anndvision

4 ms·

1. 1.▲
   by anndvision 2mo ago

   Post author here. Thanks for reading! The agent in this run did exactly that, natively. It maintained MEMORY.md as a table of contents: one line per note with a name and a short description, and that index was spliced into the system prompt
2. 2.▲
   by anndvision 2mo ago

   Thanks for reading! I am the author of this post. You are right, context helping is known. The investigation started with what the agent reached for when asked to get better. It could have reached for code execution, subagents, web search,
3. 3.▲
   Claude Code self-improved through conversation, not the memories it saved(shojin.dev)

   4 points by anndvision 2mo ago|4 comments
4. 4.▲
   Claude Code self-improved on business workflows using conversation, not memories(andrewjesson.com)

   1 points by anndvision 2mo ago|0 comments
5. 5.▲
   When Does Data Help Automated Agent Engineering?(andrewjesson.com)

   1 points by anndvision 3mo ago|0 comments
6. 6.▲
   The engineering practices Claude Code and Codex use to improve AI agents(andrewjesson.com)

   2 points by anndvision 4mo ago|0 comments
7. 7.▲
   'Distealed' LLMs: smarter, 5-30x cheaper inference(tensorzero.com)

   3 points by anndvision 1y ago|0 comments
8. 8.▲
   by anndvision 1y ago

   thanks
9. 9.▲
   by anndvision 1y ago

   We recently ran similar experiments and saw that fine-tuning small models on automatically curated high-quality outputs from a large model can beat large-model performance while reducing inference costs by up to 30x and inference time by up