Yeah I'm with you here, and I guess we're close to this point already in some areas.
bsky.app/profile/emol...
bsky.app/profile/emol...
If progress continues, the ability to figure out the AI frontier will slip from most of us.
For example, I am not a good enough musician or a skilled enough critical listener to know if Suno v4 is actually as good as it sounds to me. I need to defer to experts (is it?) suno.com/song/6a93b06...
For example, I am not a good enough musician or a skilled enough critical listener to know if Suno v4 is actually as good as it sounds to me. I need to defer to experts (is it?) suno.com/song/6a93b06...
Star light by @ryh | Suno
vintage soul, R&B, analog, rhodes, spoken word song. Listen and make your own with Suno.
suno.com
November 26, 2024 at 1:14 PM
Yeah I'm with you here, and I guess we're close to this point already in some areas.
bsky.app/profile/emol...
bsky.app/profile/emol...
I'm not sure I agree. I think on the scale of <worst possible> to <best possible> output for the majority of areas, most humans would be in agreement up to the top-end of the scale. Like, it's easy to spot horrendous poetry / art / code / advice / whatever.
November 25, 2024 at 11:29 PM
I'm not sure I agree. I think on the scale of <worst possible> to <best possible> output for the majority of areas, most humans would be in agreement up to the top-end of the scale. Like, it's easy to spot horrendous poetry / art / code / advice / whatever.
Ah fair, though I'm struggling to think of meaningful examples where this wouldn't be the case beyond "areas where humans widely disagree with each other" and "areas where labeling is expensive".
November 25, 2024 at 11:06 PM
Ah fair, though I'm struggling to think of meaningful examples where this wouldn't be the case beyond "areas where humans widely disagree with each other" and "areas where labeling is expensive".
Why isn't it enough just to be able to tell one better output from another? Are you saying this approach doesn't work as well as we think? Or are you talking about something else?
November 25, 2024 at 10:26 PM
Why isn't it enough just to be able to tell one better output from another? Are you saying this approach doesn't work as well as we think? Or are you talking about something else?
I'm confused why it would be more useful to call the local cluster the "feature" than a linear direction in this case, can you explain?
November 23, 2024 at 9:48 AM
I'm confused why it would be more useful to call the local cluster the "feature" than a linear direction in this case, can you explain?