generalization
The process of abstracting a rule from many examples so that one can correctly classify or predict new, previously unseen examples.
As defined in 272 | Leslie Valiant on Learning and Educability in Computers and People
Other definitions
In machine learning, the ability of a trained system to perform well on new data it has never seen before; the gap between performance on training data and test data indicates generalization capability
AI: Grappling with a New Kind of Intelligence | World Science Festival
Where it comes up · 9
- Francois Chollet | Why abstraction is the key to intelligence, and what we’re still missing7 claims
- Richard Sutton – Father of RL thinks LLMs are a dead end5 claims
- Ilya Sutskever – We're moving from the age of scaling to the age of research4 claims
- John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI4 claims
- Richard Sutton – Father of RL thinks LLMs are a dead end3 claims
- Dwarkesh Patel Podcast - Noam Brown – Agent swarms, alignment, & recursive self-improvement1 claim
- 272 | Leslie Valiant on Learning and Educability in Computers and People1 claim
- World's Top Researcher on AI, LLMs, and Robot Intelligence1 claim
and 1 more
Most important claims · 10 of 27Search all claims for “generalization” →
- The PAC (Probably Approximately Correct) model defines learning success as accurately predicting future examples from past examples, while requiring that the effort to reach that prediction is moderate (polynomial time) and that increased effort yields proportionally better accuracy, formalized as the error being a constant exponent of the effort.272 | Leslie Valiant on Learning and Educability in Computers and People
- The shortcut rule is a principle of system design stating that if you optimize for a specific metric, you will take shortcuts along every possible dimension of your problem that is not captured by your metric; in AI, this means that focusing on task-specific skill yields task-specific skill but not generalizable cognitive abilities, because skill is orthogonal to generalization.Francois Chollet | Why abstraction is the key to intelligence, and what we’re still missing
- Fine-tuning the assistant entirely on English data leads to reasonable, helpful behavior in other languages like Spanish—the model latches onto the right helpful persona and automatically does the right thing across languages—demonstrating strong post-training generalization.John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI
- Early ChatGPT models wrongly claimed capabilities they lacked (e.g. sending emails, calling an Uber); a tiny dataset of around 30 examples showing the model doesn't have those capabilities, mixed into the training data, generalized well to fix this across all sorts of capabilities they didn't explicitly train for.John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI
- When we have generalization in deep learning models, it results from human researchers sculpting representations; gradient descent will solve the problem but won't make you generalize well to new data.Richard Sutton – Father of RL thinks LLMs are a dead end
- You don't need closely matched labels or domain experts for every capability: the base model has already seen documentation, code, shell scripts and man pages, so a few-shot prompt can elicit answers like FFmpeg usage, and training a preference model purely for helpfulness generalizes to STEM and programming even without any STEM or programming examples.John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI
- Within a manifold, it is always possible to interpolate between two inputs by morphing one into another via a continuous path along which all points fall on the manifold; the ability to interpolate between samples is the key to understanding generalization in deep learning because it allows making sense of points never seen by relating them to nearby points on the manifold.Francois Chollet | Why abstraction is the key to intelligence, and what we’re still missing
- Easier verifiable RL training problems generalized strongly enough that the model could contribute to a very hard problem like a Millennium Prize Problem, even though training likely never involved problems that ambitious, and Brown agrees the gap exists but says training on some hard tasks has shown more ambitious abilities.Dwarkesh Patel Podcast - Noam Brown – Agent swarms, alignment, & recursive self-improvement
- The word 'scaling' is linguistically powerful and shaped the field's thinking by directing everyone to pursue scaling recipes, but now that scale is very large, 100x more compute would not solve fundamental problems like generalization, requiring different approaches.Ilya Sutskever – We're moving from the age of scaling to the age of research
- As models get better they become more sample-efficient at recovering from errors and edge cases—rather than needing tons of data, a little data or generalization from other abilities lets them get back on track, whereas current models get stuck and lost.John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI