Top AI Product

Meta FAIR just made reading minds a software problem instead of a surgery problem. Brain2Qwerty v2 is a brain-to-text model that decodes the sentences you silently type — straight from brain activity, no implant, no opening the skull. You wear a MEG (magnetoencephalography) helmet, think the words, and the model spits out the text.

What it actually does

The number that matters: 61% average word accuracy (39% WER). The previous best for non-invasive methods was 8%. That’s not an improvement, that’s a different category. The best participant hit 78% word accuracy, with over half their sentences off by one word or less — near the territory that used to require electrodes in your brain.

Under the hood it’s a clean end-to-end pipeline: convolutional encoder, transformer, character-level language model. Trained on ~22,000 sentences from nine volunteers, ~10 hours of recording each.

Why it’s worth watching

Meta open-sourced the training code and published in Nature. The catch: you still need a room-sized MEG machine, so there’s no plug-and-play API. But this is the first non-invasive system to get near-real-time whole-sentence decoding at accuracy that used to demand a craniotomy. The hardware shrinks next.

You Might Also Like

• Code Arena Finally Gives Developers a Fair way to Judge ai Coding Models
• Nvidia Personaplex Just Made Every Voice ai Pipeline Feel Obsolete
• Adderboard how Small can a Transformer get and Still do Math
• Meta Just Cracked Open Whatsapp for Third Party ai Chatbots but Theres a Catch
• Cognee Raises 7 5m Backed by Openai and Meta ai Founders can Knowledge Graphs fix Rags 40 Failure Rate