been letting CC+Codex take on my long-running hobby of collecting and parsing random books on Indian subcontinental languages, and now I randomly get emails like this
- yo the og transformer circuits thread is backEven if we have ways to break up neural networks into interpretable parts, the largest weights between those parts can be confusing. Why is that? In our new research note, we study the *weight* superposition that causes interference weights.
- noticing more that Claude prefers using “one” over normal determiners like “the” or “a”, such a marked choice
- has someone already made a pun that compares architectural changes post-Vaswani paper and the ship of Theseus?We will have an architecture that obsoletes the transformer within the next three years.
- to maximise status in SF culture, everything you post must signal that you too have solved codeforces div 1 problem b “vasya needs convex hull” at some point in your lifeIt seems ultimately correct to say that LLMs are not "creative" in the same way as humans, given all the effort that's been spent on maxing out their mathematical abilities, and the still-true observation that they seem firmly stuck in the convex hull of human achievement.
