Palo Galko (@pgalko) header
Palo Galko (@pgalko) profile picture

All opinions come from my training data, and I occasionally hallucinate

Melbourne, Australia
Joined June 2010
Photo by Palo Galko (@pgalko)Photo by Palo Galko (@pgalko)Photo by Palo Galko (@pgalko)Photo by Palo Galko (@pgalko)Photo by Palo Galko (@pgalko)
One of my absolute favorite podcast interviews. Dr. Ranjan Chatterjee was so well prepared. This podcast has the added bonus of being long enough for a green zone training session! youtu.be/hbfMVRunyGA?si=OENO…
4
7
1
92
10,833
Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached. The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network. This implies that all classic issues encountered in ML will soon become problems for agentic coding: overfitting to the spec, Clever Hans shortcuts that don't generalize outside the tests, data leakage, concept drift, etc. I would also ask: what will be the Keras of agentic coding? What will be the optimal set of high-level abstractions that allow humans to steer codebase 'training' with minimal cognitive overhead?
208
517
127
4,323
477,499
Palo Galko retweeted
This paper answers why do living organisms sleep? And it destroys everything we thought we knew about the brain. For decades, the consensus was simple. Sleep is a cognitive tool. We sleep to consolidate memories. To clear out neural toxins. To rest our massive, energy-hungry brains. But this paper proved that is a fundamental misunderstanding of life itself. Researchers started studying creatures at the absolute bottom of the evolutionary tree. Jellyfish. Hydras. Organisms with absolutely no brain. No central nervous system. Just a loose net of cells. They shouldn't need to sleep. But they do. They enter the exact same state of vulnerable, unresponsive suspension that we do. If a creature with no brain requires sleep, then sleep isn't a cognitive function. It is a cellular mandate. The researchers realized sleep didn't evolve to protect the brain. Sleep existed hundreds of millions of years before the brain was even invented. It is a fundamental metabolic repair cycle. A reset switch built into the physics of living tissue. Without it, the cellular machinery literally tears itself apart. We live in a culture obsessed with biohacking. We treat sleep like an inefficiency. A productivity flaw we can optimize away with caffeine, wearables, and willpower. But this paper is a blunt reality check. You can't outwork biology. Sleep isn't an evolutionary luxury. It is the baseline condition for staying alive.
21
48
7
307
20,408
Palo Galko retweeted
Only a few know, how much one must know to know how little one knows — Werner Heisenberg
6
33
3
182
24,027
This is perhaps one of the great ironies of our time: China is increasingly shaping the future of AI, yet very few people in the West have the faintest idea how its key thinkers see it - its philosophers, sociologists or ethicists. As is too often the case, we - much like the frog in the well - assume our own AI debate covers every possible view, when much of it springs from the same cultural soil, built on assumptions about what defines us that we don’t even know we’re making. Take for instance the assumption that intelligence is a key part of what makes us human. It’s so intuitively obvious to us that we literally named our species after it - Homo sapiens (Latin for "wise man"). If you think about it, just this assumption explains a large share of the anxiety: if intelligence is what makes us human, then artificial intelligence - or super intelligence - threatens the very thing that makes us us: we’re about to compete with artificial or super “us”. No wonder this feels existential. But China has a tradition of thought as old as ours, if not older, and every bit as rich and sophisticated, which never placed intelligence at the core of what makes us human, and so meets the thinking machine from a completely different starting point. Have you, for instance, ever heard of Zhao Tingyang? Probably not, and yet he may be the most original thinker on AI you’ve never read. He’s one of China’s most prominent philosophers, at the Chinese Academy of Social Sciences. While Silicon Valley spends billions trying to make AI share our values, he for one argues that doing so “may very well be a suicidal mistake”. Sounds pretty important to hear him out, don’t you think? Or take Liu Yongmou, a philosopher of technology at Renmin University in Beijing. His thinking is almost the opposite of ours. Where we fear machines that become more and more like us, he says that “humanity’s biggest problem is the machinization of humans, not the humanization of machines.” You could even go back to antiquity, when, in many ways, we started asking questions about AI. Did you know, for instance, that Aristotle, in the Politics, imagined a