Alterego: Thought to Text

Original Article ↗ Hacker News Discussion ↗

Technical Approach & Plausibility

  • Commenters infer it’s EMG-style sensing of neuromuscular activity around jaw/face/neck (“silent speech” / subvocalization), not direct brain reading.
  • Linked MIT publications and FAQ describe an older prototype with multiple facial electrodes, user-specific training, and ~90–92% accuracy on limited vocabularies (e.g., digits, math tasks).
  • Some note the new hardware seems to use fewer electrodes around the ears, raising questions about how accuracy is maintained and whether LLMs are compensating for weak signals.
  • Several point out that video demos of this kind of tech are trivial to fake, especially when connected to an unseen computer.

Accuracy, Speed & Practical Limits

  • Many see accuracy as the real bottleneck: even 95–99% word accuracy is considered frustrating for continuous input, especially for users who can already speak or type.
  • Others counter that modern LLM-based speech pipelines can “repair” imperfect input and may tolerate more noise.
  • Observers note the demo looks slow and effortful, with noticeable facial tension; not “speed of thought,” more “silent speech.”
  • There’s debate about whether typing speed is actually a bottleneck; some say their thinking is slower than typing, others that typing severely limits idea flow, especially on phones or while multitasking.

Use Cases & UX

  • Proposed uses: private note-taking, “telepathic” chats, smart-home control, AR/VR HUD control, hands-busy scenarios (cycling, washing dishes, working in respirators), and quiet participation in meetings or cafés.
  • Several emphasize the UX win of silent over spoken commands in public, where voice assistants are socially awkward.

Accessibility & Literacy

  • Strong interest in applications for locked-in patients, motor neuron disease, paralysis, and speech or hand impairments, with caveats that the relevant muscles must still function.
  • Debate over whether such tech reduces the need for literacy: some argue it enables non-readers; others clarify it still requires language fluency and doesn’t inherently remove the value of reading/writing.

Trust, Hype & Vaporware Concerns

  • Multiple commenters compare the launch style to peak-crypto whitepaper hype and call it potential vaporware or even a “grift,” citing lack of current technical detail, public benchmarks, or tryable demos.
  • Others are cautiously optimistic, praising the core EMG-to-text idea and hoping the company hasn’t oversold beyond the underlying research.

Social, Ethical & Dystopian Concerns

  • Fears include:
    • “Thought policing” or “thought crime” scenarios if inner speech becomes observable.
    • Governments or corporations nudging or surveilling users’ internal monologue.
    • Misuse on brain-dead patients to manipulate families, or charlatan-style “spirit box” applications.
  • Some worry that offloading more cognition to AI (e.g., autocompleting fuzzy thoughts) could subtly shape or suppress people’s own thinking.

Future Computing & AR Integration

  • Several see the real potential when combined with AR glasses and on-device LLMs: eyes-up computing, hands-free interaction, and conversational coding or control at (near) thought speed.
  • Others argue that if it’s only equivalent to quiet speech-to-text, its niche might remain narrow outside accessibility and specific privacy-sensitive contexts.