RE: https://mastodon.social/@dangillmor/116345730533077431

OpenAI hired the OpenClaw guy, with Sam Altman declaring him a "genius with a lot of amazing ideas," in case you were wondering about the AGI company's ability to evaluate intelligence.

@bodil in tests that measure fluid intelligence (i.e. activities that depend only minimally on prior learning) it seems like LLMs struggle, until you give them a REPL loop and ability to recursively call themselves, which is pretty much what OpenClaw is doing, and why it works.

I don't think it's a bad acquisition by OAI, and academia is now formalising the OpenClaw tech bro's praxis via RLM and other nomenclature

https://www.symbolica.ai/blog/arcgentica

https://www.symbolica.ai/blog/arc-agi-3

SotA ARC-AGI-2 Results with REPL Agents

Exploiting the reasoning capabilities of code mode agents and RLMs with the Agentica framework.