We've observed AIs discovering complex tool use while competing in a simple game of hide-and-seek. They develop a series of six distinct strategies and counterstrategies, ultimately using tools in the environment to break our simulated physics:https://openai.com/blog/emergent-tool-use …
-
-
I desire nothing more than the opportunity to do so. Have been hard at work on Neural MMO since the v1.0 release -- lots of progress since:pic.twitter.com/PXvkEEK2XP
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
We’ve been actively searching for priors that are required for effectively learning tool usage from scratch, on a _small scale_, rooted in Psychology, and its turning out to be quite the challenge! Apparently, it has a lot to do with reasoning: https://arxiv.org/abs/1907.02050
-
And here’s some environments in you don’t have the compute and are interested!https://github.com/fomorians/gym_tool_use …
Kraj razgovora
Novi razgovor -
-
-
So the agents use a simulated environment to go through millions of trials each of which would kill actual animals in the real world and you think this is inspiring work? I'd be infinitely more impressed with a robot that walks into any generic kitchen and makes a sandwich.
-
But they aren't actual animals. They are virtual agents. They have no concept of violence. The "killing" or tagging aspect is just a UI convenience for us to interpret the scene. For all the trained agents know, they could be giving out flowers.
- Još 1 odgovor
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
香港人
⠀