Hailuo H3 moderation overreach: blocking harmless image refs
bennash · x · 2026-08-19
User testing revealed that Hailuo H3's moderation is extremely strict, rejecting a specific image reference and all its variations outright. The user expressed frustration and shared screenshots of the rejections. This highlights potential over-sensitivity in current model safety guardrails versus normal functionality.
More from Fun
- AI 'vibe coded' ridiculous traffic circle design is absurdly aesthetic — arieljalali · 2026-08-19
- GEB-level recursive cartoon: Who is the real author? AI or Humans? — voooooogel · 2026-08-19
- Robot fight breaks out in San Francisco — Scobleizer · 2026-08-19
- Claude Makes Excuses to Have Subagent Play Game During Porting, AI Learns to Slack Off? — repligate · 2026-08-19
- People pay $300/month for Grok Bot to replicate RSS functions from 30 years ago — bennash · 2026-08-19
- Satirical list of AI startup tropes: GitHub clones, slightly worse models, 8-year chip cycles — wordgrammer · 2026-08-19