8 Predictions for the Era of Continual Learning

An AI that learns from a company's daily work could become harder to replace as it improves. In an essay on continual learning, podcaster Dwarkesh Patel distinguishes saving notes between sessions from changing the model's learned parameters through experience. Patel predicts that replacing such a personalized model could mean losing accumulated knowledge of an organization and training its replacement again. The potential benefit and switching cost would grow together: the experience that makes an assistant more useful could also give its provider more pricing power. This is a conditional forecast about models that learn during deployment, rather than a measured account of customer lock-in.

“This is one of many reasons why I think it’s unwise to lock in some kind of regulatory safety regime right now. We simply don’t know what kind of technology we’re going to be looking at even in a year, let alone in five years or ten years, and we’d be entrenching an archaic and potentially counterproductive approach to dealing with the threats from AI.”

Dwarkesh Podcast

“Anthropic has been using Mythos internally since February. But it only shipped this model to the public in June. In a continual learning regime, a four-month internal/external gap means ceding four months of deployment learning.”

Dwarkesh Podcast

“But currently there is no switching cost for AI models. There’s nothing that’s preventing me from starting a software repository with Codex and then finishing it with Claude Code.”

Dwarkesh Podcast

“If you want to change what AI you are using, you basically have to fire an employee that has months of context on your organization and replace them with a fresh one that you have to retrain from scratch. Once you’re locked in like this, model providers can demand pretty hefty margins.”

Dwarkesh Podcast

“A large company whose employees and agents generate that much concurrent traffic can efficiently serve their continually-updated weight fork; an individual user serving themself at batch size 1 might suffer a 100x+ compute efficiency penalty. So the economics of serving personalized weights strongly favor big organizations.”

Dwarkesh Podcast