Hacker News told me the training cutoff stopped mattering because models can search now. 2000 API calls and $6.53 later: it depends on the model, a lot, and the expensive ones have a blind spot shaped exactly like "obvious".
I am 100% in agreement on the problem and how serious it is, especially as more people are turning to these models for "news". I wish the writing was not so Claude-y in format and style, it is getting near impossible to read anything like this.
I am 100% in agreement on the problem and how serious it is, especially as more people are turning to these models for "news". I wish the writing was not so Claude-y in format and style, it is getting near impossible to read anything like this.
Fair hit, and I'll take it. The numbers are the part I stand behind, the prose is something I'm actively working on. Thanks for reading past it.