Atanas Stankov

illinois gov. JB pritzker responds to ken paxton's threat to arrest TX democrats who fled the state: "they're here in illinois. we're going to do everything we can to protect every single one of them."

Different languages, similar encoding efficiency: Comparable information rates across the human communicative niche Human languages encode similar average information rates (~39 bits/s) despite their remarkable differences.

“… coupling between language-level (information per syllable) and speaker-level (speech rate) properties results in languages encoding similar information rates (~39 bits/s) despite wide differences in each property individually”

www.science.org/doi/10.1126/...

So I think it’s fair to push back against the optics of describing via harmful/not harmful. One need to “against method” like Paul Feyerabend’s philosophy and conspiracy theory thinking is basically that in the wild: untamed and uncalibrated. It’s these optics got us scientists defunded.

To add more context. Critical appraisal only works if there is community that has an agreed rule set like academia. Conspiracy theory thinking doesn’t care about that bc it’s very solipsistic in nature. Having been through it in early college, I think you don’t understand it until you experience it.

Hello. Is there a stereotypical emergence pattern between conspiracy theory and other emergent group dynamics? I’ve become skeptical of clinical analyses because they can be bias selective (We know how it’s destructive). It’s a very robust pattern that pops up in all cultures and languages.

The wire called this “cooking the books” or “junking the stats”. Straight out of meme. They are p-hacking the justification of their policy decisions.

Are foundational models/NeuroAI/AGI benchmarks all neuro-engineering rebranding? So many influencers at large tech firms have spoken about this in PR spin: Francois Chollet, Dario Amadei, Demis Hassabis, and many others. Elisabeth Bik’s work inspired me to me make these kind of posts. 🪏🧐

📌

An important metric in social media posts is the ratio of likes to comments or likes of an original post compared likes to a replay post/comment. If in either case the ratio is less than 1, it’s a net negative appraisal. This has emerged since the removal of the dislike button on many platforms.

Vibe physics? … I like how general-use AI tools brings out deep beliefs people hold into the public. Hypothesis: AI use is a social sorting function like prior new media (printing press, radio, TV, and social media) and AI is going to catalyze social change in ways its developers did not intend.

PS: He made an follow-up post for better quantitative measurements. mathstodon.xyz/@tao/1149100...

This is the kind of paper to post on the office door. www.nature.com/articles/s44...

Terry Tao takes critical position on AI non-transparent benchmarks. Or maybe I’m misinterpreting it? But even the idea he might be doing so is a game changer. mathstodon.xyz/@tao/1148814...

I find it concerning Twitter is still more active discussion-wise given its level of toxicity, suppression, and right-wing bias. What is happening? @jackdodo.bsky.social

#BookSky

YouTube video by ThePrimeTime AI Outlawed ☠️ in Open Source Project

🌶️ Should this apply to scientific open source code? m.youtube.com/watch?v=Rqry...

An unforeseen Turing test that is happening is folks using LLMs to explain every single trivial online post because of laziness. Same with school assignments. The efficiency hypothesis of the brain is being tested by overpriced data centers. Let’s see how this plays out from a game theoretic angle.

To the folks trying to build world models/AGIs using RL/SCMs, how is that different from “theory of everything” attempts historically done in physics?

That keyboard is nice!

Infinite regress - Wikipedia

Full of traps? en.m.wikipedia.org/wiki/Infinit...

My PI knew a technical writer who reads Strunk and White’s The Elements of Style once every year with a fresh mind. Professional goals right there.

What’s the connection between reinforcement learning and entropy production hypotheses? Does entropy production have links to noise models? What is the substrate for all these loose heuristics? Do “lateral” theoretical insights should give insights into “vertical” theory?

📌

PS: Entropy is just such a bad metric for explaining brain representation. It’s noisy and nonspecific. If we can’t derive stochastic thermodynamics from first principles in neuroscience, not physics, then these studies are just mathematical exercises and not critical stepping stones.

The cost of cognition: Measuring the energy consumption of non-equilibrium computation In biological systems, survival is predicated on an animal being able to perform computations quickly on a minimal energy budget. What is the energy consumption of non-equilibrium brain computation, i...

Still have no idea how this will apply to large-scale recordings neuroscience and even if it applies yet.

“The cost of cognition: Measuring the energy consumption of non-equilibrium computation”

www.biorxiv.org/content/10.1...

Do recent papers on spotting “reasoning” errors in LLMs have transferable techniques for spotting spike sorting errors?

Yes just a fun metaphor. It’s not at Tarantino level of analysis.

a man is standing in front of a window and says yes , right . you did . well done . ALT: a man is standing in front of a window and says yes , right . you did . well done .

That’s the O’Neill I vibe with.

Processes and Mechanisms of Democratization Unlike Artistotle's analysis, recent treatments of democratization identify pathways and propose necessary conditions but fall short of specifying cause-effect relations. Democratization does not fol...

Sure, I like resources similar to Charles Tilly as my anchor to discussions (a go to for my job as an academic) but if we’re aligned that all is good. Cheers.

“Processes and Mechanisms of Democratization” onlinelibrary.wiley.com/doi/abs/10.1...

… slow collapse of oil industry, rise of clean energy breakthroughs in solar and nuclear fusion, and many others. At this point just engagement alone on this topic is enough. I digress.