latent space
The internal vector space of learned representations within a model where input data points are embedded.
As defined in 280 | François Chollet on Deep Learning and the Meaning of Intelligence
Other definitions
A compressed, abstract representation of data learned by a model, often lower-dimensional than the original data
Metas AI Boss Says He DONE With LLMS...The abstract mathematical space that a trained neural network uses internally to represent concepts and meaning—a high-dimensional compression of patterns in training data. Companies like Meta and Google are now selling advertising within this internal space.
Emad Mostaque: AGI, AI Agents & the Future of Work
Where it comes up · 11
- Jeremiah Lowin: Explaining the New AI Paradigm - [Invest Like the Best, EP.307]11 claims
- What Is Yann LeCun Cooking? JEPA Explained Simply6 claims
- Michael Levin4 claims
- LeWorldModel: Stable End-to-End JEPA World Models From Pixels2 claims
- Gwern — Anonymous writer who predicted AI trajectory on $12K/year salary2 claims
- Metas AI Boss Says He DONE With LLMS...2 claims
- Inference, Diffusion, World Models, and More | YC Paper Club2 claims
- Why MCP Won the Agent Tooling Wars (And How It Will Speed Up Agents)1 claim
and 3 more
Most important claims · 10 of 33Search all claims for “latent space” →
- In a well-built latent space each concept (such as hand count, character color, or watermelon color) sits in its own numeric dimension, so changing one number, for example turning a watermelon from green to orange, changes only that attribute in the output and makes the output step far easier.Jeremiah Lowin: Explaining the New AI Paradigm - [Invest Like the Best, EP.307]
- Generative models can produce garbage as easily as high-quality images because both reside in different regions of the same latent space, so the model must learn what bad outputs look like and users must steer away from them.Jeremiah Lowin: Explaining the New AI Paradigm - [Invest Like the Best, EP.307]
- Stable Diffusion beat DALL-E because it runs diffusion in latent space rather than pixel space; this frees the model from the spatial constraints of the image during generation, so it can merge image content first and then settle details like five fingers at the end, which gave it much greater expressive power.Jeremiah Lowin: Explaining the New AI Paradigm - [Invest Like the Best, EP.307]
- Modern diffusion latent spaces are too large for a single slider per attribute, but this does not make the models less interpretable; it only makes them more complicated, and a machine can find the combination of numbers that yields an attribute such as a bigger smile.Jeremiah Lowin: Explaining the New AI Paradigm - [Invest Like the Best, EP.307]
- Calling novel capabilities 'emergent' does nothing for a research program — it merely means you got surprised; the pessimistic physicalist response is to catalog these surprises in a 'big book of emergence', whereas the productive (optimistic) move, like mathematicians, is to assume these patterns come from a structured latent space that can be systematically mapped.Michael Levin
- Going directly from text to pixel coordinates, which expand from a few bytes of characters to thousands or tens of thousands of pixel coordinates, is too large a transformation to be done with precision, so models must pass through intermediate latent representations that preserve the input's meaning.Jeremiah Lowin: Explaining the New AI Paradigm - [Invest Like the Best, EP.307]
- Levin's research program treats all physical things — cells, embryos, robots, biobots, language models, simple machines — as interfaces or pointers; depending on what interface you build, different patterns 'ingress' through it from the Platonic space, and the program is to map the relationship between physical pointers and the patterns (including minds) that come through.Michael Levin
- Traditional pixel-prediction-based world models waste computational power on irrelevant visual details, whereas latent space prediction with abstract representations is far more efficient because it captures the essential concept (e.g., 'tree is swaying') rather than exact pixel coordinates.LeWorldModel: Stable End-to-End JEPA World Models From Pixels
- Meta is already selling advertising space in model latent space by selling against language model outputs at CPMs 10-20x higher than normal Google ads, and Google announced in earnings it will sell space inside latent space, commodifying model internals.Emad Mostaque: AGI, AI Agents & the Future of Work
- Real invention and creativity by an algorithm has historically occurred only when an open-ended search process operates over a relatively unconstrained search space; LLMs cannot invent anything because they are limited to interpolations, whereas genetic algorithms with the right search space and fitness function can invent entirely novel systems no human could anticipate—such as the famously novel antenna design.280 | François Chollet on Deep Learning and the Meaning of Intelligence