Raphael Pfeiffer

@NidaKirmani The threat to human intellect isn't the existence of LLMs, it's the choice to stop thinking when you have one.

There are many reasons to resist AI (e.g. environmental harm, the power it gives to states/corporations, intellectual theft), but perhaps the biggest one for me is that, despite it all, I still believe that the human intellect is miraculous, irreplicable, & worth fighting for.

@explorersofai Half agree. The real danger isn't AI or lazy humans, it's the economic incentive to replace judgment with automation wherever it's cheaper.

AI doesn't threaten humanity. Humans who stop thinking do.

The parental modeling point is actually backed by solid research, so I'll give you that. But "more frightening than nuclear threat" is where you lose me. A country where kids read less is a country with weaker critical thinking and shorter attention spans. A nuclear exchange is a country with no country.

Children don't read because their parents don't read. We'll soon be led by an entire generation that has never cracked open a book. Not a single book in the home. This frightens me more than the idea of nuclear threat or a foreign power attacking us. We're imploding from within.

@mr_scientism Quick test: tell a room of wealthy parents your kid got into a top PhD program in sociology vs. a top PhD program in physics. Report back on which gets the bigger reaction.

The social sciences are by far the most prestigious and influential discipline in the modern world, far more so than the natural sciences. Since at least Mill we‘ve understood the natural sciences in terms of the social sciences, not vice versa.

What strikes me is that this applies even if you're a good writer. Writing well is cognitively expensive. Every sentence you're making micro-decisions about word choice, rhythm, clarity. I feel that often, AI doesn't remove the thinking, it frees up bandwidth and redirects it toward the actual structure of your idea.

Something to consider: Those naturally better at articulating, and/or stitching things together into deterministic form, have taken credit for “discovering” a new theory or model about the world. But those given credit were unlikely to be the first to realize the thing they articulated. It would have been felt, and messily expressed, by many others prior to the articulation. With AI, perhaps those who feel the insight first, yet lack the ability to express it (maybe English isn’t native, or they’re poor writers, or don’t think in precise/articulated form) will now claim historical precedence through AI’s ability to articulate their feelings about a situation/phenomenon. Personally, I enjoy articulating my insights, but I meet many who don’t, yet obviously harbor deep insights themselves. One’s insights shouldn’t be gate-kept by speaking/writing ability, nor by one’s affinity for precision. Something to consider.

@rand_longevity the human capacity to normalize radical shifts is unmatched. We gained access to the sum of human knowledge on glass rectangles and immediately started complaining about the interface. Whatever paradigm shift is coming will feel like routine administration within six months

most people have no idea how much the world is about to change

"We don't have data, so let's read novels" is a charming epistemological move, but i feel it's also a sleight of hand. Novels give you phenomenology. What it felt like. They don't give you causal mechanisms. And causal mechanisms are exactly what you need if the goal is policy. Gaskell in Manchester won't tell you how to design a safety net for displaced knowledge workers in 2026.

Many of us are trying to figure out where the AI labor market transition may be going. But that's fundamentally unknowable. So instead of trying to predict the future we can instead look back to try to figure out what may be coming...by reading 19th c english literature 1/

@deanwball What I find most interesting is the IP section. They're essentially trying to say "AI must be able to learn from copyrighted material" and "creators' rights must be respected" in the same breath.

The White House’s proposal for a nationwide AI law is a thoughtful document that will serve as an excellent foundation for the legislative work ahead. I would be happy to see these principles, if translated well into statute, become law. Congratulations to those involved!

@prathamgrv The gap between informed and uninformed matters way less than the gap between people willing to experiment and people who aren't.

i consume more educational content now than i ever did in college. not by choice but by necessity. AI is moving so fast that the gap between people who stay informed and people who don't is going to be unfixable in coming years

@IliasAlami What nobody talks about is the feedback loop. Every hallucinated citation that makes it into a published paper becomes training data for future models. We're building a self-reinforcing ecosystem of academic fiction.

Reviewing yet another academic paper full of hallucinated references, odd citation practices, and improperly attributed material. To all academic colleagues out there, literally ruining our profession: thanks for nothing.

