Image created with Flux Pro v1.1 Ultra. Image prompt: Meta, beach bag and sunglasses, lenses reflecting horizon shaped like a smooth infinite loop, warm glow, photorealistic, editorial, minimal, landscape, vacation, no text overlays
Updated & turned my Big LLM Architecture Comparison article into a narrated video lecture. The 11 LLM architectures covered in this video: 1. DeepSeek V3/R1 2. OLMo 2 3. Gemma 3 4. Mistral Small 3.1 5. Llama 4 6. Qwen3 7. SmolLM3 8. Kimi 2 9. GPT-OSS 10. Grok 2.5 11. GLM-4.5 https://x.com/rasbt/status/1965798055141429523
What if you kept asking an LLM to “”make it better””? In some recent work at FAIR, we investigate how we can efficiently use RL to fine-tune LLMs to iteratively self-improve on their previous solutions at inference-time. Training for iterated self-improvement can be costly. The https://x.com/MinqiJiang/status/1965055909605916892
Meta hid harms to children from VR products, whistleblowers allege | Meta | The Guardian https://www.theguardian.com/technology/2025/sep/08/meta-virtual-reality-whistleblowers
Writing fast GPU kernels is important, though not nearly as important as writing correct ones. That’s why @marksaroufim and folks from Meta have released BackendBench. Now BackendBench also lives on @PrimeIntellect environment hub. 1/3 https://x.com/m_sirovatka/status/1965891832942047350
Introducing LlamaIndex Classify: Rules-Based Document Classification Made Simple Learn how to automatically classify your documents with LlamaIndex’s newest beta feature! In this quick demo, Laurie walks through the Classify service – a powerful tool for preprocessing documents https://x.com/llama_index/status/1963263366086172719
Writing fast GPU kernels is important, though not nearly as important as writing correct ones. That’s why the folks from Meta have released BackendBench.https://x.com/johannes_hage/status/1965945249274151107
Meta researchers just unveiled Set Block Decoding on Hugging Face. It’s a game-changer for language model inference, delivering 3-5x speedup in token generation with existing models. No architectural changes needed, matches previous performance. https://x.com/HuggingPapers/status/1965084731839513059
