Training frontier knowledge work agents: A 397B RL training guide with SkyRL (www.mercor.com via hn) 17h
event
LLM Model Competition
-
Training frontier knowledge work agents: A 397B RL training guide with SkyRL A step-by-step account of how we post-trained Qwen3.5-397B-A17B: we focus on the often overlooked infrastructure and derisking work preceding the hero run, show e…
-
Editing one row of an LLM's output layer to stop it overthinking (think-graft.vercel.app via hn) deepseek 20h
Editing one row of an LLM's output layer to stop it overthinking A linear probe folded into the </think> row lets the model's own hidden state decide when its thinking is done. Tested on Qwen3.5-4B and on DeepSeek V4.1 Flash (763B, running…
-
Dynamic Agentic Harness with Recursive Self-Improvement (RSI) A universal, model-agnostic Dynamic Agentic Harness that wraps frozen LLMs (from local 9B–80B Ollama models to frontier cloud models) in a Just-In-Time (JIT) Self-Configuring &…
-
LM Studio vs. Ollama on Linux: same model files, measured on an RTX 3090 (homebrewserver.club via hn) ollama 2d
LM Studio and Ollama both run local language models, and both serve an API that other programs can use. This page runs the same two Qwen3.5 GGUF files in both, in Docker, on one RTX 3090.
-
Show HN: JackAI Experimenting Local Models on iPhone (apps.apple.com via hn) gemmaqwencodex+1 2d
Hello HN! I've spent the last few months building solutions for friends and small businesses and creating mobile applications using Claude Code and Codex.
-
Show HN: OpenPocket – iOS app for running Qwen3.5, LFM and other LLMs offline (apps.apple.com via hn) 4d
I built OpenPocket, a completely free iOS app for running open-weight LLMs locally on iPhones and iPads using Apple's MLX framework.