惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

小众软件
小众软件
A
About on SuperTechFans
博客园 - Franky
Engineering at Meta
Engineering at Meta
Recent Announcements
Recent Announcements
云风的 BLOG
云风的 BLOG
B
Blog
Microsoft Security Blog
Microsoft Security Blog
L
LangChain Blog
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
U
Unit 42
Martin Fowler
Martin Fowler
Y
Y Combinator Blog
Stack Overflow Blog
Stack Overflow Blog
博客园 - 叶小钗
Vercel News
Vercel News
Apple Machine Learning Research
Apple Machine Learning Research
The Cloudflare Blog
Last Week in AI
Last Week in AI
腾讯CDC
Microsoft Azure Blog
Microsoft Azure Blog
爱范儿
爱范儿
V
V2EX
G
Google Developers Blog

GetRest News

Physicists Build a Working “Quantum Lie Detector” A Tiny Astronomer Lives Inside Vacheron Constantin’s La Quête du Temps Money Machines: Why Wall Street’s “Wolves” Are Disappearing Quantum stretches its wings with Google’s newest chip Google Gemini Live Can Now “See” Through Your Phone Camera—Accurate, On-Screen, and Near-Instant GENESIS, But Make It Real: How Europe Wants Cleaner Chips (And Why I’m Oddly Excited) Google Pixel Buds 2A: Focus on Noise Cancellation and Performance Apple Prepares Major Product Lineup Refresh for Fall 2025 Why I Spent $180 on a Gaming Mouse for One PC Game — And Would Do It Again
Brain Implant Decodes Inner Speech — With a Built-In “Pas...
Justin Hart · 2025-09-23 · via GetRest News

Inner Speech

 September 23, 2025

|   8:27 am

What exactly was shown (and when)?

In August 2025, a Stanford-led team reported a brain–computer interface (BCI) that turns deliberately imagined phrases into on-screen text with up to ~74% accuracy. Four volunteers who’d lost speech after stroke or neurodegenerative disease took part. The decoder only wakes when users think of a preset passphrase — a playful “Chitty-Chitty-Bang-Bang”! Privacy first, then performance!

Where in the brain?

Researchers implanted microelectrode arrays in the motor cortex — areas that normally coordinate mouth, tongue, and larynx. Attempted speech and inner speech sparked overlapping patterns there; AI models learned those patterns and mapped them to phonemes, words, then sentences. Concrete, not sci-fi.

Why the “mental password” matters

Mind-reading? No. The system ignores everyday thoughts and responds only to an intentional cue. In tests the password detector worked with ~98–99% accuracy, then the BCI decoded short, consciously imagined phrases from a vocabulary reaching as high as 125,000 words. Faster than letter boards; less fussy than eye tracking; closer to real conversation again.

Limits you should know

Free-flowing monologue? Not yet. Signals are faint, vocabularies still constrained, and the hardware is invasive and wired — suitable for clinical trials, not home use. Spontaneous thoughts remain out of scope by design. Caution first.

How this compares and why it matters

Earlier assistive setups required users to try to speak, which is fatiguing and slower. Here, “inner speech” itself is the signal — quicker, quieter, potentially less tiring. By September 2025, related BCI efforts (e.g., Neuralink’s cursor control in daily life) underlined a shift from lab demos to live function. Different paths, same goal: restore autonomy.

What’s next

Teams are pushing for wireless implants, richer vocabularies, and real-time dialog. Equally urgent: ethical guardrails — consent, audit trails, and tamper-resistant locks — so decoding stays under the user’s control. That’s the headline. And the homework.