Sign in

Arnav Das

@arnav7.bsky.social
2K followers 13K following 1.2K posts

My posts here will be a mishmash of personal opinions and smol curious excursions..

PostsRepliesMedia
Reposted by Arnav Das
William Umboh @swumboh.bsky.social · 6h
Training a Large Language Meowdel by osmosis
Cat sitting on textbook
0101
Reposted by Arnav Das
Aaron Rupar @atrupar.com · 6h
I for one think it’s bad that the president can just say crazy stuff all day every day and 99% of it doesn’t make the news because we’ve normalized having an insane president
678181784264
Reposted by Arnav Das
sara @slb.sh · 8h
just discovered a delightful website where you pick the venue and it shows crowdsourced photos of views from different seats aviewfrommyseat.com/photo/445964...
aviewfrommyseat.com
Carnegie Hall, section Dress Circle C, row AA, seat 1 - Víkingur Ólafsson
Great less expensive seats for a piano recital. Unobstructed view
051
Reposted by Arnav Das
Grace @gracekind.net · 7h
Full name reveal
arxiv.org
GRACE: Reinforcement Learning for Grounded Response and Abstention under Contextual Evidence
Retrieval-Augmented Generation (RAG) integrates external knowledge to enhance Large Language Models (LLMs), yet systems remain susceptible to two critical flaws: providing correct answers without expl...
41863
Reposted by Arnav Das
Michael "Shapes Dude" Betancourt @betanalpha.bsky.social · 8h
The product spaces chapter has been reorganized and expanded. This includes a discussion of the rectangles and cylinders and new, cleaner notation and terminology for product spaces and their cross sections. HTML: betanalpha.github.io/assets/chapt... PDF: betanalpha.github.io/assets/chapt...
betanalpha.github.io
Product Spaces
111
Reposted by Arnav Das
Michael "Shapes Dude" Betancourt @betanalpha.bsky.social · 8h
I have finally finished a new probability theory chapter. In fact, two of them. Two and a half if you consider the substantial updates to my product spaces chapter.
171
Reposted by Arnav Das
koenfucius @koenfucius.bsky.social · 8h
The 69-year old guy who memorized pi to 111,000 digits forgot his wife’s birthday. The biggest misconception about memory is that it should be perfect, says @lisagenova.bsky.social—everyone forgets. All you wanted to know about your memory and how to protect it: buff.ly/EhnfSZx
011
Reposted by Arnav Das
Marco @mcognetta.bsky.social · 12h
🚨 [Token][ization] Paper Alert 🚨 Tokenization is a wildly understudied area of language modeling despite it having effects across all of NLP. Over the past ~8 months, 32 (!) tokenizer researchers put together the most comprehensive survey of the field. Check it out!
110128
Reposted by Arnav Das
Earl K. Miller @earlkmiller.bsky.social · 12h
Surprisingly Complex Waves Reveal the Brain’s Inner Workings www.quantamagazine.org/surprisingly... #neuroscience
quantamagazine.org
Surprisingly Complex Waves Reveal the Brain’s Inner Workings | Quanta Magazine
Unexpected patterns traveling across the human brain may be reorganizing its activity in real time.
1158
Reposted by Arnav Das
koenfucius @koenfucius.bsky.social · 11h
We are all nodes in a Dunbar network—but not all nodes are the same. Most of us focus on the close ties; superspreaders keep a balance between strong and weak connections. Such diversity is key for effective communication, writes Gerald Ashley: buff.ly/w0i1q7B
011
Reposted by Arnav Das
koenfucius @koenfucius.bsky.social · 10h
Sleep out of sync with your circadian clock? Catch some 🌞 early in the day. Researchers in Brazil found adults who spent more time in the morning sun had a more aligned internal biological clock, which in turn can promote healthier sleep patterns: buff.ly/152iFDx
021
Reposted by Arnav Das
Experimental Philosophy @xphilosopher.bsky.social · 15h
People tend to attribute knowledge in what philosophers call “fake barn” cases These cases don’t satisfy the “safety” condition — meaning that people are happy to attribute knowledge even when this condition is not satisfied
link.springer.com
Knowledge and luck - Psychonomic Bulletin & Review
Nearly all success is due to some mix of ability and luck. But some successes we attribute to the agent’s ability, whereas others we attribute to luck. To better understand the criteria distinguishing...
