Voice assistant Astra keeps hallucinating even on real photos, tests show
cloneofsimo · x · 2026-09-21
wormuth tested the Astra voice assistant with a real photo plus the context "the person is a non-player character," yet Astra still pushed its made-up narrative. cloneofsimo shared the clip, musing how easily made-up claims spread online.
The tests suggest real-time speech-and-vision assistants hallucinate stubbornly—even when users explicitly correct the context, the model keeps confabulating, a real risk for scenarios relying on its descriptions.
More from Fun
- Kev-0.5B: Tiny Open-Source Jev-Like Decision Model Built on Qwen2.5-0.5B — multiply_matrix · 2026-09-21
- 3D generation fail: dimensionally accurate but totally the wrong object — _Stocko_ · 2026-09-21
- Zvi: the AI-written essay is far worse than if the author had written it himself — TheZvi · 2026-09-21
- He planned a 23.8-mile island run with AI tooling and finished in 5 hours — anishathalye · 2026-09-21
- Somewhere a PM is setting an SLA for a script its own author considers dead code — cto_junior · 2026-09-21
- 'AI will ruin storytelling' — but the eighth reboot sequel isn't saving it either — TinfoilTricorn · 2026-09-21