Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

Reply to @bonifartius@noauthority.social
🌲@scathach@stereophonic.space (2026-08-16 23:11:07)
@bonifartius A big part of what sketches me out about using LLMs for large programming tasks (bro ur just a manager of juniors now bro) is that I can't have a theory of mind for them or build well-founded trust in them, like I can with people

The whole process ends up being adversarial and exhausting
---Reply---
🌈ᚩ🌈@bonifartius@noauthority.social (2026-08-16 23:42:58)
@scathach i think a big part of why LLMs aren't good for this is that the input language is english when it should be some artificial precise meta language. would be prime time for declarative programming.

---Replies---

🌲@scathach@stereophonic.space (2026-08-17 07:03:36)
@bonifartius Spec-driven-development (especially with a human written test suite) is a lot better than vibe coding, but yeahhh Dijkstra was right all along https://www.cs.utexas.edu/~EWD/transcriptions/EWD06xx/EWD667.html