This breakthrough not only turned the code of "think it, do it" into reality, but also marked a crucial step forward for humanity on the path toward future human-machine collaboration.

Article author, source: AIBase

Recently, Elon Musk’s brain-computer interface company, Neuralink, achieved a groundbreaking technological milestone—directly integrating the AI model Grok into the human brain. This marks a leap from brain-computer interfaces merely enabling “thought-controlled cursor movement” to a new era of “thought-controlled AI.”

72-hour non-stop marathon: Engineers moved into the patient’s home to develop eight major features

Behind all of this lies a recent intensive real-world effort by the Neuralink team, during which several core engineers moved into the subject’s home and, through 72 hours of intense, closed-door development, rapidly produced eight entirely new technical prototypes.

Subject Kenneth is a Neuralink P21 clinical trial participant who has been living with ALS for exactly seven years. Facing his condition, he has maintained an exceptionally optimistic outlook. Through the collaborative efforts of engineers, the technological breakthroughs achieved over these 72 hours have not only enabled the capture of underlying neural signals but also extended to deep integration at the system-level interaction and top-tier AI large model layers.

From底层 signals to top-level AI: Comprehensive analysis of core features

Neural Voice Activity Detection (Neural VAD): In the past, patients wishing to speak by thought alone had to manually press a switch button—a significant burden for those with high-level paralysis or ALS. To address this, the team developed the “Neural Voice Activity Detection” system, acting like a 24/7 neural sentinel that continuously monitors the stream of brain signals transmitted by the implant. The system automatically distinguishes between rest and speech preparation states, instantly activating decoding within milliseconds upon detecting an intention pulse, then automatically returning to sleep as soon as speech ends—achieving true zero-button, pure thought-based “speak as you think.”

Real-time neural signal visualization: To address the black-box challenge of “how to think in the mind to trigger the strongest neural firing,” the team converts the signal intensity and spatial distribution detected by the implant into real-time visual feedback for the patient. This creates a closed-loop neural feedback system, enabling patients to inversely train and adapt the decoder through intuitive, immediate visual challenges.

Mobile brain-computer interface input method integrated with lock screen: The team further embedded the speech brain-computer decoding system into a smartphone and successfully completed an accessible ordering test in a real café setting. Additionally, by leveraging Apple iOS low-level APIs combined with Live Activities and Dynamic Island functionality, they developed a system-level brain-computer interface input method with a persistent lock screen interface. Patients can instantly produce speech on the lock screen with a thought, without needing to unlock their phone.

Grok Agent Communicates with Thought: The most science-fiction-like aspect of this upgrade lies in the deep integration of brain-computer interfaces with large models. Patients need only silently think “Hey Neuralink” followed by their desired task to instantly awaken the Grok agent on their computer. Grok then automatically takes over the screen and executes the operation, amplifying a single thought into a series of complex computer actions. Additionally, the team has directly integrated the speech generated by brain-computer decoding into the iPhone’s call microphone audio stream. Kenneth, who has been paralyzed for seven years, successfully used his thoughts to call his wife, regaining the long-missed experience of everyday communication through synthesized speech from brain signals.

Industry analysis points out that the biggest bottleneck for brain-computer interfaces has previously been limited bandwidth; by introducing agents like Grok, the brain only needs to output core intentions, while all complex step-by-step breakdowns and interface operations are fully managed by AI. This breakthrough not only turns the idea of “thinking is doing” into reality but also marks a crucial step forward for humanity on the path toward future human-AI collaboration.