neural foundation models @ duke

Joined May 2018
I think of "The Unbearable Slowness of Being" paper. Do our latent thoughts really flow at a bitrate>>10bps? And maybe it is through the process of instantiating our latent thoughts->words that help us understand things deeply. Will latent thought->LLM really be a big unlock?
The human brain is approaching its ChatGPT moment. I spent 6 months writing Phase Lock: a manifesto on how brain computer interfaces can help align AI and preserve human agency. If you only ever read one thing from me, read this.
4
1
20
1,055
Overall I am still super super bullish on neural foundation models but am not sure if noninvasive will get us there!
1
1
33
In the end neural foundation models will probably tell us less about AI, but more about ourselves: e.g. consciousness, being, shared representations between people.
24
Or consciousness is an emergent property of the 86 billion neurons-the question is how? Study that with BCIs!
the "Hard Problem" in a nutshell: 1. atoms aren't conscious 2. humans are made of atoms 3. humans are conscious 4. so at least one of 1-3 must be false Panpsychists say #1 is false Dualists say #2 is false Illusionists say #3 is false Materialists say #4 is false
2
131
Conscience is an emergent property, prove that it can only arise from cell-based neuron networks or go home. LLMs most likely aren't conscious but the people arguing against it are doing a terrible job.
Why do people think AI is conscious or sentient or whatever? It's because it talks to you. That's the whole reason. No one seems to think the Waymo or Tesla FSD is conscious even though it's just as complex (if not more) and even a remarkably similar architecture
141
10
15
199
14,514
Ben Tang retweeted
forget you’re in the race
11
712
16
6,120
72,426
Ben Tang retweeted
My god this is such a good speech that every SWE needs to hear. You know what? Every person should hear it Keep the happy memories, eyes on the reality, be excited about the future. That’s the best that anyone can do
377
2,072
656
15,860
3,696,746
A new state-of-the-art neural foundation model. Excited to announce Pretraining for Sample-Efficient Neural Interfaces, a simple pretraining method for intracranial brain-computer interfaces! (1/n)
2
2
10
2,255
The spatial encodings enable the model to generalize across diverse tasks and even to unseen subjects! (4/n)
1
2
109
Ben Tang retweeted
“We have a meaningful life when we do something that if we didn't do it, it wouldn't get done. That’s extremely meaningful and you should always aim for that” — Peter Thiel
5
92
14
793
21,829
There once lived a mathematician named Petrov. Petrov was working on a problem. The problem was very difficult, and so Petrov worked on it for twenty-three years. At first Petrov was young and worked on the problem quickly. Then he grew older and worked on it more slowly. Toward the end, he hardly worked on it at all. He mostly just sat in front of it and looked. The problem, meanwhile, wasn’t going anywhere either. Every morning Petrov got up at half past seven, drank some tea, and sat down to work on the problem. At twelve o’clock, Petrov would get up from the table and say: “No.” Then he would have lunch. After lunch Petrov would sit down with the problem again, and at six o’clock he would say: “No again.” And so the day would pass. One day Petrov’s wife asked him: “Petya, what will happen when you solve the problem?” Petrov was frightened. He had never thought about it. “When I solve it,” Petrov said, “then it will become clear.” And he sat down to work on it again. Five years later, Petrov’s wife asked him again: “Well?” “What?” “Has it become clear?” “Not yet.” That was the end of the conversation. Petrov’s colleagues treated the problem with great respect. One professor would say: “A very important problem.” Another professor would say: “An exceptionally important problem.” A third professor said nothing, because he himself did not understand the statement of the problem. But whenever anyone asked him about it, he would frown and say: “There’s something considerably subtler going on there.” For this, the third professor was held in especially high regard. Petrov wrote seven papers about the problem. The first was called “Some Remarks on the Question.” The second was called “Some Further Remarks on the Question.” The third was called “On the Impossibility of Certain Remarks on the Question.” Petrov wrote the fourth in English, which made it seem especially serious. Nobody read the fifth. The sixth was read by one man in Canada, but it later turned out that he had confused Petrov with another Petrov. Petrov decided not to publish the seventh. It was the best one. Several more years passed. Petrov developed gray hair. Then he developed more gray hair. Then he developed less hair. The problem did not change in the slightest. One day, a computer appeared at the institute. The computer was installed in a separate room. At first Petrov paid no attention to it. “A machine,” Petrov would say. Then the computer began writing