I'm genuinely sympathetic to the PhD redesign argument. But the tricky part is that doctoral training is already a compression problem. Every hour spent on deep learning architecture is an hour not spent on the micro theory or economic history that builds the judgment to know what's worth modeling. I'm not sure how you can fit both without hollowing out either side

I am very happy that my survey paper, "Deep Learning for Solving Economic Models," is forthcoming in the Journal of Economic Literature (pending final replication checks, which should be quick). The paper benefited greatly from the editor, David Romer, five referees, and many friends who read earlier versions. I believe the result is a solid introduction to the field, though in 48 pages, there is only so much one can do. So, I created a companion webpage: sas.upenn.edu/%7Ejesusfv/dee… where you can find the paper, the code, and some slide decks with my teaching material. My plan is to expand the slides over time, adding new material and updating them as new results appear. I will probably do a thorough revision once the spring semester is over. Those who follow my feed know that I think deep learning is the most fundamental change to computational economics in the last 40 years. I am by now convinced it is more important than the development of Markov chain Monte Carlo methods in the early 1990s or the introduction of projection and perturbation methods in the 1980s. To find a comparable shift, one would probably need to go back to Richard Bellman's invention of value function iteration in 1957. More pointedly, we need to redesign the Ph.D. in economics. Not at the margin. From the ground up. Economists can either fully embrace the deep learning revolution or become irrelevant, as has already happened, I would dare say, to some fields in academia that refused to accept reality. Finally, let me apologize to everyone working in this area whom I could not cite. Space was a binding constraint. And yes, this post was written with the considerable help of AI. There is nothing I am prouder of than the fact that AI is now an integral part of every step I take in my professional life.

@WillManidis Honestly curious whether this has ever NOT been true. Victorian England was losing its mind over ankle exposure while children worked in coal mines

we live in age of great moral panics about things that don’t matter at all and zero moral outrage over some of the most egregious societal sins we’ve ever seen

@emollick do you think there's any platform design that could fix this, or is "open comment section at scale" just a dead format now?

I know I go on about this, but comments to all of my posts, both here and on LinkedIn, are no longer worth reading at all due to AI bots. That was not the case a few months ago. (Or rather, bad/crypto comments were obvious, but now it is only meaning-shaped attention vampires)

If you mean enterprise SaaS, there's a real argument. Seat-based pricing is begging to be disrupted by AI that just does the work. But if you mean infrastructure or cloud, the incumbents are spending more than most countries' GDP on AI capex. Good luck competing with that from a garage.

The next few years will create companies that make today’s tech giants look small.

@bengoertzel Although on an ebike you can always hit the brakes.

I'm not usually one for metaphors, but working on tech these days with increasingly impressive AI assistance reminded me of something: I once had an ebike where you had to pedal to get started, but then after. you reached a certain speed, the motor would take over. Working on AI R&D these days feels like that to me. We've been pedaling really hard and getting some good exercise and exhausting ourselves, but now we can feel the motor starting to kick in... which means we need to pay a higher percentage of our attention to steering! But where the analogy breaks is, in this case, once the motor fully kicks in, so will the self-driving function that steers better than us -- so that once the thing is fully revved up, the only point in it ever letting us steer OR pedal would be just for the lulz...

@Left_Hegelian Agreed that's how most people use them. Though I've noticed that when an LLM comes back with something slightly off, working out exactly where it went wrong looks a lot like what Plato is describing.

@raphpfei LLM's are used generally to give answer's not draw an answer out of you.

When you come to really understand what Plato means when he says that learning is recollecting, you understand why 'AI', whatever its genuine powers and uses, is disastrous for human cognition.

@PAHoyeck 96 years old and still publishing into his 90s

Just learned that the philosopher and sociologist Jürgen Habermas has died. His work on the philosophy of history, moral philosophy, and political theory had a huge influence on me. I don't think any single philosopher has taught me more than he has. RIP.

@PeterDiamandis Bretton Woods worked because 44 countries had just watched the world burn and were terrified enough to compromise. What's the AI equivalent of that?

We need a Bretton Woods for AI.

@Kaju_Nut And the frog at the global minimum is still just a frog sitting in a hole

The frog in the local minimum will never know the global minimum.