The amount of skeptical people in this thread is astonishing, it is natural to questions pile of complexity even if it is well established. In an unlikely event it goes south, author will have much easier time debugging…
i think the solution is dataset integrated in realigning process (RHLF). the was a game trending in HN about extracting password from chat model, we will not reach 100% but i think we could probably go 99.99%
[flagged]
The amount of skeptical people in this thread is astonishing, it is natural to questions pile of complexity even if it is well established. In an unlikely event it goes south, author will have much easier time debugging…
i think the solution is dataset integrated in realigning process (RHLF). the was a game trending in HN about extracting password from chat model, we will not reach 100% but i think we could probably go 99.99%
[flagged]