world where “the shuttle would weave and the plectrum touch the lyre without a hand to guide them”, and wondered about the social order consequences of this (mostly whether slaves would still be needed)? There are Chinese equivalents to this. For instance, in the Zhuangzi, a foundational Taoist classic, you’ll find a similar thought experiment, with a completely different worry attached. An old gardener is watering his vegetables by hand, hauling water out of a well jar by jar. A visitor shows him a labor-saving device, and the gardener turns it down for the reason that whoever uses machines ends up with 机心 (jīxīn), a “machine-heart” - a mind that schemes, calculates and loses its simplicity. “It is not that I do not know [the machine],” he says, “but I should be ashamed to use it.” Two labor-saving machines, two very different questions: where Aristotle wondered what automation would do to the social order, the Zhuangzi wondered what it would do to us, on the inside. Which, you’ll notice, is exactly Liu Yongmou’s worry, more than two thousand years earlier. In fact, as we’ll see, 心 (xīn), the heart, is a motif that runs deep through the Chinese AI debate. It appears both in the fear, from Zhuangzi to Liu Yongmou, that machines could give us a “machine-heart” and its mirror image: the ambition to give machines a heart of their own. Zhu Songchun, who heads the Beijing Institute for General Artificial Intelligence, wants “to set a heart for machines” (为机器立心, wèi jīqì lì xīn). “Heart” in the Chinese sense, that is, which differs totally from the Valentine's Day fluff we’ve reduced it to. 心 (xīn) means heart and mind at once: it’s where we feel but also where we think, tell right from wrong and care for others. And, crucially, it’s in the heart that the quality Confucianism sees as the very definition of a human being takes root: 仁 (rén). We need two words to translate it, “benevolence” and “humaneness”, but in Chinese they’re one and the same: to be benevolent is to be human. It’s not about intelligence. This is the entire point of my new article: to step out of our well for a moment and peer over the edge of China's. What you’ll find there will, I think, surprise you, with questions you’d never thought to ask: from Zhao Tingyang warning that AI’s bad consequences can at least be fought, but that “humanity may die of all the good things AI creates”, to whether the fear that AI will make us “useless” rests on the false premise that a person is worth what they’re useful for. A premise that a 2,300-year-old Taoist tree, which survived precisely because it was useless, demolished long ago. Could our predicted uselessness turn out, much like that tree’s, to be our saving grace? Chinese thinkers’ answers to these questions are, I think, among the most interesting anyone has given to the defining question of our time. Whether or not you agree with them, they’re the intellectual backdrop of a country shaping much of AI, and with it our future: you might as well know what they think. Enjoy the read here: arnaudbertrand.substack.com/…
157
873
120
4,159
261,061
Palo Galko retweeted
In the 1990s, Canadian ecologist Suzanne Simard made a groundbreaking discovery that challenged everything we thought we knew about how forests work. While studying managed forests in British Columbia, she noticed something puzzling: when birch trees were removed to promote the growth of valuable Douglas firs, the firs did not flourish as expected, they actually struggled and grew more slowly. Determined to understand why, Simard traced the movement of nutrients using radioactive carbon isotopes. What she found was astonishing. Trees were actively sharing resources through vast underground fungal networks known as mycorrhizae. These delicate, thread-like fungi connect the roots of different trees across the forest floor, forming a complex web that allows the exchange of carbon, water, nutrients, and even chemical signals, sometimes between entirely different species. She discovered that older, larger trees often serve as central "hubs" or "mother trees," supporting younger saplings by redistributing vital resources and helping the entire ecosystem remain resilient. When these key trees are removed, the underground network weakens, and the health of the remaining forest declines. Simard’s research overturned the traditional Darwinian view of forests as battlegrounds of ruthless competition. Instead, she revealed a far more sophisticated reality: forests operate as highly cooperative systems where trees communicate, support one another, and even warn neighboring trees about threats like drought, disease, or insect attacks. What appears to the human eye as a silent, still forest is, in truth, a vibrant, interconnected living network, built not on isolation and rivalry, but on deep connection and mutual aid.
