Re:Searchformer: Beyond a* Better Planning with Transformers...
by nextaccountic on Saturday April 27, 2024 @04:28AM
(#40178320)
Well when I asked "How does the worst case for an algorithm like this look like?" I fully expected the answer to be "in practice this never happens" (that is, there are bad cases but since we aren't drawing inputs at random, in practice we never hit them)
But how likely is that?
Re:Searchformer: Beyond a* Better Planning with Transformers...
by woopsn on Saturday April 27, 2024 @02:24PM
(#40182227)
Yes significantly. But take for example the stock market, which is obviously of interest, to OAI in particular, but which cannot over the medium-long term be extrapolated accurately from historical data.
It boils down to "you can't predict random numbers".
I'm sure it's useful to have formalized but it has no real impact on real world situations because most problems we care about have some kind of structure we can exploit.