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

Note Detail

Reply to @icedquinn@blob.cat
Tired Bunny :comf: @tiredbun@akko.wtf (2026-10-05 13:13:23)
@icedquinn 

Except LLMs have a tendency to outright lie and make shit up, which in shell could result from stuff just not working to fucking up everything you could reach.

The answer is better syntax (though even it cannot ever be as easy as natural language, and attempts at making it more like natural language can backfire badly) and better (incl. more discoverable) documentation/teaching material. Though yeah, that's not something to blame users for.
---Reply---
Blurry Moon@sun@shitposter.world (2026-10-05 13:48:26)
@tiredbun @icedquinn have you never had this happen with a copy-pasted script from somewhere, before AI? anyway often a script is easier to verify correct than write so if you're responsible there's no problem with it right?

---Replies---

Tired Bunny :comf: @tiredbun@akko.wtf (2026-10-05 13:56:13)
@sun @icedquinn You now have an AI that spouts lies and no one to correct it before someone else takes that wrong advice, and no way to tell others when you encountered that issue. And also it responds mostly identically regardless of whether it actually had something like that in training data (which is the exact "somewhere" you can find) or just interpolated plausible response.

So it made the issue worse while appearing more versatile.