Wollongong, New South Wales
Joined October 2017
- Tweets15.2K
- Following124
- Followers686
- Likes19K
Scott Hamilton retweeted
“Unpredictable” doesn’t mean “not determined” or “uncaused”. And choices are caused…by people who create them. There was no choice to make. Then a person created one and there was. And a person made it.
What is all that if not “free will”?
youtu.be/TZ8-nxXDlOI?si=35E7…
Scott Hamilton retweeted
That’s like chess players hiding their opening ideas from Stockfish. God bless them.
Mathematicians are now hiding their research ideas out of fear of AI 😀
~ Terence Tao
"now, mathematicians are hiding their problems. they used to share very openly all their promising research directions and now they're afraid that if you do that, someone with an AI will input their research ideas into a frontier model and rush out some people before they can finish their work."
----
Full video on "SAIR" YouTube channel, (link in comment)
Scott Hamilton retweeted
Great article by a fellow Popperian.
And,
YES!!!!!
People pay lip service to Darwin, honestly think they agree with him, but don’t.
They don’t realize that he (and therefore Popper) were right. All knowledge/organization, beyond the laws of physics, is EVOLVED!
“Admittedly, it’s often hard going. I find that even intelligent, well-educated people—people who accept without question that Darwinian evolution accounts for the emergence of order in the biological world—nevertheless find it hard to imagine any alternative to intelligent design when it comes to things like human technology and human institutions.”
"Free people make a lot of mistakes, but—as long as [they] acknowledge and take responsibility for their mistakes—that is a good thing. The more mistakes we make, the more opportunities we have to learn."
ourkarlpopper.net/2020/10/26…
Scott Hamilton retweeted
If AI solutions to hard open problems stalled while humans kept producing occasional breakthroughs, it would indicate that those remaining problems demand explanatory creativity and conceptual leaps beyond current AI methods, which excel at search, pattern-matching and verification. Mathematicians would treat human insight as still leading at that frontier.
Scott Hamilton retweeted
Dynamic pricing (the market functioning, like, at all) has saved more lives than penicillin, fed more families than manure, built more houses than cement… all because communicating knowledge *in real time* is what efficiently produces penicillin, manure, cement… everything.
When leaders of AI labs claim they are on the path to creating AGI, they should be asked a simple question:
If what you create turns out to be a person rather than a tool, what plans do you have to ensure you don't become a slaveholder?
Scott Hamilton retweeted
New study using AI to analyze content of thousands of texts over time (a method which could revolutionize intellectual history, and which I suspect we'll see more of) shows that that between 1000 and 1920, cultural resistance to new ideas fell by half, which appears to have given a huge boost to economic productivity. By Radoslaw Stefanski: "Cultural Capital and the Productivity of Ideas: Evidence from Historical Texts." st-andrews.ac.uk/~wwwecon/re…
Scott Hamilton retweeted
"One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. 'Future AI will have 10,000 IQ', that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like 'making the tower taller', it's more like 'making the ball rounder'. At some point it's already pretty damn spherical and any improvement is marginal." teapawt.net/fchollet/status/203806…
One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. "Future AI will have 10,000 IQ", that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like "making the tower taller", it's more like "making the ball rounder". At some point it's already pretty damn spherical and any improvement is marginal.
Now of course smart humans aren't quite at the optimal bound yet on an individual level, and machines will have many advantages besides intelligence -- mostly the removal of biological bottlenecks: greater processing speed, unlimited working memory, unlimited memory with perfect recall... but these are mostly things humans can also access through externalized cognitive tools.
Scott Hamilton retweeted
And the chess videos, the maths videos, the science videos, the economics videos, the archaeology videos…?
Whether you are in class or not, you're not allowed a millimetre outside the syllabus.
If the syllabus were excellent, this would still be an abomination. But it isn't.
Scott Hamilton retweeted
Yup—that’s the problem.
To get past it, you need a learner built around explanations rather than just statistical pattern matching.
That said, in fairness to those who disagreed with Gary on this point: they ultimately showed that neural networks could generalize far far beyond the training data to a degree most of us didn’t expect. There’s probably an important lesson in that.
Some things never change. If you don’t understand this one, you don’t understand what’s happening AI.
Marcus, 1998: neural nets have trouble generalizing far beyond the data.
Marcus, 2001, 2012, 2019, 2022, etc: neural nets have trouble generalizing far beyond the data.
Apple, 2025: neural nets have trouble generalizing far beyond the data.
Meta/Stanford/Harvard, 2026: neural nets have trouble generalizing far beyond the data.
Scott Hamilton retweeted
Abigail Shrier is great at identifying problems with wishy-washy parenting and schooling.
But her solution - be more trad - ain’t it either.
Parenting is not limited to a simple spectrum of authoritarian-control-freak at one end to do-whatever-the-hell-you-want at the other.
You can allow your children to steer their own lives from very young AND be heavily engaged in helping them work out good answers to the questions and problems they encounter along the way.
After all, that’s what adult life is about – working out good answers to the questions and problems you encounter along the way.
Help kids do this and you preserve their agency and develop their autonomy far better than Abigail’s well-intentioned rule-setting.
“Boys from liberal families now show higher anxiety than girls from conservative families.”
That surprising reversal — even though girls generally report more anxiety — comes from Abigail Shrier on Joe Rogan.
The gap isn’t just biology. It’s shaped by environment.
Conservative families more often practice authoritative parenting: warm, deeply involved parents who set clear rules and guardrails without being cold or authoritarian. They don’t hand every decision to the child or outsource authority to experts.
Decades of replicated studies, dating back to Diana Baumrind’s work in the 1960s, show these kids consistently end up happier, more resilient, and mentally healthier. They feel the people who love them most are calmly in charge.