181
Reposted by Arnav Das
Experimental Philosophy @xphilosopher.bsky.social · 15h
What the studies keep showing, again and again, is that ordinary folks are happy to attribute knowledge in cases where philosophers would deny knowledge The ordinary concept *does not* seem to involve all of the complex conditions that philosophers have invoked in their theories
1101
Reposted by Arnav Das
Experimental Philosophy @xphilosopher.bsky.social · 15h
There has been a huge amount of experimental philosophy research on the concept of KNOWLEDGE This research finds that the ordinary concept doesn’t work at all the way philosophers said it does A thread with some examples: 🧵
34013
Reposted by Arnav Das
Jay Van Bavel, PhD @jayvanbavel.bsky.social · 11h
Words change behavior more than we think People cooperate more when a game is called the “Community Game” than when it is called the “Wall Street Game” Language changes the emotions & social norms we associate with it—and therefore what we choose to do. www.powerofusnewsletter.com/p/shaping-be...
173
Reposted by Arnav Das
Denise J. Cai, Ph.D. @denisejcai.bsky.social · 12h
NEW: The anterior hypothalamic nucleus acts like a volume knob for stress, scaling the brain’s response to threatening events 🧠 Read the Cai Lab's latest paper, led by @zachtpennington.bsky.social, in @nature.com 🎉 www.nature.com/articles/s41... More: bsky.app/profile/did:...
26427
Reposted by Arnav Das
A. Feder Cooper @afedercooper.bsky.social · 11h
We make very carefully scoped claims, supported by an extensive set of baselines on non-training data. Without baselines, it's not possible to rigorously separate a model's ability to generate fluent text from extraction of training data. (We see no false positives for non-training books.)
We can measure extraction coverage: the percentage of a book that is memorized in the model by seeing how much text is extractable (with respect to 50-token suffixes) out of the whole book. When we run extraction with larger prompts (longer prefixes of the 50-token suffix), this exposes more underlying memorization (the discoverability phenomenon, see Carlini et al. 2023). Here, the plot shows that extraction coverage goes up (to a point) as prefix length increases for books in the training data. For the shown control book (which is non-training data), there is no observed extraction at all prefix lengths. This supports the validity of our measurement procedure.
133
Reposted by Arnav Das
A. Feder Cooper @afedercooper.bsky.social · 11h
Our large-scale study of memorization of books in open-weight LLMs (e.g., Llama, Qwen) will appear at the 2026 Conference on Language Modeling as an oral. We made a website for exploring our results on 200 books and 14 models: books-memorization.github.io
books-memorization.github.io
How much do open-weight LLMs memorize specific books?
Open-weight LLMs memorize books far more than previously believed. Memorization varies by model family, model size, and book. In extreme cases, entire books are memorized, and we can generate them eff...
25330
Reposted by Arnav Das
Justin Weinberg @dailynous.com · 16h
Choosing not to sleep. Philosophy of paleontology. How AI's represent. Sixty years of philosophy in America. The pathologizing tendency. Undergrads teaching undergrads philosophy. AI, writing, thinking, & learning...
dailynous.com
Mini-Heap - Daily Nous
Links of to interesting stuff elsewhere… Mini-Heap posts usually appear when several new items accumulate in the Heap of Links, a collection of items from around […]
031
Reposted by Arnav Das
Neuroskeptic @neuroskeptic.bsky.social · 15h
The word 'empathy' is a broad umbrella term that should be used "more sparingly" pubmed.ncbi.nlm.nih.gov/42789468/ Yes
074
Reposted by Arnav Das
Michael "Shapes Dude" Betancourt @betanalpha.bsky.social · 30/09/2026
It came from the deep.
An ensemble of red, squiggly lines diverging from a common point on the bottom left corner of the image.
073
Reposted by Arnav Das
Cat Hicks @grimalkina.bsky.social · 30/09/2026
Nobody past the age of twenty should be obsessed with what people majored in in college and see it as the single defining choice that entirely shapes someone, deterministically, for all time
713715
Reposted by Arnav Das
koenfucius @koenfucius.bsky.social · 23h
Higher IQ is correlated with success, but that is often confounded by incentives at school or work (intelligent people might also be working harder). New research suggests high-IQ people are successful also when incentives don’t matter, writes @ewdolan: buff.ly/Xp2jluD
011
Reposted by Arnav Das
Edouard Machery @edouardmachery.bsky.social · 30/09/2026
you must read this by @katemanne.bsky.social. Surely, it is no accident that Cornell was so eager to make a deal with trump and that its appalling president ran over a propalestinian student with his car. There is something rotten at @cornelluniversity.bsky.social.
thecut.com
What It Feels Like at Cornell Right Now
As a professor at the Ivy League school, Kate Manne has known many, many victims. There’s a palpable feeling on campus, she writes, that this is big.
