Tessera

Giosuè Baggio@giosuebaggio.bsky.social · last mo.

New chapter 🔓 w/ Matteo Radaelli: ‘Language and intelligence in minds and machines’, in Humanities Perspectives on AI (SUP). An introduction for first-year humanities students, built around a few less conventional ideas we think are needed to understand language AI 🧶👇 www.scup.com/doi/10.18261...

Scandinavian University Press

scup.com

’Humanities Perspectives on AI’ started as reading material for our first-year faculty-wide course on AI, now it’s a unique book available to everyone. Chapters on AI history, philosophy, ethics, language, media, art, literature etc. #AI #humanities #NTNU www.universitetsforlaget.no/humanities-p...

Thomas Brandt @thomabr.bsky.social · last mo.

New book out: Humanities Perspectives on AI. An Interdisciplinary Introduction. It brings together media scholars, philosophers, linguists, musicologists, and historians to provide analysis and understanding of AI. Open access here: www.scup.com/doi/book/10....

Humanities Perspectives on AI | Books

scup.com

bioRxivpreprint@biorxivpreprint.bsky.social · 2mo ago

‘An inverse effect of syntactic complexity on syntax- and semantics-related voxels: the harder stimuli were to parse, the more they deactivated semantic-related voxels. This opposing sensitivity suggests a functional trade-off or a competitive allocation of resources within the language network.’ 🔀

Distinct cortical patches for syntactic and semantic composition in the human brain https://www.biorxiv.org/content/10.64898/2026.06.18.732834v1

David Bessis@davidbessis.bsky.social · 4mo ago

Nice essay 👏 Mathematical discovery is another case, beyond language, where form and meaning come apart unnaturally in AI. The viability of AI for math depends on closing the gap with systems that can take AI-generated proofs too complex for humans to parse and present them so they make sense to us.

open.substack.com/pub/davidbes...

Nigel Flower@nigelflower.bsky.social · last mo.

Interesting 👉 Using RPVP, larger portions of the language network now show up in MEG, with intriguing differences within and between temporal and frontal areas in what is computed, when, and in how many stages. Nice to see new/alternative paradigms used to study composition! Looking forward to more.

📢 New paper out in Neurobiology of Language! Using MEG, we found that the brain's responses to full sentences are strikingly similar to those elicited by single words. Our results suggest a "global-to-sequential" process: first a snapshot of structure, then left-to-right composition.

Humanities Perspectives on AI

universitetsforlaget.no

Ethan Mollick@emollick.bsky.social · last mo.

Scale aside, AI is like neuroimaging 25 years ago: if you cried ‘hype!’ you were a denier, not a critic. But some of that criticism improved cognitive neuroscience. That’s why overblown claims and misunderstandings need to be called out early, when a new technology is at the height of its prestige.

It is weird that there is still a substantial set of people who believe "AI is mostly hype" at this stage: Five Eyes is warning about AI, exponential revenue & token use at the AI labs, unit distance/Erdos proofs, and so on... There are many real issues with AI, not being real is not one of them.

Esther Mondragón@e-mondragon.bsky.social · 2mo ago

Don’t let people tell you AI killed the academic essay. We ran a structured essay process in class this year, with successive problem finding, literature search, multiple drafting, and peer review sessions. It worked, and the vibe in class was the best I’ve seen in a long time. Repeat next year.

The 'multiversity' ethic and the spirit of Capitalism, AI version. With apologies to M. Weber for a terrible paraphrasing. "When a student can produce a plausible term paper in twenty minutes using Claude Opus or Google Gemini,... 1/n www.persuasion.community/p/the-multiv...

The language of science (LoS) is often taken to be just ordinary language (OL) extended with new vocabulary, specialized constructions etc. 📑 In this new paper, I argue that the LoS also *restricts* OL, and that some restrictions are cognitively motivated 🪡 www.degruyterbrill.com/document/doi...

degruyterbrill.com

New chapter w/ @jonrbrennan.bsky.social and @andreaeyleen.bsky.social in the Oxford Handbook of Philosophy of Linguistics. Explanation can help us integrate linguistics and neuroscience, if we exploit the diversity of directions and types of explanations available 🧵👇 books.google.no/books?hl=en&...

The Oxford Handbook of Philosophy of Linguistics

This Handbook presents cutting edge articles on the present state-of-the-art in the philosophy of linguistics. It features prominent linguists, philosophers, and cognitive scientists investigating fou...

books.google.no

Jo Wolff@jowolff.bsky.social · 2mo ago

“Science as a practice is not necessarily equally welcoming to everyone. And this can be true even when no practice of discrimination is blatant enough to violate discrimination law, and is enough to make the question of a human right to participate a live one.” @jowolff.bsky.social

Very pleased to see my 'What could a human right to participate in science be?' in Royal Society Open Science (open access) This paper certainly would not exist had it not been for the gentle persuasion and strong support of @michelamassimi.bsky.social royalsocietypublishing.org/rsos/article...

