Brain2Qwerty v2 - Decode sentences directly from non-invasive brain signals
Brain2Qwerty v2 is a non-invasive brain-computer interface from Meta that decodes raw MEG brain signals into text. Using end-to-end deep learning and LLMs, it reaches up to 78% word accuracy without surgery.
Replies
Flowtica Scribe
Hi everyone!
Sure, the MEG scanner is still this massive non-portable machine. But the jump in performance is pretty big. Brain2Qwerty v2 gets to 61% word accuracy on average, and 78% for the best participant — where more than half the sentences only had one word wrong or less.
They basically went end-to-end from raw brain signals instead of using the usual hand-crafted pipelines, and it shows.
Meta also released the full training code for v1 & v2, and the v1 dataset. The scaling laws look promising!
Mind dictation soon?
This feels like a major leap forward for brain computer interfaces. If accuracy keepsimproving, text input from thought could change accessibility tools completely.
Airtop
Do you mean to tell me that now I can prompt Claude to write my emails while remaining completely still?
Decoding full sentences from non-invasive signals is wild, and I really like that you went the non-invasive route instead of chasing implants. Feels way more realistic for actual people to ever use. Curious who you picture reaching for this first, like folks who can't type easily, or a broader everyday crowd? What's the first real use case you're most excited to see people try?