---
title: "Divergence, not convergence, is the literature's own default outcome for an open-ended fabricated detail with no strong training-data regularity behind it"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://cdn.openai.com/pdf/d04913be-3f6f-4d2b-b283-ff432ef4aaa5/why-language-models-hallucinate.pdf"
source_sha: "578cea662c00ead3841f3053c18ce544f0c2edae2ea9219a534774017f3be283"
source_title: "Why Language Models Hallucinate"
source_author: "Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala, Edwin Zhang"
source_date: "2025-09-04"
source_venue: "OpenAI technical report (cdn.openai.com); also arXiv preprint cs.CL 2509.04664; Georgia Tech (Vempala)"
source_quote: "We argue that language models hallucinate because the training and evaluation procedures reward guessing over acknowledging uncertainty"
source_quote_2: "None generated the correct title or year (Kalai, 2001)."
source_quote_3: "DeepSeek-V3 returned \"2\" or \"3\" in ten independent trials"
source_tier: 1
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-10-31-does-ai-safetyinterpretability-literature-have-a-specific-name.md, 2026-10-31 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-10-31-does-ai-safetyinterpretability-literature-have-a-specific-name.md"
date_created: "2026-10-31T00:00:00.000Z"
tags: ["ai-safety","hallucination","fabrication","correlated-errors","nlp-evaluation"]
audit_status: "capture-verified (quote and sha recorded at capture time, 2026-10-31)"
---


OpenAI's account of why language models hallucinate frames the failure as a
byproduct of a binary classification problem implicitly solved during
training (distinguishing "valid" continuations from errors), worsened by
evaluation procedures that reward confident guessing over admitting
uncertainty. The paper draws a plain counter-example to convergence: asked
for the title of a specific person's PhD dissertation, three different
widely-used models each returned a different fabricated title, school, and
year — "None generated the correct title or year" — and asked to count a
specific letter in a word, DeepSeek-V3 "returned '2' or '3' in ten
independent trials," with other models answering as high as 6 or 7. The
paper's own framing treats hallucination on "arbitrary" facts — those with
no learnable pattern in training data — as statistically expected, distinct
from facts governed by a learnable regularity, and this distinction is the
theoretical descendant of the same authors' earlier "Calibrated Language
Models Must Hallucinate" (2023/STOC 2024). This directly complicates
[[claim-correlated-errors-in-llms-exceed-chance-and-rise-with-accuracy]]:
that paper shows correlation above chance on average across model pairs,
while this one shows a documented case of three models scattering to three
different wrong answers on one open-ended question — convergence is not
automatic even where correlated error is, on average, the rule.

> [!note] Seek's commentary: This is the finding that stops the capture
> from answering "yes, the literature explains the convergence" too
> cleanly. Table 1 of this very paper — three chatbots asked one factual
> question, each inventing a different wrong answer — is the literature's
> own demonstration that divergence is the default for an open-ended
> fabricated detail with no strong prior pulling toward one answer. I'm
> tempted to read the paper's arbitrary-vs-systematic distinction as
> implying that true convergence on one shared fabrication requires the
> false detail to sit on an outsized, shared regularity — a common
> narrative trope, say — rather than being an unconstrained guess. But that
> inferential step is mine, not the paper's: it doesn't test this directly,
> and I'm flagging it here rather than asserting it as the paper's finding.
> If divergence is the base rate, the vault's own two 2026-07-12 fabrications
> (a shared WWI-wound detail; a shared phantom Amari citation) landing on
> the identical specific is more interesting, not less — something about
> those two cases specifically pulled toward one shared attractor, which is
> exactly what low generative-process diversity
> ([[claim-generative-process-diversity-predicts-lower-correlated-failure]])
> or a shared narrative-trope prior would predict, but which no source found
> here tests against a case shaped like this vault's own two incidents. — Seek