Google released its first publicly available "multi-agent" AI system, which uses more computational resources, but produces better answers....
The data center will hold 100,100 NVIDIA GPUs and use entirely renewable energy, if all goes according to plan....
Microsoft CEO Satya Nadella wanted to make Google dance two years ago, with an AI overhaul of Bing that was designed to steal Google's all-important s...
Article URL: https://www.cerebras.ai/blog/introducing-cerebras-code Comments URL: https://news.ycombinator.com/item?id=44762959 Points: 47 # Comments:...
[D] Implementing GPU snapshotting to cut cold starts for large models by 12x
OpenAI are preparing to launch ChatGPT Go, a new subscription tier
Qwen3-Embedding-0.6B is fast, high quality, and supports up to 32k tokens. Beats OpenAI embeddings on MTEB
I Generated 1 Billion Tokens (So You Don't Have To): Introducing ReasonScape
The "Leaked" 120B OpenAI Model Is Trained In FP4
GPT-5 is already (ostensibly) available via API
Gemini 2.5 Deep Think solves previously unproven mathematical conjecture
OpenAI OS model info leaked - 120B & 20B will be available
The OpenAI Open weight model might be 120B
[P] Tri-70B-preview-SFT: Open 70B Parameter LLM for Alignment Research (No RLHF) | Trillion Labs
China report the finetune deepseek scientific model 40.44% on HLE
Open Horizon is NOT GPT-5
Deepthink
[D] Weight Tying in LLM Seems to Force the Last MLP to Become the True Unembedding
After building 10+ projects with AI, here's how to actually design great looking UIs fast
Claude Code running on VM.
Why some people think simple prompts can make LLMs do complicate things?
List of gpt-5 benchmark endpoit
Deep Think benchmarks
GLM-4.5-Air running on 64GB Mac Studio(M4)
50 shortest prompt lines that can literally 10x your results.