In the Alice In Wonderland (github.com/LAION-AI/AIW) reasoning and generalization benchmark, DeepSeek R1 appears to perform much more like o1 mini than o1 -preview. (Plot from laion-ai)

This made me look up chess capabilities 40 years ago, turns out they were already expert level then, and didn’t really become super human until 10 years after Kasparov. So 40 years is pretty good for your analogy.

The way you can tell if an image is AI generated or not is by looking at the hands. If the hands look weird they’re probably human drawn.

I'm still trying to figure out what's happening in their model. They state that "latent frames from the video are passed to a large transformer dynamics model, trained with a causal mask". Do they do diffusion through time or use a autoregressive model for time and diffusion per frame?

I'm being lazy, I want to understand the algorithm without doing a deep dive into the paper. I appreciate the extra work to share it here.

Me on the other hand, I once thought that putting this in a paper was a good idea 🫥

You got off easy 😅

To be fair, it wasn’t really a full convo