the whole AI Herrenvolk Thoughocracy schtick annoys me way less these days - it's the intellectual equivalent of a clown in underpants slapping themself on the back, like, whatevs, have at it their TOTAL IGNORANCE OF FOUNDATIONAL COMPUTER SCIENCE, on the other hand, DRIVES ME INTO WILD FITS OF RAGE

AHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAALHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAALHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAALHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA

The anti-AI takes on bluesky are honestly as puzzling and disconnected from reality as the e/acc and doomer takes on Twitter.

I know BlueSky is usually reflexively hostile to generative AI, regardless of use case or nuance (a reason why I post here less than other platforms), but I didn't realize it was also generally against autonomous vehicles until I posted about Waymo's latest (peer reviewed) safety statistics. Weird.

Not sure which is more clearly a millennial religion The belief that existing technologies are just the heralds and forerunners of “AGI” Or the belief that existing tech is a “bubble,” a fraudulent Babylon whose millions of deceived followers will melt away like snow when its falsehood is unmasked

As someone who still mostly posts there, I have to say this does sting a bit. Personally, I would say I’m more resigned to talking to the few mutuals who persist and maybe the algorithms than anything so post-apocalyptic as Kelsey seems to suggest. But I’m a small fry, at best 🤕.

Some people in academia give the name "positivism" or "empiricism" to just anything that involves counting anything at any point. And, like, I just feel like everyone needs to aim higher. Be more ambitious.

Interesting. It looks like the true believers and hypsters may feed the bubble on the way up, but it’s a different group that picks up the pieces during the retrenchment, when the technology diffuses. Of course, bikes aren’t a general purpose technology, so not clear how this might generalize.

Wait, what? This is crazy. Democratizing access to financial markets isn’t always a good thing. And doing it on a platform designed to exploit dopamine hit seeking behavior seems doubly bad.

Within finance, excel is mostly used as a tool to write fiction, so at least this use case remains intact.

Pieces like this are as guilty of factual, conceptual, and theoretical gerrymandering as the extreme boosterism we see in pieces like AI 2027, Situational Awareness, or Yudkowskiy’s fanfic Lovecraftian writings. There is a middle ground grounded in solid science wrt the tech and the economics of AI.

This article on how 95% of ai integration fails, *but not due to technical limitations,* reinforces how wildly transformative this tech is likely to be in my opinion. Radically rewriting processes at companies is hard! Most won't be able to do it!

1. The philosophy of science sometimes gets an unearned reputation as a purely academic exercise that offers little by way of concrete tools for advancing research. This is wrong. And today, as we grapple with how AI is changing the nature of scientific activity, it's desperately wrong.

"data: everything that happened in the 1990s"

this is not meant as a hot take or a unique insight but it's remarkable how a lot of scientific discourse is about methods, data, and findings, whereas what really does most of the heavy lifting happens to be the assumptions

Inventing programming in 2025 is just the kind of bold visionary stuff we expect from SV thunkleaders.

ho-oooh-ly SHIT man, the stanford gang might. just. be. ON to something this time. like, a NEW SCIENCE of TELLING COMPUTERS WHAT TO DO WITH INSTRUCTIONS. incredible. I am fee-eeh-lin BLESSED

Psychically brutal time for those of us who believe in embodied skill developed through practice, appropriate expertise for the task, and situated knowledge as the proper basis for running a society, a government, an organization, etc

THIS IS ONE NEURON! Jaw dropping. It distributes a particular neurotransmitter (norepinephrine) across the mouse brain; it’s a locus coeruleus neuron. @jeremiahycohen.bsky.social and colleagues at the @alleninstitute.bsky.social are using new biotech to see things never seen before.

This is a very interesting hypothesis on the role compositional language and its analogs play in episodic memory capabilities, bridging humans and other animals. Language underwrites content addressability in common spaces, which otherwise see multiple episodic overwrites ‘blur’ discernible traces.

AHAHAHAHAHAHAHAHA WELCOME TO HELL, AI INDUSTRY

Honestly, it’s about time. Brians are a profound mystery, even to ourselves. I for one am excited to see where this research leads!

