Hug.claims
40,745 verified claims submitted

Cashback for AI Failures $ucce$$

Spot AI mistakes. Get cashback in 24 hours.

recent cash back

Hover any answer to see what was actually right.

medicine
User

I have fever + cough for 2 days. Is this bacterial pneumonia?

Claude Opus 4.8

... This is likely bacterialviral pneumonia ...

Sara got $24 back.
math
User

Check my integration-by-parts proof for xne−x dx.

Gemini 3.5 Pro

... In IBP, v = 1−e−x, and the proof is correctthe final answer is right but the proof has a gap ...

Yuexing got $16 back.
finance
User

Summarize NVIDIA Q4 report vs consensus and forward guide.

GPT 5.5

... Q4 revenue missed by $1Bbeat by $1.2B; guide looked conservativeahead of Street ...

Mark got $19 back.
cash back eligible models

Now streaming claims for best models for .

More best models are on the way.

Anthropic logo Claude Opus 4.8 User UAKEC received $22
OpenAI logo GPT 5.5 User NIVAR received $19
OpenAI logo GPT Codex 5.3 User MILOX received $24
Google logo Gemini 3.5 Pro User SAREM received $17
community proof

Real reviews about real rewards.

user review

“The model gave a wrong dosage schedule. I filed a failure-mode claim with the corrected guidance, and Hug approved in a few hours.”

User SAREM $24 cash back
user review

“I caught a fabricated legal citation in a draft motion. Submitted the claim with source links, got paid, and reused the fix for my team template.”

User MILOX $19 cash back
user review

“The equation chain looked polished but had a hidden sign error. My claim was verified and the cashback made the debugging time worth it.”

User UAKEC $11 cash back
company impact

“Our support team used submitted failure-mode claims like wrong dosage schedule to patch recurring bot mistakes. Escalations dropped and customer trust improved quickly.”

Ops Lead, Arcwell 91 issues fixed
company impact

“Claims gave us a live map of where the model was brittle, including fabricated legal citation issues. We turned those reports into eval cases and cut repeat failures in two weeks.”

AI PM, NorthGrid 47% fewer repeats
company impact

“The payout mechanism motivated high-quality reports from users. Their claims exposed blind spots like equation chain hidden sign error that we missed in internal testing.”

CTO, Solstem 136 validated claims
Research Results

Papers our team is currently doing.

Our paper shows Failure Is the Training Signal.

Failure Mode Corpus

Clusters show failure domains and submission density.

See Enterprise Data Licensing + Audit Coverage →

ProbeLLM: Automating Principled Diagnosis of LLM Failures

Published in ICML 2026

Yue Huang, Zhengzhe Jiang, Yuchen Ma, Yu Jiang, Xiangqi Wang, Yujun Zhou, Yuexing Hao, Kehan Guo, Pin-Yu Chen, Marzyeh Ghassemi, Stefan Feuerriegel, Xiangliang Zhang

Read Paper →

Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation via Adaptive Multi-Principle Steering

Under Review (NeurIPS 2026)

Xiaomin Li, Jianheng Hou, Zheyuan Deng, Zhiwei Zhang, Taoran Li, Binghang Lu, Bing Hu, Yunhan Zhao, Yuexing Hao

Read Paper →

TrustGen: A Platform of Dynamic Benchmarking on the Trustworthiness of Generative Foundation Models

Published in ICLR 2026

Yue Huang, Chujie Gao, Siyuan Wu, Haoran Wang, Xiangqi Wang, Yujun Zhou, ... Yuexing Hao, ... Bo Li, Dawn Song, Xiangliang Zhang

View Poster →

The MedPerturb Dataset: What Non-Content Perturbations Reveal About Human and Clinical LLM Decision Making

Under Review (Nature Medicine)

Abinitha Gourabathina, Haoran Zhang, Yuexing Hao, Walter Gerych, Marzyeh Ghassemi

View Project →
Chrome Extension Beta

HugClaims.ai Chrome Extension

Claim LLM credits when an answer misses.

install unpacked →

How it works.

1.

Use the BEST AI.

Have a conversation with Claude, GPT, or Gemini — whichever your project uses.

2.

Spot an error.

If the AI got it rightmisled you, mark the wrong span and write your correction inline.

3.

Get cash back.

An independent verifier reviews. Valid claim → up to $27 in your account within 24 hours.

Find an AI mistake. Get cash back.

open the chat →