What is the bad in this ruling?

I’m probably in the very very rarely camp but generally agree with you

Is high effort a model or a setting? This definitely feels unsustainable

I believe Oracle is included in that

Can’t all of those things be worse when they are explicitly controlled by an adversarial government?

Well that’s ByteDance/ the Chinese government being stubborn

They are voluntarily taking it offline. The ban impacts Apple and Google listing TikTok in the App Store and Oracle hosting their data. At least that’s my understanding

How about you?

By Thursday I was able to remember my password AND what my job is

I think there is an underestimation of how much can be learned from commonalities in images when the training data is sufficiently large. And the mistakes these models still make highlight their lack of true understanding

But it’s not actually learning any of that for the images. Just the patterns of what the pixels usually are in images with similar labels. I think that’s what lead to the extra fingers in earlier models

Alt text and other labeled images were a huge source of training data for diffusion models.

Then noise is introduced to an image and the model is trained to remove the noise.

In some ways they are simpler than text because the model is working within fixed dimensions with fixed value options

I use a spy belt and love it. Would highly recommend

Whoops misread. It was the new o1

I didn't realize they launched a new version of 4o this week. I haven't heard anything about it other than this. That's a pretty massive improvement.

Yeah I struggled with that one from a time complexity perspective. Did you get through it?

That’s fair. I’d assume this is a starting point and not the end game of their monetization.

If they’re offering premium features why would they need to solicit donations?

I feel pretty good about recognizing zero names on this list