Is the unicode really the best way to display the file structure? The special unicode characters are encoded into 2 tokens, so I doubt it would function better overall for larger repos.
Loved the part at the end of the video about the Johnson-Lindenstrauss Lemma. Really great visualizations of how high dimensionality can encode so much information.
The roll outs are quite slow. Sometimes it takes weeks for them to release a feature to my account while I know others who get access immediately. I understand that it helps them control the quality of ChatGPT, but I wish I got access earlier as I have been subscribed since the beginning.
That's an interesting idea, would be interesting to see how well it can follow the rules for creating insulation. I think one limitation would be that it is simply following predefined rules for neighboring cells, which would mean it might not produce the mathematically best result given just the neighboring cells.
Of course, it would be interesting to see all of these iterative growths from a simple set of rules.