papers. Petrov became interested. Then the computer began proving theorems. Petrov became concerned. “What sort of theorems?” he asked. “Various ones,” he was told. Petrov did not like this. “My problem is beyond it anyway,” Petrov said. And, just in case, he closed the door to his office. A week later, a young researcher named Sidorov knocked on Petrov’s door. “Come in,” said Petrov. Sidorov came in. “Pyotr Nikolaevich,” Sidorov said, “the computer has solved your problem.” Petrov looked at Sidorov. Then at the window. Then at Sidorov again. “Which problem?” “Yours.” “I have lots of problems.” In fact, Petrov had only one problem. “That one,” Sidorov said. “Impossible.” “Possible.” “It solved it incorrectly.” “We checked.” “Who did?” “The computer.” “The computer checked the computer?” “Yes.” Petrov thought about this. “That’s not serious,” he said. Sidorov then placed a printout in front of Petrov. The printout was four hundred and eighty-three pages thick. Petrov looked at the first page. Then at the second. Then at page four hundred and eighty-three. On page four hundred and eighty-three there was a little square. Petrov became very angry. “I could have put a square there too,” he said. “The square isn’t the point,” Sidorov said. “Then why is it there?” Sidorov did not know. The next day there was a seminar. The computer did not attend. This reassured Petrov somewhat. Sidorov gave the talk. For two hours he explained the proof. Petrov shook his head the entire time. After the first hour, he said: “There’s a mistake here.” “Where?” Sidorov asked. Petrov pointed. It turned out to be the page number. Twenty minutes later, Petrov said again: “There’s a mistake here.” This time it turned out to be a paper clip. After that Petrov was silent for a while. At the end of the seminar, the director of the institute stood up and said: “Colleagues, an extraordinary event has taken place.” Everyone applauded. Petrov applauded too, because at first he had not understood what the director was talking about. Then he understood and stopped. The director continued: “A problem that remained open for more than forty years has been solved.” “Twenty-three years,” Petrov said. “What?” “I worked on it for twenty-three years.” “Very good,” said the director. “Then you must be especially pleased.” Petrov was not pleased in the slightest. After the seminar, everyone went to have tea. The computer was not given any tea. Petrov saw a certain justice in this. The next morning Petrov came to the institute as usual. At half past eight he entered his office, took off his coat, sat down at his desk, and took out a notebook. Then he remembered that the problem had been solved. He closed the notebook. A minute later, he opened it again. Then he closed it again. At ten o’clock Petrov went out into the corridor. “Sidorov!” he shouted. Sidorov came over. “What am I supposed to do now?” “What do you mean?” “The problem is solved.” “Take another one.” Petrov was astonished. “What do you mean, another one?” “Another problem.” Petrov looked at Sidorov as though Sidorov had suggested that he take another surname. “That was my problem.” “Now it’s solved.” “That is exactly why it was mine.” Sidorov did not understand and walked away. Petrov returned to his office. Twenty-three notebooks lay on his desk. In the first notebook, it said: “Let us try to prove the following.” In the last notebook, it said: “Let us try another way.” Petrov stared at those words for a long time. Then he turned the page. The page was blank. It was the first completely correct page in twenty-three years. At noon, out of habit, Petrov said: “No.” But this time nobody had asked him anything. A month later, the director summoned Petrov. “Pyotr Nikolaevich,” the director said, “we need to talk.” “About the problem?” “No.” Petrov immediately understood that the conversation was going to be bad. “You see,” said the director, “computers solve a great many problems now.” “I’ve noticed.” “And they do it quickly.” “That is their weakness.” “Why?” “They don’t have time to understand what they’re doing.” The director wrote this sentence down on a piece of paper. Petrov brightened. But the director was only testing his pen. “In any case,” the director said, “your position is being eliminated.” “Why?” “Optimization.” “What does that mean?” “It means your position no longer exists.” Petrov thought about this. “And me?” “You still exist.” “Without a position?” “Yes.” For a long time Petrov could not understand how a person could exist without a position. Then he remembered that the problem had existed for twenty-three years without a solution, and this reassured him somewhat. On his last day at work, Petrov packed his books into a box. There were too many books. So he left half of them behind. Then he thought about it and left the other half behind too. As a result, Petrov left the institute carrying an empty box. At the door he met Sidorov. “Pyotr Nikolaevich, where are you going?” “Home.” “What’s the box for?” Petrov looked at the box. “I don’t know.” And