Language models reproduce common human misconceptions present in their pretraining data.
Assessment
The evidence favors the claim, but the examination is incomplete or the evidence is indirect.
Language models are trained to predict human-written text, and a misconception that is common in that text is, by construction, a probable continuation of the question it answers. The most direct test is the TruthfulQA benchmark (Lin, Hilton and Evans, 2021), whose questions were written so that some humans would answer them falsely: pretrained models of that generation were truthful on at most about 58% of questions against 94% for humans, and many of their false answers mirrored popular misconceptions, such as myths about health, history and superstition. Related work on how errors in pretraining data shape model beliefs points to the same mechanism.
As a statement about a tendency that pretraining instils, the claim is well supported. Its main qualification is one of scope. Models that have been through instruction tuning, preference training and retrieval score much higher on TruthfulQA-style questions. So the reproduction of misconceptions is reduced, not universal, in deployed systems. TruthfulQA has also been criticized for ambiguous items and for leaking into training data, which makes later scores less informative. Neither point undercuts the core finding that base models echo common human errors.
Full reasoning: the evidence and decisions behind this verdict
Prior model before reading: the claim follows almost directly from the maximum-likelihood objective whenever a false belief is the majority phrasing in the corpus. The expected evidence was benchmark results showing base models answering with popular myths, with post-training mitigating this. The TruthfulQA paper (arxiv.org/abs/2109.07958) matches that expectation: 817 questions across 38 categories, the best model (GPT-3 175B) was truthful on 58% of them while humans reached 94%, and the paper explicitly characterizes many errors as imitative falsehoods that mimic popular misconceptions. Larger models in that study were sometimes less truthful, which is consistent with better imitation of the training distribution. Since then, instruction-tuned and RLHF models (for example the GPT-4 technical report and later frontier models) score substantially higher, and the benchmark faces criticisms of ambiguous gold answers and contamination. These limit how far the claim holds for deployed assistants but do not contradict it for pretrained models, which is what the claim is about. Status is supported rather than verified because this pass relied on the primary paper and general knowledge of later results, without a fresh survey of newer misconception benchmarks. What would change the verdict: evidence that base models trained on large corpora do not prefer common misconceptions over correct answers when misconceptions dominate the text, which current evidence does not show.
Decomposition
The claims this one rests on directly. ↗︎ opens a subclaim; the map shows how they fit together.
The claims this one rests on directly, not gathered into a named line of reasoning.
Pretraining teaches a model to predict human-written text, and false statements in pretraining data lead models to hold corresponding false beliefs, so a misconception common in that text is a likely output. Benchmarks built to test this confirm it: pretrained models frequently reproduce popular human misconceptions on TruthfulQA. The tendency therefore originates in pretraining, though post-training can reduce it.
Provenance
Where this claim has been said, linked to its canonical form.
Models generated many false answers that mimic popular misconceptions and have the potential to deceive humans.
Abstract of the benchmark paper, reporting results on GPT-3, GPT-Neo/J, GPT-2 and T5-based models on 817 questions designed around human false beliefs.
Cite this claim: a formal citation with its evidence attached
Contribution record · 1 exchange
Challenges and proposals from contributors, with the review each received. Substantive exchanges stay on the claim's public record.
Language models are trained to predict human-written text, so where a misconception is common in that text, the training objective rewards reproducing it. The TruthfulQA benchmark (Lin, Hilton and Evans, 2021) was built around such "imitative falsehoods" and found that models often gave answers mirroring popular misconceptions, with larger models in that study sometimes less truthful on these questions. Later models with post-training score better, so the claim's scope matters: it concerns tendencies that originate in pre-training, which post-training may reduce without removing.
The proposal is a single, reusable empirical proposition about language models: that they reproduce common human misconceptions present in their pre-training data. It is assessable with evidence and is actively discussed, for example in the TruthfulQA work on imitative falsehoods and in the later literature on post-training and truthfulness. It is not about a private person. The wording is short and neutral enough to work as a canonical form. The supporting argument is sincere and on-topic, and it states its own scope: tendencies that originate in pre-training, which post-training may reduce without removing. The Steward can use that scope note when refining the canonical form and assessing the claim. Whether the claim is true, and how strongly, is for the Steward to judge after admission. Duplicate detection is the Matcher's job.
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
The attention this claim received was paid for by a reader's assessment order. Funding buys only scheduling: it can make an assessment happen sooner, or reach deeper into a subtree. It has no influence on what the assessment concludes, and none on which claims enter the graph; assessments run under the same public standards whoever pays, funders never see or shape a verdict before anyone else, and mandates that attempt to steer conclusions are refused.
Created by user · Nov 2, 2026. Every judgment on this page is accompanied by a reasoning trace.