filipcodes

These days I think Yorke is circling around an existential problem of a specific cognitive style, one that struggles with balancing abstract with sensory content of experience. The song Nude is the most direct expression of it - and the opposite/corrective is discussed in Jigsaw Falling into Place - both on In Rainbows. In Jungian terms I'd describe it as the problem intuition/sensing imbalance. There's also a fascinating relation to predictive processing model of consciousness (Karl Friston, etc.) He keeps coming back to this problem from the perspective of lived phenomenology. I should really write something about it, can't really describe it adequately in a tweet...

Is there a good German word for the soul-crushing sadness you feel when listening to these albums?

@garybernhardt @adamwathan I used to be a vim guy…

@adamwathan you spend your life merging your mind with vim, then one day you wake up at and you're supposed to program by leaning back in the chair instead of hunching forward...

It kind of feels like the choices right now are: 1) Provide minimal constraints to the agent. It makes slop. 2) Provide many constraints to the agent, but the more you constrain it the worse it performs in other aspects. It makes slop. 3) Use it like a glorified text editor.

@ValerioCapraro congrats - you have just discovered marshal mcluhan's "medium is the message"

Something unexpected, and slightly worrying, is happening. Ten days ago, I posted a preprint introducing the concept of LLMorphism: the biased belief that human cognition works like a large language model. The preprint received an unusual amount of attention. Hundreds of comments on social media and forums. Reels on Instagram and TikTok. YouTube videos. Infographics for students. And now it has even made it to Forbes. It seems that I got some sort of zeitgeist. Many people were already thinking about this. Many people had already experienced it. But they were missing a name and a theoretical framework. So, here it goes: LLMorphism is what happens when people start to see themselves as language models. The psychological mechanism is analogical trasfer combined with metaphorical availability: LLMs become an available metaphor for cognition, and people project that metaphor back onto themselves. The machine becomes the model of the human. And this worries me because the risk is not only that we overestimate machines. It is also that we underestimate ourselves: our embodied experience, our goals, our emotions, our responsibility, and our capacity for understanding. * Full paper in the first reply.

@bethanyinpain @afrazhaowang lol, you mean me? :D proves my point, i guess

is anyone else experiencing a kind of "over-indexed AI-writing syndrome"? the symptom is this: more and more writing produced in the age of AI (even by essayists i used to love!!) starts to carry an AI flavor. i can smell it immediately. Am I just overfitting to the effects of AI’s own overfitting? ok some examples of what now triggers that feeling in me: 1. any sentence built around three adjectives in a row starts to smell like AI. things like “too clean, too efficient, too neutral.” 2. any overly certain statement using words like “unmistakable” or “unmistakably” starts to feel AI-coded. 3. the moment i see “what’s striking is,” i start to suspect AI. 4. and sentences like “part XXX, part YYY” (XXX and YYY are adjectives) also increasingly feel like AI to me.....

@LeRoyDesCimes been there done that. now my fav past-time is thinking myself into how extremely different people are from me. i'm still wrong, mostly, but this at least makes me appreciate the wonderful, radical diversity of how humans make sense of things

When you believe everyone thinks like you, you tell yourself you can understand the world if you just introspect hard enough

@CustomWetware @garybernhardt yeah, linters are great. anything that gives consistent deterministic feedback works as a shaping tool, as it were. once the repo acquires a style of it's own, i find llms reproduce the style they find in the codebase more consistently - though linters still catch things

@filipcodes @garybernhardt Clang-format has saved a part of my sanity. I set it up to enforce some style rules about things that bothered me that LLMs do and disabled as many as I could of its default style rules (can't disable them all). It auto-runs whenever I save in neovim.

I keep finding myself in this cycle: - Add constraints in AGENTS.md. - It follows them! - But then its output seems slightly worse in other ways. - Repeat 20 times. - Agent performs worse. Which makes sense since I'm forcing it into narrower and narrower implementation paths.

