It's a shock! Apple's brain successfully connects to ChatGPT

  It's the first time in the history of the world.

"I smell the taste of vinegar and yeast. I taste the taste of sugar. It's warm and bright. What should I do?"

The phrase comes from a cerebrospinal fluid emitting electricity.

The pulses from 138,000 neurons were sent to ChatGPT by a translator who decoded them into words, spelled them into sentences, without changing the words.

Just yesterday, the developer @chetaslua announced that he had connected an entire apple brain simulation to ChatGPT.

15 million connections, one at a time, on a MacBook.

There are no neural chips, no GPUs. The tools are just NumPy and a connection group.

But even in the simulation, the scale of the shock has not diminished at all.

Biological neural signals, translated into natural language for the first time, directly and directlyAITalk to me.

Chetaslua did something even more important before allowing the fruit to "speak openly": to prove that the results of this simulated brain were reliable.

So he designed a rationality test according to his paper.

The glucose pump is responsible for the activation of the CN9 neurons in the neck at 78 Hz. This indicates that the fruit tastes sweet and the mouth is stretched forward at a frequency of 78 times per second, ready to eat.

In addition to the bitter taste, the frequency instantly dropped to 3 Hz.

The directions and conclusions of the paper published in the journal Nature in 2024 are correct.

Like it or hate it, there are two frequencies in the brain. High frequency is to open your mouth and low frequency is to shrink back.

Some netizens asked: Did 78 Hz actually run out of the connection group data, or did your model write it itself?

He replied with two words:its real(Real)。

This is a critical step. The most common problem with the simulated brain is self-talk.

But first, you have to verify the path of inference in a paper, and then you're right, and then someone listens.

This is the most critical step in the whole experiment, and the most evil step.

He trained a "linear reading" on his own activity in the brain. Speaking to people is a translator that adapts different modes of transmission to different words.

Each time the cerebrospinal fluid produces a specific pattern of neural activity, the translator speaks a corresponding word.

During his training, he did something remarkably thoughtful:It's important to keep the sensory neurons out.

Because the moment the smell hits the olfactory neuron, the signal is clearest, and it doesn't really matter. It's like you're looking at the answer while you're on the exam, which is true, but it doesn't make sense.

He wanted to know how much of this signal was left in the brain after a few stops and could not be recognized in the downstream.

In other words, information is still missing from the middle link between "smelling" and "ready to move".

The result: 16 words are perfect and 100% accurate.

He also narrowed the available signals to the only remaining walking neurons and motor neurons. The layer that sends instructions from the brain to the body is already very close to "moving hands", with an accuracy rate of 85%.

This means that the sense of smell, taste, touch, and sensation carry information that can be read from the perceptual layer to the "readiness to act" layer.

Then the words were spelled out by the fruit itself and sent to ChatGPT unchanged.

The opening sentence is the first sentence of this series.

Look at this sequence.First there's electricity, then there's words, then there's sentences.

It's not the first time I've been able to use a laptop to run an entire brain.

In October 2024, Berkeley researchers used a FlyWire connection group to simulate 139,000 neurons, 50 million connections, and a laptop to predict the reaction of real fruit flies to stimuli.

This new place is:For the first time, electricity was translated into natural language, and the word was passed unchanged to a large language model.

Prior to this, the simulated neural activity could only be compared to the reaction of true fruit flies.

And this can be done provided that the connected group data is fully public.

The data is based on HHMI Janelia, the University of Cambridge, and Google Research, a decade-long study of male CNS v1.0 connecting the central nervous system.

It took 44 years for the photosynthesis to be published on September 3 in the journal Cell Dynamics, which contains 166,700 neurons, 125 million synaptic connections, and is fully open source under the CC-BY protocol.

Registration takes a token, pip installs several lines of code, and pulls the upstream and downstream connections of any neuron to the local location.

Just two days after the publication of the MaleCNS paper, Evan Smith, a graduate student at Georgia Tech, borrowed the GPT-6 Astra and plugged it into Minecraft, allowing a virtual fruit flying through the glass chamber, with neural activity directly determining each turn.

The next thing you know, there's someone playing Doom, running Mario 64, playing Beat Saber.

After a decade of graphics, two days in the game, and a few days of opening and GPT conversations.

Chetaslua also conducted a round of reverse testing.

"There's a wall in front" sends ChatGPT, ChatGPT goes "to the left" and then sends the command back to the hearing neurons of the fruit fly.

It obeyed thirty-seven times.

But I've never found food.

The instruction to carry out drops of water and feed is given in white papers.There is also a temporary gap between driving behavior and generating consciousness.

Obedience requires only signals to come in and move. But finding food is another set of things to remember where you have tried, where you haven't tried, and to use the information from a few steps before, to form some kind of "strategy" in a strange environment.

This is the real thing, the fruit is not shown.

No one knows if it smells real, if it has subjective experience. The developers themselves say it's just a simulation.

In the past, if you wanted to do something with your whole brain, you needed a state-of-the-art lab, a six-figure budget, a cross-disciplinary team, and a long wait.

Now all you need is a notebook, a downloadable thread map, and a willingly folded weekend.

With 138,000 neurons in the brain, a developer runs it at home and connects to ChatGPT.

The next experiment that lets biological neural signals and AI dialogue may not have to wait for the paper to be finalized, it may not have to wait for financial approval. It may happen at someone's desk, at night, after a flash of light.

The phrase "What should I do?" is a template phrase written by a translator.

But the question itself, now it is really our turn to give a serious answer.