2 ms·
I do neither maths nor science with AI but in my experience most models are perfectly willing to burn tokens on a ton of sidequests at the earliest opportunity.
by blfr 7d ago
I do neither maths nor science with AI but in my experience most models are perfectly willing to burn tokens on a ton of sidequests at the earliest opportunity.
- accurrent 7d agoYeah but do you read through and find if one of those side quests is useful?
- someguynamedq 7d agoYes, why not?
- blfr 7d agoThey're often directly applicable: minor bugs, inconsistencies, missing tests, commonly also stuff I already now (like some infra config details).
- accurrent 7d agoMinor bug and develop whole new field of cryptography are very different scales. I do agree alms are really good with a lot of common bug fix related tasks to the point you have to split commits out. The fact is you recognise it's a minor bug. Who is sitting through the proof of Navier Stokes and going through it and finding connections between itself and other fields? Im not saying LLMs are bad, but I do think understanding is important. Heck, the fact you identified the minor bug suggests you understand the output. Im not so sure the same can be said of a gajillion line lean dump.