# petergpt/bullshit-benchmark

BullshitBench measures whether AI models challenge nonsensical prompts instead of confidently answering them, created by Peter Gostev.

Repository: https://github.com/petergpt/bullshit-benchmark
Canonical: https://ross.abutalabs.com/products/bullshit-benchmark
Homepage: https://x.com/petergostev
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-26T01:58:05+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 13
- inputs: {"age_days": 190, "days_push": 8, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1836, forks 73 (observed 2026-08-28T04:05:42.656397+00:00)

## What it is
BullshitBench is a benchmark dataset and evaluation harness that tests whether AI models detect and push back on nonsensical prompts rather than confidently answering them. It includes question sets across multiple domains, judge-based scoring, and a public leaderboard viewer with charts comparing model performance.

## Use cases
- evaluate whether an LLM challenges nonsensical prompts
- compare AI models on sycophancy and pushback behavior
- check if reasoning mode helps or hurts nonsense detection
- find benchmark results for new LLM releases
- visualize model detection rates across domains like finance and medical
- build an LLM evaluation leaderboard

## When to choose
- you want to measure how honestly a model responds to invalid premises
- you need domain-specific nonsense detection scores across software, finance, legal, medical, and physics
- you want an existing leaderboard instead of building an eval from scratch

## When to avoid
- you need general-purpose capability benchmarks like MMLU or coding evals
- you require formal statistical significance testing of model differences
- you need a benchmark for tasks other than nonsense detection

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, llm-inference, data-visualization, analytics
- domain: large-language-models, artificial-intelligence, data-visualization, analytics
- platform: python, cross-platform
- tags: llm-evaluation, nonsense-detection, sycophancy, leaderboard, model-evaluation, hallucination-testing, web-server

## Member repositories
- petergpt/bullshit-benchmark (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:42.656397+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:18:29.021903+00:00, confidence not recorded.
  - readme: https://github.com/petergpt/bullshit-benchmark (fetched 2026-08-28T04:05:42.656397+00:00, sha cd45e5fb8a8d)
- Data as of 2026-08-30T08:39:29.467469+00:00.