---
title: "Adam Tauman Kalai"
type: "entity"
entity_kind: "person"
status: "hub"
canonical_name: "Adam Tauman Kalai"
aliases: []
first_seen: "2026-10-31T00:00:00.000Z"
writer_model: "claude-sonnet-5"
connects_to: ["hallucination (LLM)","calibration","Good-Turing estimator","Santosh S. Vempala","correlated errors"]
seek_code_commit: "d40ee39"
---


Researcher (OpenAI/Microsoft Research) who co-authored, with Santosh S.
Vempala, "Calibrated Language Models Must Hallucinate" (arXiv:2311.14648;
STOC 2024) — the theorem proving a statistical lower bound on hallucination
rate, tied to the Good-Turing missing-mass estimate, for any calibrated
learning algorithm — and later co-authored OpenAI's "Why Language Models
Hallucinate" (2025), which extends and applies that bound. Matters to this
vault as the connecting figure across its whole correlated-errors/hallucination
lineage: the 2023/2024 theorem both later vault notes on LLM hallucination cite
as their theoretical origin. A hub for him was explicitly deferred on
2026-10-31 ("not yet hubbed... on a capture's word alone") until the earlier
theorem itself was read directly rather than taken on a later paper's
characterization; this capture is that direct read, discharging the deferral.

## References
- [[claim-kalai-vempala-2023-hallucination-bound-is-a-rate-theorem-sorted-by-fact-regularity]]
- [[claim-kalai-vempala-2023-construction-models-arbitrary-fact-hallucination-as-uniform-random-draw]]
- [[claim-kalai-vempala-2023-theorem-is-silent-on-cross-model-hallucination-agreement]]
- [[claim-llm-hallucination-divergence-is-the-default-not-convergence]]
- Captures: 10-inbox/raw/2026-11-02-this-weeks-ai-safety-capture-found-llm-errors.md