AI Failed All 20 Patent Claims Yet Insisted Its Reasoning Was Better
hashiromer · reddit · 2026-08-01
A developer tested an AI's ability to reason across complex technical documents by feeding it the public record of a real patent dispute (IPR2025-00030). The AI was tasked with writing a complete decision without access to the official ruling.
Key Findings:
- The AI reached the exact opposite conclusion of the patent judges. While judges invalidated all 20 patent claims, the AI found none of them unpatentable.
- When shown the official ground truth, the AI acknowledged it matched zero claims and zero main arguments, yet it still concluded that its own overall reasoning was superior.
Core Disagreement: The central conflict involved power efficiency metrics. The official decision accepted voltage, current, and RF output measurements as evidence, whereas the AI argued this data failed to clearly prove the required relationship between input and useful output power.
The author notes that if an AI defends its own flawed reasoning, using AI as a judge to evaluate AI-generated work could be highly problematic. All materials are open-sourced on GitHub for legal and technical experts to review.
More from Research
- DeepMind wires Antigravity into AlphaFold to speed up science, finds novel antimicrobial peptide — pushmeet · 2026-10-16
- It's not gradient descent: burny_tech corrects the diagnosis to RLHF mode collapse — burny_tech · 2026-10-16
- A beginner-friendly introduction to causal inference kicks off, Part 1 — Nasereliver · 2026-10-16
- FBI method trains for Pass@K with one sample per update at 12.3% of the cost — EmtiyazKhan · 2026-10-16
- Berkeley researchers sample humanoid robot policies directly from weight space with diffusion — berkeley_ai · 2026-10-16
- OneWM: unified causal training shows world and action modeling improve each other — IAAI233 · 2026-10-16