I had a funny idea to upload '16 trillion tokens' to huggingface but decided to stop before frying the SSD with 90GB of data.

The latest talk at ZurichAI also exists as a video presentation since this morning. 🌂

I welcome feedback on the format, what works, and what doesn't. 🙏

youtu.be/yeEZpf4BlDA

How do you even fit such a big language model into 70 bytes?

the Matplotlib era of pollock

AC: what's the problem?

me: the review is clearly ChatGPT generated

reviewer: diving into the discourse tapestry, it is crucial to unweave the possibilit-

(Research heavily inspired by 'highlighting token-level probabilities tells us nothing'.) 🐸

Thank you ACLRollingReview that the options aren't 0, 4, 5, 6, 7, 8 anymore.

Arbitrary digits in an arbitrary number system but still thank you everyone for the past and present support. 🙂🙏