serving/rollout up to 3.9x faster losslessly! ⚡️Say hello to ThunderAgent, a fast, simple, and program-aware agentic Inference System. 🥇 We propose a program abstraction to schedule all GPU and CPU resources, the first principled approach for distributed agentic inference and rollout. 🌐 Blog: thunderagent.ai 💻 Code: github.com/ThunderAgent-o… 📜 Paper: arxiv.org/pdf/2602.13692 #AI #ThunderAgent #LLMAgent #Mlsys 1/n

Excited to see SkyRL being used by systems research to study how agentic RL workload can be optimized!! github.com/ThunderAgent-o…

@sakshambatraa Awesome work saksham!

i spent the last few months building microMLC, a machine learning compiler without any prior experience. the result is an interactive educational resource that documents my journey and findings. here's a breakdown of what i learned!

@suryasure05 no ascii diagrams surely 🙏

if you haven’t tried using codex or claude code to generate model architecture diagrams (with tensor size annotations too), you should seriously give it a shot

SkyRL 🤝 Harbor is the best way to train terminal agents.

Source: Benchmark's 2020 fund is now worth 10x+ and 2024 fund is 3x what investors put in, based on cash distributions and the paper value of its investments (@nmasc_ / Bloomberg) bloomberg.com/news/articles/… #a260217p46" target="_blank" rel="nofollow noopener">techmeme.com/260217/p46#a26… 📥 Send tips! techmeme.com/contact

🔥Excited to see SkyRL bringing Tinker to local GPUs. Standardizing training APIs lowers the barrier for research and infrastructure innovation. vLLM is proud to power the inference layer behind high-throughput RL training. 🚀

@fazalmittu_ very cool, love the figure you made!

recently been interested in Meta's JEPA, an architecture for predictive models to be able to extract/understand higher level details from any modality of data. walkthrough on my website: fazalmittu.com/reports/ijepa going to be doing more of these going forward!

super excited about our release of the tinker API in skyrl! come try it out for your next training run or if you're interested in building an open source Tinker engine!

We are excited to announce that SkyRL now implements the Tinker API. Run Tinker training scripts on your own hardware with zero code changes. Try it out today: novasky-ai.notion.site/skyrl-tinker

We’re super excited to have SkyRL be fully implementing Tinker!