Contrary to what you might infer from the initial spate of downvotes, it doesn’t strike me as necessarily nuts that a project of this kind could have value and so I applaud the effort, though unfortunately this first video does leave a fair bit to be desired methinks. At a minimum, one would hope that the audio isn’t actually somehow less comprehensible than the original, yet in comparing the beginnings as well as skipping to where Eliezer begins speaking, I find that both seem significantly more effortful to parse (and so then also have the negative side-effect of making claims in the title like “HQ” feel rather icky).
Also btw, your description mentions timestamps and chapter markers, yet it seems they’ve not actually been included (additionally on the subject of inclusions there, I suppose I’ll mention too that it would seem both useful and polite if you would link back to the original vid from your description as well ☺).
it doesn’t strike me as necessarily nuts that a project of this kind
What am I looking at here? I was surprised by the downvotes, and wondered if the problem was that the whole thing was AI synthesized. However, it seems to be genuine, the original video being this. Indeed the sound is better (but still bad), and I find both videos pretty unendurable to watch with all the ums and ers and retakes. If I had been there I believe I would have found the live performance little better.
This is from the automatic transcription (from 2:12), and even leaving aside the transcription errors it’s...it’s...
so to quickly rephrase the question is about the natural abstraction hypothesis so unfortunately neither you nor Ellas or John Wendworth who is the on theory let I’m going to try to do my best to chat with John here and quickly explain what both and quickly explain what the natural extraction hypothesis is so the natural abstraction hypothesis is something like there are structures in the universe that are useful to think about so there are some concepts that are contention you know whether or not humans in anime is a property of our culture you know maybe we come up with it maybe we don’t but the concept of a tree is quite universal trees are things in the world and even if we take say neural network systems we train them on large amounts of images even if they’re
And I wish Connor Leahy would stop tapping his foot up and down, especially when someone else is speaking.
Contrary to what you might infer from the initial spate of downvotes, it doesn’t strike me as necessarily nuts that a project of this kind could have value and so I applaud the effort, though unfortunately this first video does leave a fair bit to be desired methinks. At a minimum, one would hope that the audio isn’t actually somehow less comprehensible than the original, yet in comparing the beginnings as well as skipping to where Eliezer begins speaking, I find that both seem significantly more effortful to parse (and so then also have the negative side-effect of making claims in the title like “HQ” feel rather icky).
Also btw, your description mentions timestamps and chapter markers, yet it seems they’ve not actually been included (additionally on the subject of inclusions there, I suppose I’ll mention too that it would seem both useful and polite if you would link back to the original vid from your description as well ☺).
What am I looking at here? I was surprised by the downvotes, and wondered if the problem was that the whole thing was AI synthesized. However, it seems to be genuine, the original video being this. Indeed the sound is better (but still bad), and I find both videos pretty unendurable to watch with all the ums and ers and retakes. If I had been there I believe I would have found the live performance little better.
This is from the automatic transcription (from 2:12), and even leaving aside the transcription errors it’s...it’s...
And I wish Connor Leahy would stop tapping his foot up and down, especially when someone else is speaking.