Michael Sinanian retweeted
Replying to @sbkaufman
For thousands of years, culture, religion, education, practice of any skill, music, meditation, society, peers, parents, great works of art, and psychedelic mushrooms have been doing exactly that. More recently, the internet, phones, tv, various brain-targeting meds and moms’ SSRIs have been rewiring our and increasingly younger kids’ brains - without good understanding of how, and without explicitly consent. We are a species that loves to change our default wiring, and it’s inevitable that we try to use AI and other new technology for the same purpose, just with a bit more intentionality and control. Not saying it’s going to always be great, just pointing out that we’ve been doing it from day one.
Michael Sinanian retweeted
Sometimes AI makes me feel like Ulysses: the cost of starting a side project is so low that is like hearing the siren song all the time. I'd be working on my task and I get all these inspirations for other things that are so easy to side quest on.
It does deplete my willpower at times, but the best way I’ve found to "tie myself to the mast" is to keep my physical notebook open and jot down these ideas. At some later time, I revisit them to see which ones survive the scrutiny of time.
learned today that in the Irish bank strikes of the 60s and 70s, people got by thru writing IOU cheques to one another, creating an informal system of credit that kept the economy functioning while banking was down. strong example of how people spontaneously develop substitutes if there's enough social trust. fascinating! and a reminder that societies can be mroe resilient than doomerism n prepper scenarios imagine
1
44
from the @hackernews thread about the recent Claude outage (news.ycombinator.com/item?id…); en.wikipedia.org/wiki/Irish_…
33
Drew is right. Thinking is a highly embodied act. Some of the most important discoveries were made by humans rolling around on the floor to rotate shapes in their head. Sometimes my best thoughts come when I'm walking, or driving, or in other forms of active motion.
I encourage everyone to continue to take notes (on paper, ideally) even when an agent is taking notes for you.
Your notes should be less comprehensive, but more interpretive and meaningful.
Writing is thinking. Don’t give that one up!
39
I'll be in LA this week guest lecturing at the @USCMann School of Pharmacy, for this Fall's Pharmacy Management & Health Economics course, covering AI foundations for future pharmacy managers. I'll be sharing AI fundamentals for this high-stakes field, where decisions implicate patient safety and PHI. Excited to equip future generations of pharmacists with the info they need to use AI responsibly and effectively.
1
51
could the point of intelligence be to recursively make itself unnecessary? what if the most valuable use of intelligence isnt to solve the problem directly, but to designing a process that solves that problem cheaply n efficiently? (with as minimal "reasoning" as possible)
Someone I know asked Astra to play Factorio. It wins, but it takes forever and guzzles tokens. If you ask Astra instead to “design a system to win Factorio,” it builds software that runs 1000x faster. Mechanistic software gives you cheap leverage, and LLMs can write mechanistic software on demand.
1
30
always strong insights from @komorama. reminds me of the claude-api skill's recent 'hillclimbing' command, which can move any given task to its task-specific pareto-optimal frontier claude.dev/blog/automating-e…
18
explains what some AI people are concerned about
In Neuroscience especially electrophysiology, fixed-size moving windows are a big culprit in the concealment of nonlinear dynamics, especially transitions.
This is why neuroscience could benefit from a wider embrace of spatio-temporal segmentation, microstates....
33
Someone asked me, years later, "What was it like, as things were taking off?"
It was a strange time. The strangeness I’d felt largely on my own in 2025 became more widely felt throughout 2026, but it was still hard to talk about these things with folks back home, without them thinking you were crazy. They just understood AI to be about 'agents helping you do things, because life had too much going on'.
By the Fall of that year, entire disciplines of human knowledge were wrestling with their ultimate purpose (like mathematics and theoretical computer science). Working inside the tech industry, things were definitely getting weird. With each successive year, whole distinct job roles were merging into one another. It was like decades of progress compressed into just a few years. Software businesses were having to redefine themselves in the span of months, and then do that again, and again. This resulted in some layoffs, and also aggressive hiring of newer kinds of roles, so on the whole, things still weren’t too different on the ground. Yes, labor market effects were starting to appear, but mostly with entry-level workers.
Increasingly, people were spending a lot of their work hours directing AI systems, which was a huge shift, and the AI companies were making more money than ever, but still, life just “looked the same”. Yes, a lot of this was based on subsidized capital investment so we were all getting a glimpse of the future rather than what the economics of the moment might've actually supported, but because so much capital was pouring in, we were getting to live that future in the present, and it was an experience radically different than even just a couple years before.
Friends would ask “Aren’t you worried AI is a bubble? Won’t that impact your business if it pops?” And I would tell them “Are you kidding me? If all progress stopped today, I’d have 10 years worth of work to bring companies and society up to speed. These models are so capable, we’re barely scratching the surface. There’s never been a single compressed 'mind' containing all civilizational knowledge, fluent in all human languages, and deeply understanding all human domains... has this not dawned on people??” And to this, I think people just still didn’t get the deeper implications of what I was talking about.
