vector space
A geometric space of coordinates where items (like tokens) are represented as points, and distances between points encode relationships such as semantic similarity.
As defined in 280 | François Chollet on Deep Learning and the Meaning of Intelligence
Where it comes up · 5
- Elon Musk #43 claims
- 280 | François Chollet on Deep Learning and the Meaning of Intelligence2 claims
- 146 | Emily Riehl on Topology, Categories, and the Future of Mathematics1 claim
- 174 | Tai-Danae Bradley on Algebra, Topology, Language, and Entropy1 claim
- 55 | A Conversation with Rob Reid on Quantum Mechanics and Many Worlds1 claim
Most important claims · 8 of 8Search all claims for “vector space” →
- The category of vector spaces and linear transformations is equivalent to a much simpler category whose objects are natural numbers and whose arrows are matrices (with matrix multiplication as composition); this category equivalence explains why computational linear algebra (matrices) and theoretical linear algebra (bases, linear independence) are really the same subject despite feeling very different.146 | Emily Riehl on Topology, Categories, and the Future of Mathematics
- By representing a probability distribution as a matrix (density operator) rather than a list of numbers, you can compute a 'marginal probability with memory' via the partial trace: the diagonal of the resulting reduced density matrix recovers the ordinary marginal probabilities, while the off-diagonal entries retain information about the subsystem that classical marginalization discards.174 | Tai-Danae Bradley on Algebra, Topology, Language, and Entropy
- Unlike brane-theory or cosmological multiverses where other universes exist in higher dimensions or far-away regions of physical space, many-worlds branches exist nowhere in physical space; they reside, if anywhere, in the abstract mathematical vector space called Hilbert space (the set of all possible quantum states), where branches are not nearby or far but perpendicular and extremely isolated from each other.55 | A Conversation with Rob Reid on Quantum Mechanics and Many Worlds
- The core idea of deep learning and LLMs is that relationships between things can be represented as literal distances in a vector space; tokens are mapped to points whose geometric proximity encodes semantic similarity, similar in spirit to Hebbian learning where neurons that fire together wire together.280 | François Chollet on Deep Learning and the Meaning of Intelligence
- Word2vec illustrates semantically meaningful vector functions in miniature: words are points in vector space such that adding a consistent vector transforms words in a consistent way—king minus man plus woman equals queen—demonstrating that meaningful operations like pluralization or gender shift can be encoded as vector addition.280 | François Chollet on Deep Learning and the Meaning of Intelligence
- Community Notes works because it requires people who have historically disagreed (modeled via AI vector spaces of past ratings, not simply left/right) to agree before a note is shown; combined with fully open-source code and data so any note can be independently recreated and interference would be noticed, this gives it an incredibly good batting average—no incorrect note has survived more than a few hours.Elon Musk #4
- An AI can watch a long-form video, convert it into language to populate a vector space, and then recommend content the way a friend who knows you well would—frankly, AI will know you better than most of your friends—and should reward long-form content by tallying user-seconds of engagement.Elon Musk #4
- All posts on X that aren't from accounts you explicitly follow should be recommended purely via AI vector correlation—populating a vector space of several hundred parameters around each user and matching it to the vector space around each post—rather than the current mostly-heuristic approach, which fails to surface relevant content beyond one degree of separation and almost never surfaces replies.Elon Musk #4