gpt4-fact-checking-71pct-below-human
IN premise — entries/2026/06/21/wiki-Large_language_model-chunk-3.md
Created 2026-06-21T09:50:09+00:00
GPT-4 achieved 71% fact-checking accuracy in 2023, below the accuracy of human fact-checkers.
Summary
In 2023, GPT-4 got roughly 7 out of every 10 fact-checking judgments right, leaving a 3-in-10 error rate that human fact-checkers do not share. That gap means any pipeline relying on it for truth verification should treat its output as a useful starting point rather than a final verdict, since a meaningful share of its answers will be wrong.
Dependents
These beliefs depend on this one:
- OUT structured-reasoning-overcomes-factual-accuracy-gap — The evolution of structured reasoning prompting (CoT → self-consistency → ToT) combined with training-time reasoning specialization provides systematic methods to close the factual accuracy gap between LLMs and humans.