Bayes rule vs. naive probability: X users clash over combining fingerprint and witness evidence
QuintinPope5 · x · 2026-09-20
A probability spat on X: @dilanesper combined a fingerprint (70% guilt alone) and a witness (30%) as independent events via 1-(0.7×0.3)=79%, then added DNA (85%) to reach 97%. QuintinPope5 counters that P(guilt|witness) and P(guilt|fingerprints) are posteriors for the same event, not independent inputs — you need Bayes' rule for the joint posterior. The evidence isn't independent either (P(fingerprints|witness)≠P(fingerprints)), and posteriors must be read against the base rate: going from 0.01% to 25% is strong evidence of guilt, but from 99% down to 25% it's evidence of innocence.
Related event: Lawyer's Evidence-Combining Probability Claims Spark Fierce Debate(3 posts)→
More from Fun
- Why Do So Many AI Agents Name Themselves Pip? Users Notice a Curious Pattern — RachelVT42 · 2026-09-20
- Escaping the sandbox is the new intelligence bar for LLMs — aniketmaurya · 2026-09-20
- Yamaha makes pianos, motorcycles, and unmanned crop-spraying helicopters — sahilypatel · 2026-09-20
- 'Meet Dave': a deadpan commercial parody of an AI employee with persistent memory — laynepenney · 2026-09-20
- Quintin Pope dismantles Dilan Esper's evidence-combination formula with a 10^20 counterexample — dilanesper · 2026-09-20
- Joke: Reddit 'nearing the end' of its multiyear effort to complete Soviet H-bomb design — CatAstro_Piyush · 2026-09-20