• [Blog] First glimpse into the work of Thinking Machines Lab. Inference systems can be made deterministic by using “batch-invariant” GPU kernels (GitHub implementation).

  • [Paper] Alternative sampling strategies are shown to be conceptually a better fit if using a text space LM on numerical regression tasks.

  • [Blog] Post-training primer.

  • [Blog] Anthropic’s new blog shows substantial gains from simply using Claude to optimize MCP servers.

  • [HuggingFace] 4B model trained on Obsidian-style markdown operations pairs performance with much larger models on memory tasks - and can be used as an MCP server to a bigger model.

  • [GitHub] Crew AI - YAAF (Yet Another Agent Framework)

  • [GitHub] ROMA - YAAF (Yet Another Agent Framework)

  • [GitHub] term.everything - Run GUI apps in terminal.

Have a great week!

Updated:

Comments

You May Also Enjoy

The week ahead: 20251006

[Blog] Thinking Machines Lab: LoRA is better to be applied to all weights, including MLP and MoE. On RL, LoRA performs on par with full training even with...

The week ahead: 20250929

[Announcement] ChatGPT Pulse. ChatGPT can now proactively push personalized content.