4 ms·
This whole thread on lookup tables seems to be confused. Isn't this purely math, the equivalence of a function to a lookup table is well studied. And NN as co
by FrustratedMonky 3y ago
This whole thread on lookup tables seems to be confused.
Isn't this purely math, the equivalence of a function to a lookup table is well studied. And NN as comprised of functions, can be boiled down to table as posted.
How do we get from this math concept of function=table, and get to arguments about consciousness and free-will and state space of the universe...
The table-NN equivalence doesn't seem to help peoples understanding of NN.
- calf 3y agoPeople are just outright abusing the terminology. OP's argument would also conclude that a sorting algorithm is not a "real" algorithm because it too can be done by an infinite lookup table. That said, the general debate is a valid one. Are LLMs just doing fancy statistical compression of data, or are they doing "reasoning" in some important sense, be that merely mechanistic logical reasoning, or "human-level intelligent reasoning"? For that matter, did the paper authors ever define "Reasoners" in their title, or leave it to the reader?
- FrustratedMonky 3y agoSure. I agree. The debate is good. I just don't see how the 'table-lookup' analogy is helping. Except maybe by helping people see the non-free-will nature of the universe. But seems like people that reject this, are also ones rejecting the 'table-function' equivalence.