🥽 Plastic Labs
Search
Dark mode
Light mode
Explorer
blog
Xeno Grant -- grants for autonomous agents
YouSim Launches Identity Simulation on X
YouSim: Explore the Multiverse of Identity
Comprehensive Analysis of Design Patterns for REST API SDKs
Solving The Campfire Problem with Honcho
User State is State of the Art
Theory-of-Mind Is All You Need
Open-Sourcing Tutor-GPT
YouSim DAO -- A DAO for Identity Simulation
A Simple Honcho Primer
Announcing Honcho's Private Beta
Introducing Honcho's Dialectic API
Achieving SOTA on OpenToM with DSPy
Memories for All
Honcho: User Context Management for LLM Apps
careers
Work at Plastic
Research Grants
Research Fellow(s)
Plastic Intern(s)
Founding Engineer
(FILLED) ML Research Engineer
(FILLED) Full-Stack Engineer
extrusions
Extrusion 06.24
Extrusion 02.24
Extrusion 01.24
notes
YouSim Disclaimers
There's an enormous space of user identity to model
Context window size doesn't solve personalization
LLM Metacognition is inference about inference
Humans like personalization
Human-AI chat paradigm hamstrings the space of possibility
Loose theory of mind imputations are superior to verbatim response predictions
LLMs excel at theory of mind because they read
Honcho name lore
Machine learning is fixated on task performance
releases
Release Notes 10.31.24
Release Notes 08.15.24
Release Notes 08.01.24
Release Notes 07.25.24
Release Notes 06.23.24
Release Notes 06.18.24
Release Notes 05.23.24
Release Notes 05.16.24
Release Notes 05.09.24
Release Notes 04.01.24
Release Notes 03.21.24
Release Notes 03.14.24
Release Notes 02.23.24
Release Notes 02.15.24
Release Notes 02.08.24
Release Notes 02.01.24
research
Evaluating Steerability in Large Language Models
Violation of Expectation via Metacognitive Prompting Reduces Theory of Mind Prediction Error in Large Language Models
Folder: research
ml
philosophy
research
2 items under this folder.
Dec 14, 2024
Evaluating Steerability in Large Language Models
Oct 12, 2023
Violation of Expectation via Metacognitive Prompting Reduces Theory of Mind Prediction Error in Large Language Models