A big shift in LLM training led to capability explosion

Found on arstechnica.com

How a big shift in training - from imitation learning towards reinforcement learning - led to a capability explosion of LLMs, especially with extended context and in agentic setups.

It was very nice of arstechnica.com to reprint this original Timothy B. Lee posting in his blog on understandingai.org ;-):

Reinforcement learning, explained with a minimum of math and jargon To create reliable agents, AI companies had to go beyond predicting the next token.