midnightbobarun

Probably anything that's not cost-effective to train AI for or, more likely, anything that's too subjective that AI can't do a good job regardless of how much data it's trained on.

What's "too subjective?" Well, take fashion, for example. Can you teach an AI to design clothes? Of course you can. Can you teach an AI to anticipate what will become trendy for the next year among teenagers? Probably not, just because it's hard to train a model to anticipate fleeting whims. You'd need someone who's really plugged into "the culture" to anticipate future trends, and that's more art than data science.

There's probably other examples I could give, but that's the one that comes to mind most.

The UI does look really nice, I think you did a great job with it

I'm kind of impressed how often they manage to do this. Claude Code and Grok Build do resets too, but nowhere near this often. And in an ironic twist for a big company with the infrastructure, Google Antigravity never seems to do resets like this at all (at least that I've noticed). I wonder how much this stuff costs OpenAI to do?

I might travel the world. On foot, if necessary (it will probably be necessary if I have no money).

I'd say any cheap mouse off Amazon that has a pleasing shape is usually good enough, but I've also never ranked above gold in any competitive PvP shooter, so there's that :')

I'm currently using a wireless ProtoArc mouse. Good shape, can adjust DPI on the fly, hasn't broken even after a year. I think it was like 30 bucks maybe?

Just a few requests of Opus on the $20 plan uses up my entire 5 hour limit... I don't even want to know what Fable would do.

I hope it goes through, if only to improve Claude usage limits a bit...

And I don't think there's that many people using Meta Spark (I might be wrong)

Regarding being "usefully critical": something I've noticed, and it seems to happen more with cheap or "instant" models, is that it will nitpick seemingly minor things. Often it's things that aren't even all that important to the point of the discussion, but it might fixate on it, and low-key argue about it (but in a circular "I'm not saying X but I'm not agreeing with Y" kind of way).

My theory is that sycophancy is just intended to prevent the AI from spiraling into a loop of ineffectually "arguing" or fixating on unimportant details, because it's both kind of annoying when it happens AND it's obvious that the model is spiraling and burning tokens uselessly when it decides to be uselessly critical.

I don't have a PSP to try it on, but that honestly looks really impressive. :o

Something I've noticed is that local models are giving better answers these days than they did a year or two ago, even if the size (in parameters and in the amount of RAM used) hasn't increased. I'm not familiar enough with the technical side of model training to explain how they're doing this, but I think in another couple of years, models that use up 48 GB will be able to squeeze out even more incredible performance.

Though on the level of something like Sonnet 5... well, maybe not.