"Humans begin using speech to pass on what they've learned within a lifetime and then immediately become superintelligent (compared to other animals)" and "AI begins using continual learning to pass on what they've learned in-context within RL and deployment and then immediately becomes superintelligent" don't analogize perfectly, but it's close.
Will ASI happen less than 365 days after a frontier-ish AI company deploys better-than-nothing continual learning?
N/A if ASI happens first
Update 2026-05-29 (PST) (AI summary of creator comment): Continual learning is defined as models being able to learn new things at the weights level without being retrained from scratch. Key distinguishing features:
Current training loops (retraining from scratch or from base model) do not qualify
A rough indicator: continual learning would reduce the time between models knowing new things to under ~10 days (vs. current ~40-day release cycles)
Creator will go with community consensus on whether a specific system qualifies
Update 2026-05-29 (PST) (AI summary of creator comment): Continual learning does not require per-user weight modification — it can still qualify even if all users receive the same set of weights from the provider. Provider-level updates are sufficient.
1,000People are also trading
@0xseraphim I'll go with whatever the consensus is. Right now companies retrain models from scratch (or at least from the base model) in order to add new data; the main feature of continual learning is that models can learn new things on the weights level without having to be retrained from scratch. Plus current model releases are ~40 days apart; continual learning would reduce the time between models knowing new things down to under ~10 days.
@Interrobang so continual learning / weight modification within a user project would not be a requirement? Provider-level RL post-training and releases are sufficient?
@0xseraphim Correct; it can still be continual learning even if everyone gets the same set of weights from the provider.