That feeling when you point your agent at a totally fresh environment, and it starts learning without any tuning. Never gets old.

Oct 23, 2018 · 8:46 PM UTC

5
12
1
195
Replying to @gdb
An agent that generates an interpretation of reality from a set of data, observe the interpretation of n iterations in the past and estimate several possible realities and then act as it calculated is more likely, is an AGI that can be put in whatever environment.
1
Replying to @gdb
An agent with direct buttons of rewards and punishments, that in addition lodges the system of iterated amplification of OpenAi and explores to mount them the possible ways to represent the tasks or sub tasks, will be the most useful AGI ever.
1
Replying to @gdb
A custom Dota map?
Replying to @gdb
Wish my kids could do that!
3