Skip to content

Comment on Benchmarks and comparison of LLM AI models and API hosting providersparent

Comments

At least for me making tests has been very frustrating, full of many "test conditions here" and "continue with the rest of the tests".

It _hates_ making assumptions about things it doesn't know for sure, I suspect because of "anti-hallucination" nonsense. Instead it has to be shoved to even try making any assumptions, even reasonable ones.

I know it's capable of making reasonable assumptions for class structures/behaviour, etc where I can just tweak it as needed to work. It just refuses too. I've even seen comments like "We'll put the rest of the code in later"

Yep, with code generation I have definitely encountered this issue as well. It will write out a function description as a comment and move on instead of actually writing out the function a lot for me. It also does this when it has properly written code but you ask it to rewrite to tweak something in my experience.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.