This start-up wants to make an AI beanie that can read your thoughts

(Credit: Sabi)

• 🧠 Sari, a Silicon Valley startup, wants to make a beanie hat with a chip that can read your internal speech, turning it into speech

• ⚡️ The hat works via EEG, which uses metal disks placed on the scalp to record the brain’s electrical activity

• 💨 It says it should start reading at 30 words a minute, which should get faster as more people use the device

• 🧢 Sari is aiming to ship the beanie by the end of 2026, and is also working on a baseball cap

The idea of simply being able to think to type, as opposed to even speaking, is a bit of a pipe dream, although one startup thinks it can put it in a fashionable beanie hat.

Silicon Valley startup Sabi is developing a brain wearable that decodes a person’s internal speech into words on a screen, with the CEO Rahul Chhabra stating that the brain-reading beanie will be available by the end of the year. The company is also designing a baseball cap version.

The idea of the brain wearable, or brain-computer-interface (BCI) is that it provides a direct method of communication between the brain and an external device. This is more of a temporary solution than the surgically implanted options that Elon Musk’s Neuralink has tested for several years.

Sabi’s brain-reading hat works via EEG, or electroencephalography, which uses metal disks placed on the scalp to record the brain’s electrical activity. Decoding imagined speech via EEG isn’t new, although it’s currently limited to small sets of words or commands, rather than more natural and continuous speech.

The problem with a wearable device rather than an implanted BCI is that the sensors have to ‘listen’ to the brain through skin and bone, which dampens neural signals. Sabi thinks that the way to solve this is to scale up the number of sensors for a more accurate reading. Most EEG devices have a dozen to a few hundred sensors – Sabi’s hat will have anywhere from 70,000 to 10,000.

Sabi is aiming for an initial typing speed of 30 or so words a minute, which is slow for most people, although it’s aiming to improve this as users spend more time with the cap.

The problem with decoding imagined speech is how much natural thought patterns can differ between people, meaning that even if two people think about saying the same phrase, their brains would fire differently.

To help decode this into actionable commands, Sabi is building a large-scale AI model called a brain foundation model that’s trained on extensive neural data to learn fundamental patterns of activity that correlate with inner speech. Chhabra says the company has so far amassed 100,000 hours’ worth of brain data from 100 volunteers.

Naturally, this kind of wearable device has raised questions about privacy and security, not least because it’s your inner thoughts. Chhabra says that when your data leaves the device and is uploaded to the cloud, it is end-to-end encrypted.

Sabi’s AI models are able to train on encrypted data, rather than raw neural data. The company is also consulting with neurosecurity experts from Stanford University and elsewhere to audit its stack to ensure the utmost privacy.

Up next: Google Gemini is now available on macOS thanks to a new native app

Reece Bithrey is a journalist with bylines for Trusted Reviews, Digital Foundry, PC Gamer, TechRadar and more. He also has his own blog, UNTITLED, and graduated from the University of Leeds with a degree in International History and Politics in 2023.