3 ms·
Because there are no intermediate internal registers with values that can be altered by a prompt, it's prompt in -> pass through deterministic computations with
by heyitsguay 3y ago
Because there are no intermediate internal registers with values that can be altered by a prompt, it's prompt in -> pass through deterministic computations with fixed trained weights -> token out. Context-dependent planning (beyond what you get from next-token prediction, which is already a lot) requires additional mutable state.
- heisenzombie 3y agoYeah but it gets the whole prompt every time, right? So the whatever state it could compute for the first forward pass, it could also compute on the subsequent one. Hence, does it matter that it can’t store the state, if it can recompute it?
- ukuina 3y agoExactly. It's inefficient for LLMs to recompute state based on previous output, but that's what we have so far, and it works, surprisingly.