273
Reposted by Arnav Das
Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 30/09/2026
Preparing a lecture on the practice of research and was reminded of this lovely essay on the importance of being comfortable with feeling stupid: fermatslibrary.com/s/the-import...
fermatslibrary.com
Fermat's Library | The importance of stupidity in scientific research annotated/explained version.
Fermat's Library is a platform for illuminating academic papers.
3234
Reposted by Arnav Das
Daniel Drucker @danieldrucker.bsky.social · 29/09/2026
@kmahowald.bsky.social and I wrote this piece on expressives in LLM agents' chains of thought: www.theatlantic.com/technology/2... We look at some surprising, emotionally expressive language that appears in AIs' "thinking" when they're doing math, or hacking. We argue it's likely functional!
theatlantic.com
2224
Reposted by Arnav Das
The Ramsey Lab @theramseylab.bsky.social · 29/09/2026
Are you interested in the past, present & future of Artificial Life (ALife) as a scientific field? 🚨 We are hosting an interdisciplinary minisymposium this December, so send us your work & come along! Please share widely! 🔗 hiw.kuleuven.be/clps/events/... #philsci #academicsky #evobio #AI #HPS
Promotional graphic for a mini-symposium titled “Life As It Could Be: Interdisciplinary Perspectives on Artificial Life,” taking place 7–8 December 2026 at Aula Symbiosis, BioScience Center, KU Leuven. The text appears over a blue-and-orange honeycomb-like cellular pattern.
13320
Reposted by Arnav Das
Shen-yi Liao 廖顯禕 @liao.shen-yi.org · 29/09/2026
@dailynous.com in the side links there is a post about whether you'd take a pill to forego sleep, and there's a philosophy paper on this exact question! doi.org/10.1017/apa....
doi.org
The Value of Sleeping | Journal of the American Philosophical Association | Cambridge Core
The Value of Sleeping - Volume 11 Issue 2
051
Reposted by Arnav Das
David Spurrett @doctorspurt.bsky.social · 29/09/2026
Sometimes The Onion really hit one right out of the park.
2116
Reposted by Arnav Das
Olivia Guest · Ολίβια Γκεστ @olivia.science · 29/09/2026
> Capitalist elites hate workers and despise worker expertise. [I] have decades of experience, sitting in rooms with (mostly) men who [...] hated that a wage servant possessed technical expertise they needed, but were too lazy, arrogant or stupid to grasp. — @cloudquistador.eurosky.social
monroelab.com
The Tech Industry: An Autopsy
The tech Industry, and the fields that grew from it, are dead.  The Industry, which, as Yasha Levine details in his book, ‘Surveillance Valley’, was always a tool of oppression, and never &#82…
59037
Reposted by Arnav Das
koenfucius @koenfucius.bsky.social · 28/09/2026
Insomnia is not necessarily about environmental or behavioural influences. What if you looked at it through the lens of the autonomous nervous system? 4 tips to regulate your autonomous nervous system to sleep better: buff.ly/AVB59L2
011
Reposted by Arnav Das
Earl K. Miller @earlkmiller.bsky.social · 28/09/2026
It's a struggle to figure out the math behind chaotic systems, especially when the data is limited and noisy. PEM-UDE can recover interpretable equations from messy, chaotic data, and it yields testable predictions about neural networks. arxiv.org/abs/2507.03631 #neuroscience
arxiv.org
Scientific Machine Learning of Chaotic Systems Learns Reduced-Order Equations for Neural Populations
Extracting interpretable mathematical models from complex dynamical systems is difficult, especially for chaotic dynamics observed with noisy experimental data. We present PEM-UDE, a method that combi...
0162
Reposted by Arnav Das
Maria Antoniak @mariaa.bsky.social · 28/09/2026
Interesting new venue for work on culture and AI, first set of essays already available: www.cambridge.org/core/journals/cam…
cambridge.org
Cambridge Forum on AI: Culture and Society | Cambridge Core
Cambridge Forum on AI: Culture and Society - Tobias Blanke, Georgina Born OBE FBA, Beth Coleman
092
Reposted by Arnav Das
Kevin Mitchell @wiringthebrain.bsky.social · 29/09/2026
It's done! 😊
1415514
Reposted by Arnav Das
Edouard Machery @edouardmachery.bsky.social · 29/09/2026
worth reading, although no, the scale is not what makes the 74,000 murders a genocide. What makes it a genocide is the repeatedly expressed intent to destroy a people by the politicians in the israeli government, many in the media, and armi leaders, with the support of many israeli.
nytimes.com
Opinion | There Is a Word So Powerful It Need Be Spoken Only Once (Gift Article)
The film that makes a powerful statement without ever raising its voice.
