Elon Musk's most groundbreaking upgrade to the brain-computer interface integrates Grok with the human brain, making it possible to control AI directly with thoughts.

The plots in sci-fi films *The Matrix* and *Transcendence* are being turned into reality little by little by Elon Musk.

Today, Neuralink released its latest progress, which is truly quite shocking —

This time, Elon Musk directly connected SpaceXSI's Grok to the human brain!

As long as the patient silently thinks the phrase "Hey Neuralink" in their mind, and then thinks about the task to be completed, the Grok Agent on the computer will be woken up instantly.

Next, Grok will take full control of the screen and directly open YouTube.

There is another most touching scene, which is a special phone call.

The chip implanted in the brain of a patient with severe paralysis and aphasia directly converts "the words they want to say" into a pure audio stream, which is injected into the underlying microphone call system of the iPhone in real time.

Seven years later, he finally made a phone call to his wife using his mind.

Humanity has taken an extremely crazy step forward towards the real "mechanical ascension".

72-Hour Legendary Record, the Most Aggressive Upgrade of Brain-Computer Interface

All of this was made possible thanks to a "tough battle" fought by the Neuralink team a few days ago.

Several core engineers directly packed their luggage, moved into the subject's home, ate and lived with him, and carried out a full 72-hour extreme closed-door development.

For three full days and nights, they completed the work that usually takes several weeks in just a few hours, and developed 8 brand new functions in one go.

In the past, our understanding of brain-computer interfaces was limited to "moving the mouse cursor with the mind".

After these 72 hours, the brain has officially become the native input terminal of AI.

The patient, named Kenneth, has been diagnosed with amyotrophic lateral sclerosis (ALS) for seven full years, and is the Neuralink P21 subject.

He has been very optimistic along the way, just as he often says, "ALS is terrible, but my life is not."

Say a Word in Your Mind, Grok Takes Over the Task

The most sci-fi-feeling feature this time is of course Grok.

Brain-computer interface patients only need to imagine saying "Hey Neuralink" in their mind, plus the task they want to complete, to trigger a Grok agent that will execute it autonomously on the computer.

There is no need to type on the keyboard with hands, nor to make a real sound with the mouth, and the instruction is sent from the neural activity.

If you break down this link, it is easy to understand: the brain-computer interface is responsible for converting the expressed intention into usable input, and the Grok agent receives the task and then operates the computer.

The so-called "access" is essentially a complete reconstruction of the interaction dimension:

Mind-controlled cursor: The brain outputs underlying actions such as "move a little to the left, click once", and completing a task requires dozens or hundreds of steps, requiring people to watch the whole process;

Mind-driven Agent: The brain only outputs a single "intention", and all steps of decomposition and interface operation are left to AI.

The biggest bottleneck of brain-computer interfaces has always been bandwidth: the number of electrodes and the decoding speed both have physical upper limits.

The value of the Agent is that it "amplifies" a single sentence into a series of operations — without adding a single electrode, it multiplies the things that can be completed by each thought.

Just as Neuralink itself said, this is only an early exploration, but it has truly proved that:

As long as the communication between the brain and the computer becomes smooth, the gap between "thinking of something" and "achieving it" can be bridged.

Disassembly of 8 Core Technologies

The 8 technical prototypes born in these 72 hours have gone through from the underlying signal capture, all the way to system-level interaction, and finally connected to the top-level AI large model.

Neural Voice Activity Detection (VAD): Throw Away the Physical Switch Completely

Before this hackathon, when Kenneth wanted to speak with his mind, he still faced a very awkward paradox — he had to manually press a button first to tell the software "I'm going to start thinking", and press it again after finishing to stop recording.

For patients with amyotrophic lateral sclerosis (ALS) or high paraplegia, pressing this button itself is a huge burden.

To this end, the engineers directly rewrote a set of "Neural Voice Activity Detection (Neural VAD)" systems from the bottom layer.

This algorithm is like a neural sentry on duty 24 hours a day, real-time monitoring the EEG signal stream sent back by the implant.

It can extremely keenly distinguish whether the brain is in the resting state of "spacing out" at the moment, or has mobilized the motor cortex to "prepare to speak".

Once the intention pulse of speaking is captured, the APP will automatically start decoding within milliseconds; as soon as the speech is finished, the system will automatically go to sleep.

The whole process is zero-button and zero-hand operation, which truly realizes the pure mind "speak as you wish".

