Notes on the representation of human taste: structure, context, time, evidence and uncertainty.https://www.tastegraph.aienhttps://www.tastegraph.ai/research/05-personal-ai-needs-a-theory-of-the-personhttps://www.tastegraph.ai/research/05-personal-ai-needs-a-theory-of-the-personPersonal AI is usually framed as a memory problem. The harder problem is deciding what accumulated evidence should mean about the person it describes.Tue, 01 Sep 2026 00:00:00 GMTPersonal AIcontact@tastegraph.ai (Stéphane Benayoun)https://www.tastegraph.ai/research/04-similarity-is-not-understandinghttps://www.tastegraph.ai/research/04-similarity-is-not-understandingModern systems are remarkably good at learning which things are related. Relatedness alone does not establish which property of an object mattered, or what should generalize from it.Tue, 11 Aug 2026 00:00:00 GMTRepresentationcontact@tastegraph.ai (Stéphane Benayoun)https://www.tastegraph.ai/research/03-context-is-part-of-the-preferencehttps://www.tastegraph.ai/research/03-context-is-part-of-the-preferencePreference is inferred from evidence produced in situations. Removing context does not remove noise. It changes what the observation says about the person.Tue, 14 Jul 2026 00:00:00 GMTContextcontact@tastegraph.ai (Stéphane Benayoun)https://www.tastegraph.ai/research/02-preference-has-a-historyhttps://www.tastegraph.ai/research/02-preference-has-a-historyAn observation is not only a value. It carries a date, a context and a relation to everything observed before it. Every preference has provenance.Tue, 09 Jun 2026 00:00:00 GMTTemporal Modelscontact@tastegraph.ai (Stéphane Benayoun)https://www.tastegraph.ai/research/01-taste-is-not-a-vectorhttps://www.tastegraph.ai/research/01-taste-is-not-a-vectorVectors are exceptional tools for computation. The difficulty begins when a representation built for one task is treated as a sufficient model of the person behind it.Tue, 12 May 2026 00:00:00 GMTRepresentationcontact@tastegraph.ai (Stéphane Benayoun)