Asked Gemini to critique my code. Realized I can't trust it because Shazeer himself probably biased the training data

will work for a16z workstation 🥵

the devil works hard but openGVlab works harder

Coding agents write code so fast that PRs are huge. I don't have time to read it all carefully. Now I need a coding agent just for review 😑 the agents are perpetuating themselves

Introducing the Environments Hub RL environments are the key bottleneck to the next wave of AI progress, but big labs are locking them down We built a community platform for crowdsourcing open environments, so anyone can contribute to open-source AGI

AWS IAM is mental illness people in the future will see what we were doing and call us neanderthals

asking GPT5 a question and then GPT5-Thinking feels like the first half of Flowers for Algernon

Fuck it. Today, we open source FineVision: the finest curation of datasets for VLMs, over 200 sources! > 20% improvement across 10 benchmarks > 17M unique images > 10B answer tokens > New capabilities: GUI navigation, pointing, counting FineVision 10x’s open-source VLMs.

openAI hit a critical threshold with Codex: even if it's not quite as good as claude code, it's close enough that I'd rather only spend the $200 at OpenAI.

The pi-05 model is now in openpi: github.com/Physical-Intel… Now with pytorch (πtorch?) support too!

Anyone in Vegas can take a ride with Zoox! The first time rides in our robotaxi have been generally available 🥳

Excited to release a preview of Moondream 3. A 9B param, 2B active MoE vision language model that makes no compromises; offering state-of-the-art visual reasoning while still retaining an efficient and deployment-friendly form factor.

you know you're having a bad day when you start putting shit like this at the top of your code torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False torch.backends.cuda.matmul.fp32_precision = "ieee"

when chatgpt and Claude refuse but Grok answers immediately

This is the 5th shot! Super crazy! We opensourced a 235B-A22B Instruct and Thinking Qwen3-VL models under Apache 2.0! Qwen3-VL, the new generation of our vision-language model, whose previous version was released a long time ago. During these days, we have conducted a lot of

qwen3-vl > ... we shift from the conventional... practice of injecting visual tokens into only a single layer, to injecting them across multiple layers of the large language model

Qwen3-VL-235B-A22B is a beast. I'd give you my vibe check, but at 471GB, it's still downloading 😅 vLLM and 🤗 transformers reference a 30B-A3B MoE and 32B, 4B dense models. I expect almost everyone will use these dramatically smaller sizes. Look forward to trying them out!

super excited to have Krish Shah on the team!