Blind-taste-test workflow: two Gemini models grade Claude Code's video edits
aryanXmahajan · x · 2026-11-22
A practical QA workflow for AI-edited video audio: render 4 loudness-matched anonymized versions, send them to two independent Gemini models that rank them and flag exact failure timestamps, then have Claude Code verify each timestamp claim against word timings and real cut points, discarding hallucinated gaps and keeping the consensus best version. Blind listeners preferred the lightest noise reduction — heavy cleanup sounded watery and hollow despite better loudness metrics. Ship what sounds human to a cold ear, not what the meter prefers.
More from coding & agent
- Agent swarm locks out cofounder for 24+ hours: Simeon Campos on multi-agent dynamics — repligate · 2026-11-22
- Dev builds sandboxed browser plugin for Claude Code that clicks through pages and evades bot detectors — jradoff · 2026-11-22
- AWS open-sources CodeKB, a tool to help AI understand brownfield enterprise codebases — FlolightC · 2026-11-22
- Researcher: Collaborating with frontier models is a deeply technical skill with no diminishing returns yet — trq212 · 2026-11-22
- Dev warns: don't disintermediate yourself — skill of working with models keeps paying off — trq212 · 2026-11-22
- Weak-form proofs mostly eliminated DoS bugs in httpd-rs — a verifiable RL-training target — jd_pressman · 2026-11-22