avatar
notayakk.bsky.social @notayakk.bsky.social

The basic design of LLVM is a lossy compression of the internet. Input and previous output tokens produces a probability distribution of the next output token, mimicing training data. Then "spice" is added to fix errors; like the LLM talks to itself first. What did I miss?

aug 10, 2025, 10:04 pm • 1 0

Replies

No replies