"I'm Talking to You With My Mind": Neuralink Just Gave an ALS Patient His Voice Back

Neuralink's president DJ Seo publicly stated the next target: passing the speech Turing test [2]. The goal isn't just functional communication — it's making the AI-decoded voice indistinguishable from a real human voice on the other end of a phone call. Specifically, they want it to sound like OG Ken — Kenneth's voice from before ALS changed it, recorded back in 2020. If they pull this off, the output won't just be intelligible. It'll be him. That's a high bar. But Seo said he thinks they can hit it this year.

The Numbers Behind the Milestone

Elon Musk announced this month that "Neuralink is restoring speech to those who have lost the ability to speak" [3] — a statement that reads as a product announcement but is grounded in actual clinical results. Here's what the numbers look like across the broader trial:

• 21 "Neural Knots" (Neuralink's term for trial participants) are now enrolled in studies worldwide
• Multiple patients are hitting cursor control speeds that match or beat able-bodied users with a regular mouse
• ALS patients like Jake are typing at up to 40 words per minute by imagining finger movements on a virtual keyboard
• Sebastian Gomez Pena, a 23-year-old medical student who was paralyzed in a diving accident, is using his implant up to 17 hours a day for academic work
• The UK trial at University College London Hospital now has 7 participants enrolled

Sebastian's story in particular is worth pausing on. Before Neuralink, he relied on voice commands to operate his computer — a slow, frustrating workaround for someone trying to finish medical school. After receiving the implant, he completed one of the best academic semesters of his life. He's back studying neuroscience. That's not a press release anecdote. That's the whole point of the technology [4].

Why This Matters More Than the Headlines Suggest

Here's the developer angle that most coverage misses: Neuralink isn't just a medical device. It's a high-bandwidth data interface between biological neurons and a compute layer. The VOICE study is, technically speaking, a sequence-to-sequence model problem. You're taking noisy, high-dimensional neural signal data, mapping it to phoneme sequences, and outputting synthesized speech in something approaching real time. The fact that they got "zero to 100 with the model" after adjusting input quality — Kenneth slowing down and enunciating — tells you exactly how much this system behaves like a conventional machine learning pipeline. Garbage in, garbage out. Better signal quality equals dramatically better model performance. What Neura Pod's coverage reveals [1] is that the system isn't fully automatic yet. It's collaborative — the patient works with the model, refines their neural output, gives feedback. That's the current state of BCI. But each refinement brings the floor up. And Neuralink's team is iterating fast: Neuralink's head of engineering has publicly stated they're working on a second-generation design with 10x the electrode density, scaling to 100x via 3D integration. More electrodes means better signal resolution. Better signal resolution means better models.

The Competition Is Real — And That's Good

Neuralink isn't alone in this space. A Shanghai-based company called NeuroExus has developed a comparable fully-implanted BCI. Korea just announced a national BCI R&D project targeting seven mission areas, including quadriplegia control systems and sensory restoration. The Korean government wants to be a "leading country in the global brain industry by 2030" — language that sounds familiar if you've been tracking China's semiconductor moves [1]. The competitive pressure is healthy. BCIs are hitting the inflection point where multiple well-funded teams are racing to the same capabilities. That usually means the technology develops faster than any single roadmap predicts. And for context on how good even the current state is: Neuralink's second patient, Alex Connley, is now controlling a robotic arm directly through his implant. When a company demoed a competing mouth-controlled assistive device (a tongue-operated trackpad that fits like a retainer), Alex responded publicly: "I'd love to put it up against my Neuralink." He wasn't bragging. He was benchmarking.

What Comes Next

Neuralink's stated "northstar metric" is maximizing the number of people helped, as quickly and safely as possible. On the production side, they've begun construction of small-scale manufacturing space at their Austin campus — an $8.2 million build-out that started March 2nd and is set to finish by end of May. That's not a lab. That's a factory being built around a medical device. The path from Kenneth saying "I'm talking to you with my mind" to hundreds of people doing the same thing is, by Neuralink's roadmap, closer than it sounds. The clinical trials are running in three countries. The models are improving with every session. The infrastructure is being built. For a former software developer turned tech journalist, watching this technology evolve feels like watching the early days of every other platform shift — except instead of apps on a phone, the interface is a chip in someone's head. The uncomfortable part isn't that it exists. It's how obvious it already is that it works.

Now, my son thinks I'm a superhero with a computer chip in his head.

— Jake, Neuralink trial participant