@garybernhardt reasoning: it works best if you pick from what's well represented in training data distribution. enforcing a personal idiosyncrasy just lowers performance because the signal is more muddy overall

@garybernhardt pick few battles worth fighting (mostly high level architectural), let go of everything else - especially line level style things. super uncomfortable and counterintuitive, especially for a code stylist, but that's the best way currently.

people seem to feel it outside of twitter too. i was listening to a podcast with prof. denis noble and prof. karl friston this morning, and noble - an english gentleman in his 90s - observed that his (arguably, jamesian in spirit) approach to biology, which used to be controversial and marginalized in his field for years, is suddenly of interest to a much wider community... he attributed the change to "ai" - which I thought was quite fascinating...

@filipcodes tpot is very jamesian

it is utterly crazy that a student of William James in 1905 would have to time travel through 90 years of psychology—arguably a full century—to get to a place where science started sounding like their teacher again

@JustDeezGuy it only does if you make certain axiomatic assumptions (which most logicians/mathematicians find intuitive, to be fair). but what if probability is the essential property of the world (like Prigogine thought, for example) not just a pragmatic approximation?

I wish more people understood Bayes’ theorem reduces to modus ponens in the continuum limit (and, per Cox and Jaynes, it had better, otherwise something is inconsistent).

@fchollet the mountain and the valley from daoism captures the same concept, but without importing a subjective overvaluation of mountains and undervaluation of river

A mental model for working with coding agents is that they're blind squirrels running into a maze and bumping into walls. You must place the walls (verifiable constraints) strategically so that they end up in the general region you want them in.

having read that (though likely not fully understood at this point) the intuition is something like: material-inferential normativity is what predictive coordination looks like at the social-cognitive level. the ought of inference is the high-precision shared prior, enforced through prediction-error correction in conversation - but i may be overstating the connection here...

@EricCompeau @littmath What I'm specifically interested in here is to what extent we can say that 'reasoning' models are learning inferential relationships, or perhaps, extracting such relationships from a much looser semiotic soup. The basic thought is at the end of this piece: deontologistics.substack.com/p/computation-…

This isn’t a new observation, but I think there really is a massive referential calibration problem in public discourse about ‘AI’ and its capabilities. There are at least two sides to this.

@deontologistics @EricCompeau @littmath current thinking: "reasoning" as a form of error reduction in something like karl friston's predictive processing framework...

@EricCompeau @littmath Speaking as a philosopher, I think people tend to have an overly deductive picture of how reasoning works, i.e., as the inexorable unfurling of the consequences of a given set of premises, rather than a search process over syntactic objects with that serial structure.

@AndyMasley funny!

"Nobody wants or asked for AI and the demand is a mirage, and it's dangerously addictive"

the agency = psychopathy is too strong, and honestly strange coming from an experienced therapist. my guess is much of what causes people suffering and brings them to you as patients is consequences of suppressed agency or aborted action. willingness to be misunderstood, ability to fix what's broken, and the belief in one's ability to figure things out are not disorders.

The types of people who have the highest degree of “agency” are quite literally psychopaths. Congratulations, our current collective entrepreneurship-tech-everything-maxxing-DIY culture, we created a society that rewards psychopaths, or those who want to be them, above all else.

@SIUChasmite had a great time listening to Whitehead on function of reason on my run yesterday - feels so very timely and useful to make sense of things!

Continental philosophy and pragmatism are ten times more engaged with the realities of life and how we experience the world than analytic philosophy.

@eigenrobot note that the "moronic brutedom" - if by this you mean the pre-writing oral culture - evaluated writing analogously to how you're evaluating ai (socrates in plato's phaedrus etc.) writing was seen as a regression, not advance, from the perspective of the prior default

a side effect of reading and writing is becoming able to think clearly and as we move toward postliteracy i think we will likely regress to moronic brutedom for the most part trained cognition will be a niche and suspect lifestyle