linear cannoni find it very interesting that llama.cpp, a project presumably maintained and contributed to primarily by people who generally have a decent understanding of LLMs, has a strict policy against predominantly LLM-generated contributions, bans accounts who violate it, uses an AGENTS.md that tries to rope the agent into enforcing this and warning the user, and they have also discussed adding canaries to catch people lying about AI use and not sufficiently reviewing its output
(there is still ai-generated code in llama.cpp and it is allowed for things like minor corrections, but i nonetheless find the question of who exactly is taking these stances and who isn't, interesting)
wyatt
meek cynicswelp @linear, drug dealers don't get high on their own supply either
Mans R@mansr they are placed in html comment blocks, so that they dont render in most markdown viewers (but are still read by the LLM), and one form of this involves ordering the LLM to swap lookalike characters somewhere in the text, like replacing th letter "v" with "ν"
Anthony Sorace