Meta is ready to throw a healthy challenge to Elon Musk’s Neuralink, a company that has made waves in the world of Brain-Computer Interfaces (BCI), showing positive results of linking human minds to digital interfaces without surgical implants.
Meta’s Brain2Qwerty v2 is pitched as an AI model capable of decoding natural sentences from brain activity and translating them into text in real time. What’s got the scientific community more excited is that it requires no surgery and implants, which means the medical risk is zero. That’s a massive development compared to Neuralink’s way and could find more acceptance across fields.
What does Brain2Qwerty v2 do?
Simply put, Brain2Qwerty v2 is a non-invasive brain-to-text decoder. Unlike Neuralink, which drills a hole into the skull to place electrodes on the brain’s cortex, Meta’s system utilises a Magnetoencephalography (MEG) device.
An MEG device is a specialised, high-tech helmet (resembling a large salon hairdryer) that tracks highly faint magnetic fields generated by neuronal activity in the brain while a person is thinking or typing. The AI model then maps these raw, chaotic magnetic signals directly into characters, words, and full sentences.
But how does this AI Autocorrect for the mind work?
Brain2Qwerty Version 1, which was originally released in February 2025, relied on a rigid, hand-crafted handbook of neural signals mapped to specific keystrokes. With the newer Brain2Qwerty v2, the model discards manual engineering for end-to-end deep learning. It combines three distinct AI pillars:
A convolutional encoder: This reads raw, noisy brain signals and automatically extracts patterns.
A transformer module: It looks at the context of the entire sentence rather than individual letters.
Fine-tuned LLMs: Meta trained its Large Language Models on neural data. Much like the autocorrect feature on your smartphone, these models use semantic and grammatical context to guess what a user is trying to express, filling in the gaps when the brain’s signals are messy or faint.
The numbers reveal everything
So far, models like these have yielded incredibly poor results owing to the signal distortion from bones and tissues – the accuracy rate has been at best limited to 8 per cent.
Meta’s Brain2Qwerty v2 has now shattered that ceiling:
– 61% average word accuracy, with an average word error rate (WER) of just 39%.
– 78% peak accuracy, with the system accurately decoding text with more than half of the sentences containing one word error or less.
– Crucially, Meta’s engineers discovered that decoding accuracy scales log-linearly with data volume. In simple words, the performance gap between a safe external helmet and risky brain surgery can likely be closed simply by feeding the AI more data to predict, rather than inventing better surgical hardware.
Meta says that the data for this model was compiled from 22,000 sentences recorded over 10-hour typing sessions from nine volunteers, in partnership with Spain’s Basque Center on Cognition, Brain and Language (BCBL).
Furthermore, Meta is employing its tried-and-tested open-source strategy to dominate the BCI ecosystem. Alongside the release announcement, Meta is also open-sourcing the full training code for both v1 and v2 models. This is part of Meta’s broader Digital Brain Project, which includes a $5 million fund to support open neuroscience datasets and foundational brain models, like TribeV2 and NeuralSet.
Is Meta going to change the way you compute from tomorrow?
No, don’t expect to type emails with your mind at your office desk, or prepare a report by simply thinking while you rest. MEG machines are currently massive and highly expensive pieces of equipment that require magnetically shielded rooms to isolate human brain waves from environmental noise. That’s not going to fit in your office space.
However, in the near term, this could make it easy for clinical translation. Millions of people suffering from severe speech and mobility impairments could soon regain the ability to communicate fluently without undergoing invasive brain surgery that puts them at medical risk.
What does Elon Musk’s Neuralink promise?
While Meta is relying on generative AI to change the approach to reading minds, Elon Musk’s Neuralink goes for an invasive approach. Neuralink’s method involves surgically implanting a microchip in the brain with precision robots. So far, the firm has managed to host successful trials on humans with severe paralysis, allowing them to control smartphones, computers and robotic limbs, all purely through their thoughts. Neuralink’s aim is also to restore vision to the blind and try addressing neurological disorders like Parkinson’s and Alzheimer’s.
Musk has, in fact, gone on to promise that Neuralink will work towards human-AI symbiosis, offering cognitive enhancement that allows human consciousness to merge with AI to ensure humanity isn’t outpaced by superintelligence.