Real-time Neural Signal Visualization: Patients and Models "Train Together"

The brain-computer interface field has long had a troublesome black-box problem: even the subject himself does not know exactly "how to think" in his mind to stimulate the strongest and clearest neuron discharge?

It should be noted that the higher the signal-to-noise ratio of the input signal, the more accurate the back-end machine learning decoder will be.

The team made the signal intensity and spatial distribution detected by the implant into a real-time picture and fed it back to Kenneth. He can try different ways to "pronounce in his brain" and see on the spot which way mobilizes the most neural activities.

This belongs to closed-loop neural feedback in neuroscience: not only the model is adapting to the brain, but the brain is also adapting to the model in turn.

The originally invisible and intangible parameter tuning has become a "level-breaking game" with immediate feedback.

Order Coffee on Phone Without Using Hands at All

In the past few months, Neuralink has been quietly polishing an exclusive application called Bloom internally, with only one goal: to put the huge speech brain-computer decoding system into a mobile phone.

Taking advantage of this hackathon, the engineers simply pushed Kenneth out of his home, brought Bloom, and went straight to a noisy public place.

In a real coffee shop, Kenneth did not rely on others to pass on messages for the first time, but used his mind to order a cup of coffee for himself through Bloom on his mobile phone.

And this is just the beginning.

For ALS patients, Apple's built-in soft keyboard and Siri voice dictation are almost unusable. So the team directly called the iOS underlying interface and handcrafted a set of system-level brain-computer input methods: BCI Keyboard Dictation.

As long as the keyboard is called up, the words Kenneth wants to say in his mind will go through neural decoding and flow continuously into any App on the iPhone like a stream of water.

Writing memos, replying to messages, and browsing social networks, there is no need to switch back and forth between different modes throughout the whole process, and you can get things done as you intend.

Of course, decoding is inevitably error-prone.

To this end, the engineers iterated several versions of the error correction interface continuously during the hackathon, and finally embedded a set of predictive modification process directly into the keyboard: once the recognition deviates, several high-probability candidate words will pop up immediately above the keyboard, and Kenneth can complete "lightning replacement" in any App with a sweep of his mind.

There is also a detail that reduces the startup threshold of daily communication to almost zero.

The engineers cleverly borrowed the iPhone's "Live Activity" and Dynamic Island mechanism to keep Bloom resident on the lock screen interface.

Kenneth does not need to unlock the phone. As soon as the screen lights up, he can see the current brain-computer connection status; with a thought, he can directly "speak" on the lock screen state, truly realizing "instant response".

He Was Paralyzed for Seven Years, and Finally Made a Phone Call With His Mind

Among all functions, the technology of "making a phone call with the mind" is not the most difficult, but it has become the most tear-jerking part of the whole event.

Technically, the team connected Bloom with iPhone calls, so that the voice generated by the brain-computer interface is directly connected to the microphone audio stream of the call.

But the significance of this matter lies not in the technology, but in the story.

The love between Kenneth and his wife started over the phone. When they first met, their feelings were slowly developed through one phone call after another. That phone line used to be the most important line between them.

Later, the disease made it difficult for him to speak again. That line also gradually fell silent.

This time, as soon as the engineers finished making the call function, he picked up his mobile phone and called his wife.

The call was connected. He didn't open his mouth. The words he wanted to say in his mind were captured by the implant, translated by the decoder, synthesized into sound, and directly sent to the call line.

From the other end of the receiver came his wife's laughter.

The two began to tease and bicker with each other, just like when they were dating over the phone in those days.

The engineers stood by and watched. Neuralink later said that this was one of the most memorable moments of the whole week.

In the engineering field, few things are more fulfilling than this. You watch a person use the code you wrote that morning to talk to the person he loves.

Mind-driven AI Has Become a Reality

In the past, the bandwidth of the brain was too narrow, and people could only "squeeze toothpaste" to the machine one letter and one click at a time.

Now, the brain only needs to output an intention, and the rest of the decomposition, execution and operation are left to AI. When Grok is connected to the brain-computer interface, the brain is expected to become the direct input end of AI for the first time.

This will become a major leap in human-computer interaction.

In the future, with more electrodes, faster decoding, and smarter Agents, "think and you can achieve" may go from the laboratory into the daily life of more people.

In *The Matrix*, Neo connected a wire and learned kung fu in a few seconds; in *Transcendence*, human consciousness is connected to the network and can directly mobilize the entire digital world.

These imaginations that once only belonged to the silver screen are being written into reality little by little.

This article is from the WeChat official account "AI Era", author: ASI Revelation, published with authorization from 36Kr.