OPEN LINK ↗
// 1h agoRESEARCH PAPER
UserHarness reframes Theory of Mind as user-mind reconstruction
UserHarness is an inference-time framework for Theory-of-Mind tasks that models a user’s partial observations, evolving beliefs, intentions, and actions instead of inferring mental state indirectly. In the paper, the approach is evaluated across five benchmarks and reaches up to 95.94% macro accuracy, with reported gains of more than 15% relative over existing inference methods and about 20% relative over the strongest prompt-only harness.
// ANALYSIS
Hot take: this reads less like a clever prompting trick and more like a useful mental-model decomposition for agents that need to reason about what a user sees, thinks, and will do next.
- –The main idea is strong because it makes the hidden state explicit: observations, beliefs, intentions, and actions.
- –The benchmark story is compelling, especially the reported ceiling of 95.94% macro accuracy across five tasks.
- –The framing suggests this could generalize beyond ToM benchmarks into agent assistants, planning, and user simulation.
- –The claim is still paper-level evidence, so the real test is whether the abstraction holds on messier real-world interaction traces.
// TAGS
theory-of-minduser-modelingagentsinference-timellmreasoningbenchmark
DISCOVERED
1h ago
2026-05-30
PUBLISHED
1h ago
2026-05-30
RELEVANCE
9/ 10
AUTHOR
Discover AI