062
Reposted by Arnav Das
Edouard Machery @edouardmachery.bsky.social · 29/09/2026
his book is open access - download it! #philsci.
global.oup.com
global.oup.com
122
Reposted by Arnav Das
Laurie Loves Data @laurelann.bsky.social · 29/09/2026
"I found out that nobody cared." - Trump talking to a NYT reporter about his many lucrative business deals since becoming president.
nytimes.com
Does Trump Worry About Conflicts of Interest? ‘I Found Out That Nobody Cared.’
American presidents have generally tried to avoid appearing to profit from the office. President Trump has chosen a different path.
094
Reposted by Arnav Das
Justin Weinberg @dailynous.com · 29/09/2026
"The story about the analytic turn is more complicated than the McCarthyist thesis suggests..."
dailynous.com
Analytic Philosophy and McCarthyism: A New Look (guest post) - Daily Nous
To what extent is the “the story about the analytic turn” in philosophy a story about politics? There’s been an increased discussion in recent years […]
0126
Reposted by Arnav Das
Sheina Lew-Levy @sheinalew.bsky.social · 29/09/2026
🚨 Our new lab paper, led by @amandinevisine.bsky.social, examines exploration, group composition, and play activities among BaYaka children and teens living in a village context! Amandine is currently on fieldwork following up on this for her PhD 🎯 www.sciencedirect.com/science/arti...
sciencedirect.com
Congolese BaYaka children's learning and exploring in a village context
Childhood is a key period for acquiring and transmitting cultural knowledge and skill. Hunter-gatherer children, such as the Congolese BaYaka, learn a…
14417
Reposted by Arnav Das
koenfucius @koenfucius.bsky.social · 29/09/2026
It may not be easy to describe what an intelligent person would look like, but we’d probably not consider someone talking loudly to themselves, or someone who uses swear words as particularly intelligent. We might be wrong: buff.ly/DCjlcZY
021
Reposted by Arnav Das
Lionel @spiindoctor.bsky.social · 29/09/2026
Evaluating time series foundation model claims robjhyndman.com/hyndsight/fo... similar issues for me with respect to tabular foundation models (compared to production decision tree models): Performance improvement at scale, which relies on scaling laws, is not there.
121
Reposted by Arnav Das
Ben Recht @beenwrekt.bsky.social · 29/09/2026
The illusion and politics of objectivity in cost-benefit analysis, a decision-making institution so ingrained that Americans can’t imagine life without it.
argmin.net
Pricing Commensurability
On the origins of cost-benefit analyses in governmental decision making
2259
Reposted by Arnav Das
Shubhendu Trivedi @shubhendu.bsky.social · 29/09/2026
Godfather of AI, Godmother of AI, Luca Brasi of AI, Commission of AI, Crime Familes of AI
192
Reposted by Arnav Das
Daryl Fairweather ⛅ @fairweatherphd.bsky.social · 29/09/2026
We should really put a pigovian tax on AI, not just for the environmental harm, but for the slop. Slop is pollution. The internet has been polluted with slop.
0192
Reposted by Arnav Das
koenfucius @koenfucius.bsky.social · 29/09/2026
Is business ethics an oxymoron? It doesn’t have to be, writes in-house ethicist John Paul Rollert, surveying how it has evolved over the last half century, and exploring how to prevent it from becoming one: buff.ly/NwHBSPL
011
Reposted by Arnav Das
Justin Weinberg @dailynous.com · 29/09/2026
For the first time in a while, a philosophy professor has been awarded one of the MacArthur genius grants...
dailynous.com
A Philosopher Among the MacArthur Fellows - Daily Nous
Felipe De Brigard, professor of philosophy, psychology, and neuroscience at Duke University, principal investigator at the Imagination and Modal Cognition Lab, and core faculty of […]
45613
Reposted by Arnav Das
Esther Mondragón @e-mondragon.bsky.social · 26/09/2026
Have we normalised this shit so much that nobody is reacting to it? What kind of society are we in, ffs!
012
Reposted by Arnav Das
Esther Mondragón @e-mondragon.bsky.social · 29/09/2026
Dear academic pals, perhaps we could stop talking about freaking AI for a bit and focus a little on our fellow women. This also matters
1154
Reposted by Arnav Das
Andrea Lathrop @cabernet.bsky.social · 29/09/2026
I promise, I'm really not live-tweeting this, but popping in again to state that I think CoT monitoring is highly overrated, simply because it has been widely shown to be divorced from the conclusion, and in some cases to be backfilled from the conclusion.
141