Tesla's Grok voice goes hoarse but can't hear itself — a look at AI engineering shortcuts
PTrubey · x · 2026-10-07
The author noticed the Grok voice assistant ("Rex") in their Tesla occasionally going hoarse, and the model denied it — because Grok cannot hear its own speech and has no direct control over sound production. Voice output is a separate subsystem inaccessible to the "thinking" model that composes words. Contrast with humans, who can hear and compensate when their larynx fails; Grok can only switch to another voice model. The author argues much of current AI is engineering bandaids, and until sensing and acting integrate orders of magnitude more efficiently, hacks like voices that can't hear themselves will persist.
More from Models
- Qwen3.8 Flash Next IQ1_M hits 55 tok/s on a 5060 Ti 16GB and still codes well — bobaburger · 2026-10-07
- EmbeddingGemma 2 runs offline multimodal RAG on phones in ~191MB–567MB of RAM — Saboo_Shubham_ · 2026-10-07
- Paradigm evals its math model across 7 hard benchmarks, releases full eval suite — tensorqt · 2026-10-07
- Limite 1B borrows nanogpt speedrun architecture: NorMuon, MUDD variant and XSA — tensorqt · 2026-10-07
- 1B model scores 94% on AIME 2026: Paradigma open-sources Limite 1B Violetto — tensorqt · 2026-10-07
- Google's open medical foundation model MedGemma published in Nature Medicine — EricTopol · 2026-10-07