huggingface is out of date, I was hesitant to update it for reasons that seem silly now. In any case, for doing this right, the codebase is more relevant. We should probably chat about getting it up to date—I’m the main bottleneck to that, but maybe someone else wants it to work badly enough to clean up my mess before I do. I want to get eigentaste and a blog post about local-vs-asymptotic alignment out the door, then may be returning to this as my main focus, though I’m not ready to promise that.
Incidentally, I have an idea for a way to restructure articles to make them more amenable to use as search results, if it works it’ll make comparing claims across articles much easier.
Did Claude’s original search turn up the Alignment Research Dataset? https://huggingface.co/datasets/StampyAI/alignment-research-dataset
huggingface is out of date, I was hesitant to update it for reasons that seem silly now. In any case, for doing this right, the codebase is more relevant. We should probably chat about getting it up to date—I’m the main bottleneck to that, but maybe someone else wants it to work badly enough to clean up my mess before I do. I want to get eigentaste and a blog post about local-vs-asymptotic alignment out the door, then may be returning to this as my main focus, though I’m not ready to promise that.
Incidentally, I have an idea for a way to restructure articles to make them more amenable to use as search results, if it works it’ll make comparing claims across articles much easier.
No it did not! I’ll take a look