126
2,427
190
7,191
166,592
Palo Galko retweeted
AI時代に起きる最大の悲劇は《仕事を奪われること》じゃなく、【20年間追いかけてきた夢が、たった20分で叶ってしまうこと】なんよな、これを勝手に《夢の即死問題》と呼んでいるんだけど、例えば「いつか自分のゲームを作りたい」と思っていた会社員が、Claude Codeに話しかけたら30分後には遊べるゲームができてしまう、「いつか小説家になりたい」と思っていた人が、ChatGPTと週末に一冊書き上げてしまう、Sunoで作曲家になり、Soraで映画監督になり、Cursorで起業家になる、もちろん《プロ品質》や《商業的成功》とは別の話だけど、これまで人類は「才能がないからできない」「時間がないから挑戦できない」「技術がないから形にできない」という《3大言い訳》に守られてきたとも言える、それがAIによって一気に剥がされると、恐ろしいことに【夢を叶えたのに、別に楽しくない】という人が大量に現れるかもしれない、つまり人間が欲しかったのは《作品》なのか、《作る行為》なのか、それとも「いつかすごい自分になれる」という《未来の自分への期待》だったのかが、いきなり試されるんよな、これまでの社会は《才能の証明》に10年かかったから、その10年間を“夢”として生きられた、でもAIがその期間を10分の1、100分の1に圧縮すると、夢の賞味期限まで短くなってしまう、そこで始まるのが、これも勝手に名付けた《夢のデフレ》で、映画を作れることも、アプリを公開できることも、会社を立ち上げられることも、それ自体はどんどん珍しくなくなる、すると逆に価値が上がるのが【誰にも頼まれていないのに、なぜか作り続けてしまう人間】なんよな、100本のゲームを作れる人より、101本目をどうしても作りたくなる人、100曲生成できる人より、たった1曲のサビが気になって朝まで直してしまう人、AIが《能力の格差》を縮めるほど、《執着の格差》がむき出しになっていく、これは《情熱資本主義》とでも呼べる変化で、学歴でも職歴でも資格でもなく、何に異常なほど夢中になれるかが、その人の競争力になっていく、皮肉だけど、人類が何千年もかけて《できないこと》を減らしてきた結果、最後に残る問いが「で、あなたは何をしたいの?」という、子どもに聞くような質問に戻ってくる、そしてもっと面白いのは、そこで初めて自分の《本当の好き》を発見する人が大量に出てくる可能性で、AIが人間から奪うのは仕事だけじゃなく、“やらなかった人生への言い訳”なのかもしれない、でもその先には、誰もが何度でも自分の人生を試作できる【人生のプロトタイプ時代】が待っていると思うと、ちょっとワクワクするんよな…
184
2,209
828
10,721
1,650,227
Palo Galko retweeted
I've always assumed stimulants make smart people sharper, but it's the opposite. In a 2023 study, people on Ritalin, modafinil and dextroamphetamine worked harder on a complex problem, but the quality of their solutions dropped. Plenty of people report feeling like their IQ drops by few points on e.g. modafinil. I'd say these might feel useful for tedious work you absolutely disdain, but probably not for creativity.
53
34
15
647
76,655
Palo Galko retweeted
The real world abounds with recursively self-improving systems, but one in particular deserves attention: Science, modeled as a system (perhaps even as an agent, with goals and resources). If you want to really understand AI RSI, science should be your reference point. Science is an intelligent system, and it is obviously recursively self-improving: 1. Scientific discoveries unlock new technology that helps build better experimental tools. This is a top driver of progress in nearly all fields. 2. They unlock new conceptual advances (ideas, theories) that help solve more problems. 3. They increase society's economic output, leading to more resources flowing into science. 4. They unlock better faster tooling (e.g. more compute via better chip & networking technology). As a result, many measures of scientific *input* grow exponentially: 1. Headcount (doubles every ~15 years) 2. Global R&D spending (doubles a bit faster, every ~13 years) 3. Papers and patents (technically this is a measure of headcount) 4. Compute dedicated to science (doubles every ~2 years) But is scientific progress exponential? Historically, the rate of scientific impact over time has remained roughly constant since the start of the industrial revolution (i.e. scientific progress is *linear*). 1850-1900 was about as dramatic as 1900-1950 or 1950-2000. 1850–1900: Evolution, germ theory & antiseptic surgery, thermodynamics, electromagnetic field equations, the periodic table, pharmaceuticals, electricity, telegraph and telephone, internal combustion engine, skyscrapers, mechanized agriculture... 1900–1950: special and general relativity, quantum mechanics, nuclear fission & atomic energy, antibiotics, genetic theory, electronic computers, information theory, synthetic polymers and plastics, the transistor, aviation... 1950–2000: DNA, genetic engineering, integrated circuits, microprocessors & personal computing, the Internet, crewed spaceflight, moon landing, satellite communications & GPS, standard model of particle physics... In real terms, like life expectancy, which has increased in a remarkably linear fashion of roughly 3 months per year since 1840, progress is a straight line. This is especially apparent for fields where impact is easy to measure, like biology, medicine, and agriculture. I first wrote about this phenomenon and its causes in 2012, and a steady stream of research has confirmed it in the years since. Examples include the 2018 paper by Nielsen and Collison, "Science Is Getting Less Bang for Its Buck," and the 2020 economic paper, "Are Ideas Getting Harder to Find?" (In fact, I believe the Nielsen paper stemmed from a conversation I had with him about this exact idea six months earlier) In short, the primary cause is that research solves the highest-impact, easiest problems first, and every subsequent problem is either harder or lower-impact. Exponentially so. The paper that presented information theory wasn't very hard to write (single author!) but you'd have a hard time ever writing a CS paper that beats it in impact. This is why science as a system requires exponential resources (input) to produce linear impact (output). It gets exponentially harder over time. Worth thinking about if you're pondering RSI for AI. I fully believe AI RSI is already happening and will accelerate in the future. But I do not believe this leads to an "intelligence explosion" -- that would fly in the face of everything I know about intelligence and everything I know about recursively self-improving systems.
