June 30, 2026
Mind-reading, but make it bulky
Meta's brain-scanning system reads sentences non-invasively, code open source
Meta says it can read sentences from brain scans, but the internet is stuck on the giant machine
TLDR: Meta says its new system can turn non-surgical brain scans into sentences and has released the code for others to build on, a big step for people who may someday need new ways to communicate. Commenters were impressed but immediately argued over the giant scanner, the lack of a live demo, and whether the breakthrough is being oversold.
Meta just dropped a very sci-fi claim: its new Brain2Qwerty v2 system can turn brain activity into written sentences without surgery, and it’s even releasing the code openly. On paper, that’s huge. The company says this could someday help people who can’t speak or type because of brain injuries. But in the comments, the crowd instantly swerved away from the glossy future and fixated on one very relatable issue: have you seen the size of this thing? One of the loudest reactions was basically, “Cool breakthrough, now make the brain scanner not the size of a small spaceship.” The machine in question is MEG, short for magnetoencephalography, a giant scanner that reads magnetic signals from the brain.
That split the discussion into two camps. The optimists praised Meta for doing the unusually generous thing and open-sourcing the code and data, with one commenter calling that the truly praiseworthy part. The skeptics, meanwhile, were not buying the hype without receipts: where’s the live demo? If this can really decode sentences in real time, people wanted video, proof, and less polished corporate mystique. Then came the armchair-lab energy: commenters debated whether this is actually groundbreaking or just a modest improvement dressed up in blockbuster language. And yes, there was classic internet comedy too — half amazement, half roasting — because nothing says “the future is here” like a miracle communication tool that currently appears to require an enormous brain-reading helmet from a supervillain’s basement.
Key Points
- •Meta introduced Brain2Qwerty v2 as a non-invasive system for real-time sentence decoding from brain recordings.
- •Meta is releasing the full training code for Brain2Qwerty v1 and v2, while partner BCBL is releasing the v1 dataset.
- •The project is positioned as a potential path toward communication assistance for people with brain lesions that prevent them from communicating.
- •Brain2Qwerty v2 was trained on about 22,000 sentences collected from nine volunteer participants, each recorded for 10 hours while typing with a MEG device.
- •Meta says the system uses end-to-end deep learning on raw brain signals, with large language models fine-tuned on neural data and AI agents used to explore pipeline optimizations.