See: /singer/top-5-most-useful-benchmarks-for-es
This resolves YES if an algorithm is found that passes one of the top 5 tests/benchmarks/protocols that win in the above question, and which is able to run on an Game Boy Advance in "realtime" (meaning, it can't be identified better than chance just from its speed performance in an anonymized dataset of human trials--please criticize if this doesn't sound right).
This question was inspired by this video:
1,000
3.00This resolves No only in 2090?
I'd argue that any reasonable benchmark requires some common sense knowledge about world or at least language and this knowledge requires much more memory than GBA could address normally.
This resolves No only in 2090?
Yep! Unless it's proven mathematically somehow beforehand.
I'd argue that any reasonable benchmark requires some common sense knowledge about world or at least language and this knowledge requires much more memory than GBA could address normally.
By memory do you mean RAM, or the size of the cartridge?
@singer Both cartridge and ram.
If the cartridge is something relatively standard, it's 32-64 MB, which is way not enough.
If you implement some cheaty bank switching that wouldn't be possible at the GBA times - loading a reasonably sized model in <32 Kb chunks will make it run at glacial speeds.
Running a model on a different machine and streaming with a WiFi cartridge is definitely possible but it's not running on GBA, you are using it as a display at that point.