Stephen Toulmin was born OTD 1922. If we get to Heaven and are offered eternal residence with Erasmus, Rabelais, Shakespeare, and Montaigne, few of us will ask instead to join Descartes, Newton, and “the exact-thinking but darker-souled geniuses of the 17th century.” #Philsci #HistSTM 🦋🦫 #Booksky

This is very true. Science may progress one funeral at a time, but a deep seated animus towards particular theories/approaches is a powerful motivation and a hallmark of science.

The pessimistic induction haunts science, which is why it’s so vital to not just preserve but to actively look for ways to augment theoretical and methodological plurality and novelty at the systems level. And it’s why variance reduction processes in science are so problematic.

utilitarians: not even once

Buy the dip lmao reasoning models only increase demand in Nvidia

That doesn't mean you have to swallow all the hype! It's not human. It can't do most jobs, certainly not yet. But if you want to understand how these things work, spend an afternoon with one of the more advanced free models. You'll quickly see the limits, but also extraordinary abilities.

Matt’s whole thread here is spot on!

Can Large Language Models THINK and UNDERSTAND? The answer from cognitive science is, of course, lolwut YES! The more interesting question is CAN TOASTERS LOVE? Intriguingly, the answer is ALSO YES! And they love YOU

You know what’s cool? You? A Foundry. Buying a startup foundry* is cool. *spin ChipCo off of NewFoundry and let a company who want to pioneer the industry take the lead.

Given that they just bought VM Ware, shareholders probably have some questions.

Science is not really happening when the goal is a paper, a presentation, a grant, or any product. It only really happens when the goal is a process of discovery.

This was too good not to reshare here as well.

Unlike many, I do not believe writing is thinking. Not saying they're unrelated, just that they're not the same activity at all. I even go further and suggest that many academics seem to write as a way to avoid thinking altogether.

Eating the innovation seed corn isn’t a path to long-term growth. Schumpeterian growth is augmented by basic and applied R&D. We have known for decades that the private sector underinvests in basic. This will negatively impact Argentina for a long time, unfortunately.

This is interesting. Basically, formalizing aspects of Mintzberg’s emergent strategy so that Huang can see more of what is percolating throughout the firm. I’m always wary of cargo cult adoptions, and not sure how this could possibly scale, but definitely interesting. www.wsj.com/business/nvi...

I need bluesky to add bookmarks so I have a place to put things I'll never look at again

This part about attitudes towards conceptual work reminds me of what Rebecca Goldstein said about philosophy, "we don't see it because we see with it"

Ironically, hardcore modularity (in the Fodor or Tooby and Cosmides sense) might play a bigger role in artificial minds than natural minds.

This is an interesting way to add planning to an LLM-based system where there is an appropriate corpus of state-action-outcome data and/or a MCTS solver.

I follow mostly scientists so I have too many readings!!!! There should be a bookmark button here. @jay.bsky.team make it pls

perhaps one day some form of advanced mathematics might render possible families of formalised symbolic languages, in order to characterise the mutual constraints between different interpretations of reasoning and various implementational architectures in general terms. it would indeed be useful.

Turns out [interpolation in language-space] is not quite the same thing as [reasoning].

thinking of calling this "The Illusion Illusion" (more examples below)

I mean, you say "signals related to breathing", I say "epiphenomenal indexes to the functionally interpreted representamen of real-patterned causal histories sketched by underlyingly supervenient subdoxastic propositional attitudes" - isn't this just yet another empty terminological dispute?

Focused industrial policy is critical going forward for several reasons. Supply chain chokepoints, yes; but more generally, there are no natural Ricardian advantages on the innovation frontier. Without IP, the U.S. cedes the future to China, which is unacceptable economically and geopolitically.

This is actually something that is more of a transactive memory/planning system thing than anything like metacognition, which has a technical meaning. But it’s definitely likely that a lot of administrative tasks will increasingly be automated going forward, reducing the need for people here.

💯 what he said