Testers call Meta's Muse Wonder "benchmaxxed" as lava-lamp test flops
Neurogence · reddit · 2026-10-13
BridgeMind and other early testers are pushing back hard on Meta's newly released Muse Wonder: despite strong official benchmark claims, on the BridgeBench lava-lamp test the model produced just a red cylinder—no glass, no lava. "I've never seen a model this benchmaxxed in my life."
Other independent testers report similar findings, with some saying Muse "is not anywhere near close to frontier." The poster suspected benchmark-gaming from the start but says the real gap is worse than expected—a stark contrast to the official marketing.
More from Fun
- Meta AI can read your Facebook but can barely delete anything, test shows — codeninja · 2026-10-13
- AI-pilled: talking out loud to nonexistent agents when your phone breaks — VoidStateKate · 2026-10-13
- 'World's first autonomous digital organism' lives on a knockoff Roku with 4GB RAM — Kyrannio · 2026-10-13
- Alexandr Wang memes 'Muse Wonder' crushing Opus 5.5 and GPT-6.1 — alexandr_wang · 2026-10-13
- "Remember Last Year When We Wrote Code Line by Line? Like Psychopaths" — jbarbier · 2026-10-13
- 'Self-replicating AI? Without a husband? In this economy?' meme goes viral — Kyrannio · 2026-10-13