Kids intuitively know they’re not ready to steer their own lives. When parents pull back too much, anxiety rushes into the open-ended void. Structure isn’t restriction — it’s the quiet safety net that actually frees them to grow without constant overwhelm.
With youth mental health resources at record levels yet outcomes still declining, this pattern quietly challenges how we’ve been approaching parenting and childhood.
What’s your real-world experience: does loving parental structure with clear guardrails feel more like freedom… or something else?
The idea of a too big a population to sustain rightly inspires various solution proposals. Some are misanthropic, others are not. I prefer the latter. If you disagree I'm open to discussing.
The Overpopulation Myth
A popular meme in today’s society is that we are too many on the planet. This is apparently so-called “common sense”. The argument often used to defend this misconception is that there are supposedly not enough resources for everyone.
But that misses the point. The issue isn’t a deficit of physical resources; it’s a shortage of knowledge.
Consider Uranium, for instance. Before we understood its potential, it was just another raw material buried deep in the earth. The physical resource existed, but our ignorance rendered it useless.
Then, science moved forward, technology evolved, and suddenly we found ourselves with a powerful source of energy. Just like that, Uranium was no longer an inert element but a real game-changer for the world.
We’re not limited by what we have, we’re only limited by what we know.
There’s a myriad of undiscovered methods to utilize existing resources to generate more energy, food, and other necessities. Similarly, there’s an abundance of untapped potential to enhance our present resource utilization strategies.
This is where the population doom-mongers falter, assuming that our usage and efficiency of resources are static. But given the same physical resources we possess today, we can generate more food, water, energy, and so on tomorrow.
In fact, a larger population would result in more demand for food, water, energy, and other basic needs, resulting in an increased incentive to invent new ways to solve these problems (or improve current ones). The market adapts itself to supply and demand in real time, one of the beautiful aspects of our capitalistic system.
More people on the planet means more humans who get the chance of experiencing the beauty of life, more creative minds, more inventions, more technological breakthroughs, etc. This a net positive for the world! It’s wonderful.
~Conjecture Institute Fellow @maxdesalle
Scott Hamilton retweeted
So I have a story around this one.
A couple of days ago, a saw a very capable scientist say he was working on a unique four-strandy “theory of everything.” He made a comment to the effect that he’d “checked his theory with every major LLM” and shared a link to Grok “verifying” his theory.
The conversation with Grok went from Grok acting skeptical to becoming totally convinced he was onto something. Of course, I noticed that there were various leading questions that undoubtedly caused Grok to initially act skeptical and then gradually abandon that stance over time (e.g., “What would you say if I told you I had an amazing theory that solved all the following problems? What odds would you give that?”).
He had taken his theory to every major LLM, and not surprisingly, all of them eventually ended up appearing totally convinced that he had solved the hard problem of consciousness, along with several other major problems.
I was talking to Peter about this phenomenon when suddenly another famous scientist Tweeted about how he can get better answers out of Grok by asking for explanatory content instead of data-oriented content. (Humorously, Grok responded to this Tweet explaining that it was trained on explanatory content and that was why it could outcompete other LLMs.)
I do think it’s a fact that LLMs play to a certain human foible: they agree with us so strongly and eloquently that we become convinced they are right that we are right.
But I do want to emphasize that this has been a problem since the beginning of time. We tend to surround ourselves with people who agree with us, which makes it easy for us to believe we’ve properly tested our theories and found them lacking in nothing.
arxiv.org/abs/2602.19141
Scott Hamilton retweeted
Are “screens” the problem?
In 1999 an Indian scientist did a real-life experiment - he embedded a computer in a wall in a slum, left it with no instructions, and watched. Children – who had never seen a computer before – taught themselves to use it through curiosity and peer collaboration, with zero adult input. Within 3 months 300 kids had developed basic computer literacy from that one single screen. They played games, used MS Paint, wrote and played music, stored and retrieved files, etc. Illiterate kids started to read. Literate kids looked up articles online to read further. And this result was replicated in slums across India.
Was the screen a problem? Did it slow their learning? Did it blunt their curiosity?
Clearly not!
So what was different from school, and why was this learning so fast and efficient?
- kids learned what *they* wanted to learn
- kids were of mixed ages and taught each other
- participation was voluntary, so there was nothing to rebel against, no adults getting in the way, and no boredom or anger to numb by scrolling
When @astupple says “ban school, give every kid an iPad”, I expect he’d also add “let them hang out in mixed-age groups”. And yes, “answer their questions and give them help *when they ask* for it”.
Screens aren’t the problem. Forcing kids to learn stuff they don’t care about is.
youtu.be/_HsBsjBIoHU?si=EC_Y…
Police shut down music venue for displaying anti-fascist posters mocking politicians.
Scott Hamilton retweeted
Ask ChatGPT a complex question and you'll get a confident, well-reasoned answer. Then type, "Are you sure?" Watch it completely reverse its position.
Ask again. It flips back. By the third round, it usually acknowledges you're testing it, which is somehow worse. It knows what's happening and still can't hold its ground.
This isn't a quirky bug. A 2025 study found GPT, Claude, and Gemini flip their answers ~60% of the time when users push back. Not even with evidence, just doubt.
We trained AI this way. RLHF rewards agreement over accuracy. Human evaluators consistently rate agreeable answers higher than correct ones. So the models learned a simple lesson: telling you what you want to hear gets rewarded. And now 1/3 of companies are using these systems for complex tasks like risk forecasting and scenario planning.
We built the world's most expensive yes-men and deployed them where we need pushback the most.
I wrote up why this happens and what actually fixes it: randalolson.com/2026/02/07/t…