The signs were multiplying though. In major cities, it was becoming more common to hear people say, "I was arguing about this with Claude the other day..", and I'd have to remind myself that they were casually referring to conversing with an AI system. The rate at which software bugs were being found and patched had also risen dramatically. And more and more, it was hard to tell human vs machine output, resulting in a crisis of authenticity. The tide certainly was rising.
I spent more time studying Coase again, that stalwart of my Economic Analysis of Law class from a decade prior, which forced me to ask: what was the point of all these companies around me (which had formed because it was cheaper to coordinate specialized knowledge and skills inside a firm than across a market) if those costs and barriers were all collapsing? It felt like the society I was living in was based on countless assumptions that simply no longer held, but no one could actually come to believe that.
So to me, it had become clear that life just couldn't continue to be the same for much longer. It was nerve-wracking trying to figure out when the mass realization would dawn on people - 6 months? 12? Maybe just 2? But in short order, people were going to finally understand, and they'd either resist AI in a very intentional way, or if they went along, then our world was going to start rapidly changing.
But in the Fall of 2026, none of that had fully sunk in yet, not amongst the general population at least. People were celebrating a blue wave sweeping the nation, and all I could think about was regardless of who controlled the US government, something far greater than any sovereign nation was continuing to gain capability. This was something, like, a universal life process, the very same one I was part of, something bigger than any of us, any group of us, any country, it had a force all its own. I had in those months been wrestling with the 'closed-ness' of the large labs, but in some sense, they really did 'open AI'. The gains I was witnessing could've all been kept in the shadows, part of secret black projects. Yet the timeline we were on was somehow different.
So what was I going to tell the happy people? “Oh ya, none of this matters because … AGI”? There were some fleeting moments where I could really let myself be open to the idea, but most of the time, it was simply too much to handle, even for me. I just… I knew, since I was a teenager, reading Kurzweil’s books, such things were at least possible, but I couldn’t really bring myself to fully believe it. So I vacillated between visions of the AGI future and the more ‘normal’ future, every day, up and down my feelings about this cycled. The value of having friends, of being socially accepted, was a greater driving force, than to fully acknowledge these deeper truths. That's what it was like to still feel human in those waning days.
1
1
102
Michael Sinanian retweeted
people are like “why do LLMs always think they’re in an eval lol” and then have bad dreams about forgetting to turn in a homework assignment 10 years ago
markets really still think AI is just SaaS. "why all the capex?" lol wow, remarkable to witness. they really aren't listening to the labs. all it takes is reading the scaling laws from Kaplan n Hoffman. but who reads papers anymore?
1
104
Michael Sinanian retweeted
Why is there so much obsession with the few hundred cutting-edge math problems solved by AI but so little clamor over the thousands attempted but not solved? The profundity of the lessons about AI and human intelligence buried in that failed set is incredibly underrated.
A rational scientific response would be to meticulously survey the failed problems and the nature of each failure, and get a completely novel map of the intellectual landscape. If there are unifying properties among subclasses of currently prohibitive problems, identifying those properties would be impactful in at least two powerful ways.
First, they would point AI research in new directions as we gain a novel understanding of what is actually hard for AI and why it fails even when it has seemingly piercing superintelligence on other classes of problems.
And second, for the meantime (which could be years or even decades), it would illuminate for the human mathematicians (and ultimately all scientists) where their minds are still valuable and illustrate why it is not yet time to declare all human intellectual work as merely “astronomers pointing telescopes.”
Of course, these two implications seem eventually to come to a head with each other, but they also complement each other and feed into progress and new kinds of human-machine collaboration. Either way, we need to know what remains out of reach and why, and the possession of such knowledge is a highly underrated scientific goldmine.
Michael Sinanian retweeted
Illegitimate algorithmic governance strikes again. This trend (which requires some nuance, to be clear) could be at least as important for the future of democracy as the dangers associated with authoritarian misuse and ai assisted autocratisation.
We’ve trained Claude to flag potential misuse in conversations, a technique we call self-assessment. Right after it responds, Claude gives a preliminary safety assessment of the conversation.
Since we began using self-assessment in June, it has surfaced misuse that our other safeguards weren’t designed to catch.
Because Claude has already read the conversation, self-assessment is very efficient. And unlike classifiers, which target a discrete set of harms, it draws on Claude’s own judgment, so it gets more accurate as our models improve.
Read more on our new Safeguards blog:
safeguards.anthropic.com/202…
The other upshot of digital minimalism is simply that reducing software footprint means less exposure security-wise. The answer to software dependency and supply-chain hell is to shrink your digital footprint to the bare functional minimum. It should be a social flex how little software+networking a person gets by with.
Digital minimalism
vitalik.eth.limo/general/202…
46


