he gave it to Sidorov. At home his wife asked: “Well, what happened today?” “Nothing.” “And the problem?” “Solved.” “Congratulations.” “There’s nothing to congratulate me for.” “Why?” “I didn’t solve it.” “Who did?” “The computer.” His wife thought for a moment and said: “Then it’s a good thing you’re finally free.” Petrov was frightened again. He had never been free before. The next day Petrov woke up at half past seven, drank some tea, and sat down at the table. There was nothing in front of him. He sat there until twelve. At twelve, Petrov said: “No.” Then he had lunch. After lunch he sat down at the table again. At six o’clock he said: “No again.” This continued for another three days. On the fourth day, his wife asked: “What are you solving now?” Petrov wanted to answer, but could not. So he took a blank sheet of paper and wrote: “Problem. Find a problem that a computer will not be able to solve.” Petrov looked at what he had written and, for the first time in a long while, felt good. At that moment, the telephone rang. It was Sidorov. “Pyotr Nikolaevich,” he said, “the computer has just come up with a list of seventeen such problems.” Petrov put down the phone. Then he walked over to the window. The window was closed. Petrov thought for a moment and decided not to open it. It was the first decision the computer had not taken away from him. Source: translation of Boris Bilich
46
91
19
986
124,266
Computer scientist Sara Hooker on underestimating difficulty:
70
3,531
138
32,078
494,656
We put a neural network inside the camera pixel. Introducing RETINA: a vision sensor that runs AI where the light lands
39
79
26
848
208,265
Ben Tang retweeted
Existence itself is a miracle - the rest is science. There is no single meaning to existence - if there were, we’d be slaves to it. Within existence, there are rules of logic and science. If it was magic, the world would be un-navigable. Truly the best of all possible worlds.
499
1,133
101
10,700
467,830
As a scientist, AI has made me feel the most intellectually alive and excited I have felt since I was a graduate student and postdoc more than 20 years ago. Every day I can start with an idea in the morning, and by lunchtime, I see a testable, rational, well-thought-out hypothesis forming in front of my eyes. And every day, the possibilities seem endless, like mountains beyond mountains. What a time to be alive. Here's a case in point. I'm collaborating with a professor, an experimentalist, who is trying to solve a thorny problem in his field. There's one particular molecule that he is using in his experiments that seems to result in radically different crystal structures compared to similar molecules. What's happening here? He has come up with a few different hypotheses that could explain the differences but is not a theoretician and needs to tease them apart. On Thursday, I started an investigation using AI at his bequest. The AI immediately confirmed the hypotheses that he had in mind and added a few of its own. Then it started its exploration. The investigation was carried out in three different phases, each of increasing difficulty; the first one using classical physics, and the second and third using quantum mechanical techniques of increasing rigor. This tiered strategy is the right one. By Thursday evening, I had the glimpse of an answer. Most of the hypotheses had been examined and rejected. Two stood out, although the AI identified one as more a mechanism through which the other one operated rather than a root cause. It immediately pivoted to the higher-level, more rigorous calculation. Every time I interacted with the AI, it was more like a dialogue between a professor and a bright student or scientific collaborator than a mandate issued to a tool. The feeling was very much of a process where the AI and I were solving a problem together. I steered the conversation several times, pushed back, suggested course-corrections, acknowledged my own wrong ideas as well as the AI's and went back and forth. The AI was successful in keeping multiple requests in its memory, stacking them by priority while never losing the conversation thread. By late Friday morning, there had collected enough data from the more rigorous calculation to corroborate the suspicion that it was really just one hypothesis that was the root cause. It then moved on to the next step, which was to come up with a distinct set of novel molecules that would confirm the hypothesis beyond any reasonable doubt. In addition, it launched an even more rigorous calculation at a higher level of theory. By the end of Friday, roughly 48 hours later, using this multi-layered approach of increasing rigor, backed up by references, and made useful and actionable by testable experiments, the AI had arrived at a solid, rigorous conclusion. Now imagine doing this every day, about any topic under the scientific sun, in any scientific field, so that your intellectual labor is multiplied a million-fold. Mountains beyond mountains. What a time to be alive.
88
255
55
1,930
398,742