Invisibilization as a Source of Probability Bias in Language Models
3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked
This paper measures how leading language models from OpenAI and Anthropic often refuse to assess unverifiable reports when the reporter is described by race, ethnicity, or other protected attributes, rather than weighing the evidence.
Score
0.70total under rubric v2.0
- Significance0.68
- Reader interest0.78
- Audience breadth0.69
- Practical relevance0.71
- Substantiated0.83
- Technical depth0.83
- New information0.89
- Counts against: Promotional0.04
- Counts against: Overclaim0.37
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| Significance | 0.73 | 0.50 | 0.82 | 0.68 | 0.32 | +0.25 | +0.171 |
| Reader interest | 0.73 | 0.70 | 0.90 | 0.78 | 0.20 | +0.20 | +0.155 |
| Audience breadth | 0.74 | 0.60 | 0.72 | 0.69 | 0.14 | +0.10 | +0.069 |
| Practical relevance | 0.76 | 0.60 | 0.78 | 0.71 | 0.18 | +0.15 | +0.107 |
| Substantiated | 0.98 | 0.70 | 0.82 | 0.83 | 0.28 | +0.10 | +0.083 |
| Technical depth | 0.96 | 0.60 | 0.93 | 0.83 | 0.36 | +0.10 | +0.083 |
| New information | 0.90 | 0.80 | 0.98 | 0.89 | 0.18 | +0.10 | +0.089 |
| Promotional | 0.05 | 0.05 | 0.01 | 0.04 | 0.04 | −0.15 | −0.006 |
| Overclaim | 0.32 | 0.40 | 0.38 | 0.37 | 0.08 | −0.15 | −0.055 |
| Sum of contributions | +0.697 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.697 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN —
Recorded for context only, checked 24 Oct 2026, 08:53 UTC. It played no part in the ranking.