#Meta Superintelligence Labs

'Early Experience uses agent-generated future states as supervision to train language agents without rewards or large demo sets, outperforming imitation learning across eight benchmarks and improving RL initialization.'

AI is advancing through self-improvement strategies like enhanced coding assistance, infrastructure optimization, and autonomous research, potentially accelerating the path to superintelligence.