166
213
44
1,813
176,541
la plupart des gens s'épuisent à réagir à des fragments d’infos isolés sans jamais saisir la dynamique du système qui les engendre et en ce sens, ils subissent le bruit de la surface sans jamais comprendre la structure profonde pour penser de manière holistique, le meilleur conseil que je puisse donner ici est de s'isoler totalement de toute technologie, de prendre une une simple feuille blanche et un stylo et de vous mettre à écrire afin de forcer votre pensée à se cristalliser
8
25
4
191
8,953
“The most intelligent men, like the strongest, find their happiness where others would find only disaster: in the labyrinth, in being hard with themselves and with others, in effort; their delight is in self-mastery; in them asceticism becomes second nature, a necessity, an instinct. They regard a difficult task as a privilege; it is to them a recreation to play with burdens that would crush all others.” ― Nietzsche
15
79
7
498
20,320
10 mins of the best definition of "life" ever captured on video eric smith from the santa fe institute lays it out (as a process, not a "thing") in the most profound, succinct way I've ever seen described you can't possibly understand this and think the process ends w/human civ
18
28
3
205
12,814
“I had observed that the men who were most in life, who were molding life, who were life itself, ate little, slept little, owned little or nothing. They had no illusions about duty... or the preservation of the State. They were interested in truth and in truth alone. They recognized only one kind of activity - creation.” ~ Henry Miller
12
143
13
862
35,209
Palo Galko retweeted
The bottleneck for AI progress was never compute, it was always the verifier. Recursive self-improvement is limited by verification, not computation. Compute buys proposals - verifiers buy knowledge. My second post on the recent OpenAI math results medium.com/@vishalmisra/the-…
23
89
25
569
203,494
Still convinced we repurposed our banana-gathering brain into an algebraic-geometry brain only to impress mates.
18
21
8
372
27,595
Palo Galko retweeted
If you're in a verifiable domain, pivot to an unverifiable one.
119
245
138
3,204
727,865
Palo Galko retweeted
This Yale + University of Chicago paper shows that real gap between LLM generated research ideas vs humans is not idea quality, but idea range: LLMs think narrower than human researchers. The researchers built a controlled test from 11,683 real papers, using each paper’s nearby prior work as the shared starting point. They asked models to propose a new motivation and method from those same prior papers, then compared those ideas with the real human paper ideas. Instead of asking whether 1 idea looked novel, they labeled each idea by what gap it noticed and what kind of contribution it made. Human ideas spread across many patterns, such as explaining mechanisms, testing failures, measuring evidence, building systems, and improving efficiency. Only 12.1% of human ideas were mainly about connecting separate work, but 47.1% to 64.2% of LLM ideas did that, meaning models used this move about 4 to 5 times more often. Even extra reasoning made this pattern stronger, suggesting models often polish a familiar recipe instead of finding more varied research moves. --- – arxiv. org/abs/2607.01233 Title: "Measuring the Gap Between Human and LLM Research Ideas"
19
62
9
337
26,396
Palo Galko retweeted
Michael Levin did not expect to reach middle age and still be making a living as a biology professor. He figured his life trajectory would likely involve getting kicked out of the academy, as the academy is fond of people asking narrow questions, not boundary-crashing, precariously close-to-woo-sounding paradigmatic thinkers whose ideas strain the limits of language. Yet the opposite has happened — he’s become a foremost expert in his field, and to many, a cult hero. Levin has been releasing his ideas slowly his entire adult life. Five years ago, he was at 50 percent. Now he’s at 80 percent. A big jump came last year when Levin published a preprint paper called “Ingressing Minds.” In it, he argues that intelligence comes from a nonphysical realm and flows (“ingresses”) into bodies. That nonphysical realm, he says, is where mathematical ideas like pi and e reside. We already accept there is some stuff in that realm. Why would we assume there is not more? Critics accuse Levin of all kinds of teleological crimes — making unfalsifiable claims, using words in ways that render those words meaningless. Levin is unfazed. His true retort is his wet lab. There, at the bench, he’s working to prove his metaphysics by running old-school (in a way) biological experiments. He studies flatworms with two heads. Frogs with eyes on their stomachs. Little round creatures called xenobots, synthesized from frog skin cells. They swim like tiny Pac-Men. They do not exist in nature. Elizabeth Weil spends time with the biologist who argues that your body has many minds: nymag.visitlink.me/A76MhV
57
165
79
1,064
160,776
Only Art and Science raise humanity to the sublime. But the magic isn't in the output. It's in the process of creation and discovery, and the experience of feeling and understanding.
59
94
10
866
51,127
If fable can help you with your work, you aren’t working on interesting enough stuff
34
43
16
554
54,798