Tag: ai

Post-AI Chess

AI Ruined Chess. Now, It’s Making the Game Beautiful Again

Kramnik saw flashes of beauty in how AlphaZero adapted to the new rules. No-castling chess provoked rich new patterns for keeping the king safe. A more extreme change, self-capture chess, in which a player can take their own pieces, proved even more alluring. The rule effectively gives a player more opportunities to sacrifice a piece to get ahead, a tactic considered a hallmark of elegant play for centuries.

TikTok and the Sorting Hat

TikTok doesn’t bump into the negative network effects of using a social graphs at scale because it doesn’t really have one. It is more of a pure interest graph, one derived from its short video content, and the beauty is its algorithm is so efficient that it its interest graph can be assembled without imposing much of a burden on the user at all. It is passive personalization, learning through consumption. Because the videos are so short, the volume of training data a user provides per unit of time is high. Because the videos are entertaining, this training process feels effortless, even enjoyable, for the user.

the reason tiktok has taken off is because it’s recommendation algorithm is really good, and it doesn’t need a social graph to thrive.

On GPT-3

GPT-3 is scary because it’s a tiny model compared to what’s possible, with a simple uniform architecture trained in the dumbest way possible (prediction of next text token) on a single impoverished modality (random Internet text dumps) on tiny data (fits on a laptop), and yet, the first version already manifests crazy runtime meta-learning—and the scaling curves still are not bending! The samples are also better than ever, whether it’s GPT-3 inventing new dick jokes or writing (mostly working) JavaScript tutorials about rotating arrays. Does it set SOTA on every task? No, of course not. But the question is not whether we can lawyerly find any way in which it might not work, but whether there is any way which it might work.

AI Symbolic Mathematics

For almost all the problems, the program took less than 1 second to generate correct solutions. And on the integration problems, it outperformed some solvers in the popular software packages Mathematica and Matlab in terms of speed and accuracy. The Facebook team reported that the neural net produced solutions to problems that neither of those commercial solvers could tackle.

AI Chip Design

A fast, high-quality, automatic chip placement method could greatly accelerate chip design and enable co-optimization with earlier stages of the chip design process. Although we evaluate primarily on accelerator chips, our proposed method is broadly applicable to any chip placement problem.

Jukebox

We’re introducing Jukebox, a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles. We’re releasing the model weights and code, along with a tool to explore the generated samples.