InTheWeights Rates People on LLM Familiarity
InTheWeights.com is scoring how recognizable people are to large language models on a 0-1000 scale, and Robin Hanson's July 6, 2026 Overcoming Bias post used his reported 956 score as a case study. For LLM practitioners, the useful signal is not personal ranking; it is a rough probe of which names and ideas are dense enough in model-training-like text to be recalled without web search. TechCrunch previously reported that the site queries multiple models and clusters descriptions into a strength score. The caveat is methodological: without open corpus and query details, these scores should guide dataset-audit questions, not serve as settled proof of reputation or provenance.
InTheWeights is useful for LDS readers as a lightweight warning about model memory, not as a definitive reputation metric. A score can hint that a person, entity, or idea is unusually visible to models, but it does not explain which documents, repetitions, or prompt behaviors created that visibility.
What happened
Robin Hanson's July 6, 2026 Overcoming Bias post says InTheWeights.com rates people on a 0-1000 scale for how well they are known by LLMs, and it uses Hanson's reported 956 score as a personal case study. TechCrunch previously reported that Thomas Dimson and Joey Flynn created the site and that it asks multiple models to identify a name, clusters similar descriptions, and assigns a strength score. The InTheWeights site itself was reachable during this audit, while Hanson's specific score claim remains sourced to his post.
Technical context
A visibility score can blend token frequency, co-occurrence patterns, duplicated commentary, model-specific memorization, and prompt sensitivity. That makes it useful as a triage signal but weak as a causal explanation. A high score may reflect broad primary-source presence, repeated secondary discussion, or model familiarity with a distinct phrase rather than durable public importance.
For practitioners
Teams auditing LLM behavior can use this kind of tool as a starting point for questions about training-data concentration, hallucination targets, and representation bias. The next step should be source sampling, prompt variation, and output-frequency checks, especially before using any score to make claims about fairness, expertise, or real-world reputation.
What to watch
Watch whether tools in this category publish methodology details such as model list, query prompts, clustering rules, score ceilings, and update cadence. Without that transparency, the scores are best treated as exploratory signals for model-audit work, not as validated benchmarks.
Key Points
- 1LLM-visibility scores can expose corpus concentration, but they do not prove why a model remembers a person.
- 2Practitioners should pair these tools with prompt audits, source sampling, and hallucination checks before drawing fairness conclusions.
- 3The Hanson case is useful as an example of model recall, not as a validated benchmark.
Scoring Rationale
This is a useful but niche LLM-audit and model-visibility story. It helps practitioners think about training-data concentration and reputation inside model outputs, but the method is opaque and the current event is anchored by a single personal case study rather than a validated benchmark release.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