Romain Brette@romainbrette.bsky.social · 2mo ago

Sadly, ‘brains are computers’ has a hidden universal quantifier: ‘all brains do is compute’. But the actual scientific claim is that *some* of what they do is compute. Are there 4, 14 or 140 tasks in which brains compute? This is the empirical research program, not a philosophical disagreement. 🧵👇

Analogies are scientifically useful: one can transfer knowledge between domains. But what is point of insisting that the brain is a computer, if we need to admit that it is completely different from the kind of computer that we are familiar with? What's the big insight?

Philip Resnik: ‘to address LLM bias, start from the hard-core interpretation of the distributional hypothesis, where meaning is distribution, as opposed to distribution being part of what goes into the characterization of meaning and inference’ @complingjournal.bsky.social doi.org/10.1162/coli...

Large Language Models Are Biased Because They Are Large Language Models

Abstract. This position paper’s primary goal is to provoke thoughtful discussion about the relationship between bias and fundamental properties of large language models (LLMs). I do this by seeking to...

doi.org

Simon Fisher @profsimonfisher.bsky.social · 3mo ago

Just read this solid MEG paper in Nature Neuroscience by Zou, Poeppel & Ding: the brain predicts upcoming words less precisely across constituent boundaries than within constituents. A few comments: 🧵👇1/6 www.nature.com/articles/s41...

Constituent-constrained word prediction during language comprehension - Nature Neuroscience

Zou et al. reveal a key difference between human brains and large language models (LLMs). While LLMs are optimized to predict the next word, the human brain modulates prediction efficiency by strategi...

nature.com

I'm collecting together my prior Bluesky threads addressing popular misconceptions about genetics. A thread of threads, beginning with this brief guide to heritability: 🧬🧵🧪

Simon Fisher@profsimonfisher.bsky.social · last yr.

Heritability = one of the most commonly confused concepts in public (mis)understanding of genetics. Let’s break it down. Imagine studying a cohort of people in a specific set of environmental circumstances. You assess everyone for an observable trait & find it varies from one person to another. 🧪1/5

Melanie Mitchell@melaniemitchell.bsky.social · 3mo ago

«Optimization pressures drive convergence—faster iteration within a constrained search space. The result may be more output but less exploration of the unexpected. (...) Optimizing each component does not guarantee the integrity of the whole. Optimization may, in fact, undermine it.» 💭🎯

"Science is not simply a sequence of tasks that can be optimized....The scientific system thrives on inefficiency: redundant efforts, failed attempts, and divergent paths. These are not costs to be eliminated but sources of discovery." Excellent essay! ⬇️ www.science.org/doi/10.1126/...

Alison Gopnik @alisongopnik.bsky.social · 3mo ago

Interesting new mathematical paper in PNAS by Jarvis, Klein, Rosman & Saxe: ‘Compositionality and systematicity emerge from iterated learning in deep linear networks’. A few thoughts. www.pnas.org/doi/10.1073/...

Compositionality and systematicity emerge from iterated learning in deep linear networks | PNAS

Humans have a remarkable ability to systematically generalize—reasoning about new situations by combining aspects of previous experiences. Language...

pnas.org

Here is the specific link to our paper with Eunice Yiu, Shiry Ginosar and Kelsey Allen, how to construct causal models through intrinsically motivated action, something kids do and LLMs don't. The whole issue on world models is very much worth reading. royalsocietypublishing.org/rsta/article...

Empowerment gain and causal model construction: children and adults are sensitive to controllability and variability in their causal interventions

Abstract. Learning about the causal structure of the world is a fundamental problem for human cognition. Causal models and especially causal learning have

royalsocietypublishing.org

MIT Press@mitpress.bsky.social · 4mo ago

Here’s what caught my attention in Grodzinsky’s new book, and why, if we want to understand and explain language in the mind/brain, NLP/LM research and generative grammar may be equally unviable, either alone or combined.👇 1/9

In "How Deeply Human Is Language?,” leading neurolinguist Yosef Grodzinsky explains linguistic theory and large language models—the top contenders for understanding human language—and evaluates them in the context of the brain: mitpress.mit.edu/978026205200...

This Nature article compares AI bills to postdoc salaries. It is legitimate to ask whether AI is worth what it costs, but running that calculation relative to the costs of training/hiring researchers is a choice I find problematic. 1/5 www.nature.com/articles/d41...

AI bills can be as big as a postdoc salary. Is the cost worth it?

Recent price hikes, usage limitations and unreliable outputs are causing some scientific researchers to think twice about using artificial intelligence.

nature.com

M.J. Crockett@mjcrockett.bsky.social · 3mo ago

“We are witnessing a panicked gold rush to automate scientific work (...) Science is a social practice that depends on cooperation and trust (...) We risk all of science if we rush to build AI scientists before we understand the value of human science.” 🎯 Congratulations, @mjcrockett.bsky.social 👏

Last weekend I received the Troland Research Award from the @nationalacademies.org. I’m so grateful to the communities who made this work possible. It's strange to receive this award at a time when much of the work being recognized is not eligible for federal funding. My remarks 👇 and a 🧵>>