Jon
musselwhite.dev
Computer science PhD student at Howard University, homelab tinkerer, digital photographer, and Oxford comma adherent. Tongue is often firmly in cheek. 📷 joncaptureslight.com 👨💻 musselwhite.dev
👤 Joined on 12 November 2024
191
Followers
211
Following
494
Total Posts
📜 Kullanıcı Geçmişi
- 12 Kas 2024, 13:45
🎉 Bluesky'e katıldı - İlk kullanıcı adı:
@musselwhite.bsky.social - 12 Kas 2024, 18:39
✏️ Kullanıcı adı değişti:
@musselwhite.bsky.social→@musselwhite.dev
🚫 Blokladığı Kullanıcılar
❤️ Beğendiği Gönderiler
like i think you should be promoting the agent to be like “if you run across a broader issue, let’s step back and talk about it” but that is something you should be promoting, not having the model infer your intentions on
Show Posttowards less and less human interventions and more and more “just do big thing”. i’m basically criticizing the whole shift towards “do more than what was asked in the prompt”
Show Postthe word “ask” here does a lot of heavy lifting. like i have definitely observed these models today just go and do the thing, after asking themselves - but not the user - if they should do so. if the model is seriously stopping and asking, it’s maybe more fine, but the labs are continuously pushing
Show Postthe models do not check in and present ramifications of a broader fix before doing so though. try letting opus 5.5 have access to prod deploys and see what happens.
Show Postthis is imo a bad thing. it’s wild to me anyone is asking for this
Show Post