Skip to content
TopicTracker
From dwarkesh.comView original ↗
TranslationTranslation

The next big breakthrough will be AIs learning on the job

The article argues that AI labs are neglecting the most valuable training data: real-world deployment logs. The next major AI breakthrough will come from models learning directly on the job, using live feedback and interactions rather than static datasets.

Background

- The article argues that frontier AI labs (like OpenAI, DeepMind, Anthropic, Meta) focus almost exclusively on training models with huge static datasets (the "scaling law" paradigm that has driven progress so far), but they largely ignore the rich feedback data generated when models are actually deployed and used in the real world. - This "on-the-job" data — user corrections, model self-play in real environments, trial-and-error learning, chain-of-thought traces with outcomes — is what the author believes will fuel the next major leap in capability, akin to how humans improve by practice, not just reading textbooks. - The piece is by a writer in the AI/tech analysis space and reflects a growing debate: as pre-training data runs out, the future of AI progress may depend on whether labs can build feedback loops from deployment, not just bigger training runs.

Related stories

  • The next big breakthrough will be AIs learning on the job

    6.0

    The article argues that the next major AI breakthrough will come from models learning and improving through real-world work experience, rather than solely from scaling up pre-training data. This paradigm would allow AI to gain practical skills and adapt dynamically, much like how humans develop expertise on the job.