What the smart people are recommending. 7867 books, tools, and products endorsed by the thinkers absorb.md tracks. Ranked by how many times each has been recommended across compiled podcasts, papers, posts, and tweets.
βIn this work, we present a surprising connection between attention sinks and compression valleys, tracing both to the formation of massive activations in the residual stream. We prove theoretically thβ¦β
βPaper: https://arxiv.org/abs/2508.13557β
βThus far, LLM-JEPA is able to outperform the standard LLM training objectives by a significant margin across models, all while being robust to overfiting.β
βCritically analyzing recent STM and transport experiments [Z. Ge, et al, arXiv:2510.12009] on 2D electron systems in the presence of random quenched impurities, we argue that the resulting low-densityβ¦β
βA recent STM experiment in 2D bilayer graphene [Y.-C. Tsui, et al., Nature 628, 287 (2024)], under a strong perpendicular magnetic field, has made a direct observation of the existence of three distinβ¦β
βA recent STM experiment in 2D bilayer graphene [Y.-C. Tsui, et al., Nature 628, 287 (2024)], under a strong perpendicular magnetic field, has made a direct observation of the existence of three distinβ¦β
βWe argue that the spatially random localized phase at low filling is the recently proposed disorder-dominated strongly localized amorphous "Anderson solid" phase [A. Babber, et al., arXiv:2601.03521],β¦β
βQuantum-Classical Separation in Bounded-Resource Tasks Arising from Measurement Contextualityβ
βWe argue that the spatially random localized phase at low filling is the recently proposed disorder-dominated strongly localized amorphous "Anderson solid" phase [A. Babber, et al., arXiv:2601.03521],β¦β
βLet me just say on on the job loss question because I think this is really important that Yale just released a study and it showed that in the 33 months after the launch of Chat GPT there was no disceβ¦β
βThere was an article in the Wall Street Journal just last week talking about the uh construction boom that's happening that's benefiting construction workers like electricians like plumbers like workeβ¦β
βDominik Hangleiter weighs in with an informative post about a much debated question: Has quantum advantage been achieved? This is the first post in a three-part series.β
βWe, therefore, introduce DexWM, a Dexterous Interaction World Model that predicts future latent states of the environment conditioned on past states and dexterous actions.β
βThis paper introduces SpidR-Adapt for rapid adaptation to new languages using minimal unlabeled data.β
βI like the Rolling Stone obituary for Kishinger.β
β# Fail Fast, Win Big: Rethinking the Drafting Strategy in Speculative Decoding via Diffusion LLMsβ
β# From Generated Human Videos to Physically Plausible Robot Trajectoriesβ
βChapter 4 in my dissertation from 1986 is "From Crowd to Public." https://pressthink.org/j/rosen-archive/dissertation/reader/#from-crowd-to-publicβ
βThe Claude Mythos Preview system card is available here: https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdfβ
βRemembrall: Leaning into Memory for Accurate Video Analytics on System-on-Chip GPUsβ
βHere, we introduce Spatiotemporal Pyramid Flows (SPF), a new class of flow matching approaches that model data hierarchically across spatial and temporal scales.β
βAtherosclerosis is not the only possible cause of a heart attack (myocardial infarction), and women are more prone to alternative causes than men. A closer look at the data tells a different story. Fuβ¦β
β# A Lightweight Library for Energy-Based Joint-Embedding Predictive Architecturesβ
βDominik Hangleiter continues his thoughtful assessment of the evidence for quantum advantage in random-circuit-sampling experiments. http://quantumfrontiers.com/2026/01/25/has-quantum-advantage-been-aβ¦β
βWe propose Radial-VCReg, which augments VCReg with a radial Gaussianization loss that aligns feature norms with the Chi distribution-a defining property of high-dimensional Gaussians.β
βScaling Text-to-Image Diffusion Transformers with Representation Autoencodersβ
βThe always eloquent Nadya Mason on the promise of Quantum 2.0 and how to build the quantum workforce.β
βI think I talked about it last time, but I wrote a doc called "The Big Blob of Compute Hypothesis".β
βYou mentioned in your talk at scaleup AI today your recent cybernetic teammate paper uh involving a field experiment with 776 professionals at Proctor and Gamble and it demonstrates that AI often funcβ¦β
βOur work establishes IndicFairFace as the first benchmark to study geographical bias in VLMs for the Indian context.β
βA study linking statins, the microbiome, and blood sugar control doesnβt change clinical practice. Full article linked below.β
βEveryone should go read the the Clawed Constitution. Uh argue with it, grapple with it. I found it a very challenging and also a very moving uh read.β
βThe best way of starting with this is to write this article and we made a video about this as well which has had like it's hundreds of thousands of views. It's pretty incredible. Death of a Salesforceβ¦β
βor you know death taxes and AI. We've done we've covered the gamut on everything around taxes.β
β# stable-worldmodel-v1: Reproducible World Modeling Research and Evaluationβ
βIn order to help the public better understand the developments pertaining to AI and mathematics, we suggest quantifying standard levels of autonomy and novelty of AI-assisted results, as well as propoβ¦β
βYou mentioned in your talk at scaleup AI today your recent cybernetic teammate paper uh involving a field experiment with 776 professionals at Proctor and Gamble and it demonstrates that AI often funcβ¦β
β# Causal-JEPA: Learning World Models through Object-Level Latent Interventionsβ
β# Heaps do lie: debugging a memory leak in vLLM. _Blog post by Mistral AI_β
βCan Private Markets Normalize?_Blog post by Ted Seides_β
βDominik Hangleiter highlights the importance of efficient verification of quantum advantage, and assesses the prospects for achieving it soon.β
βWe thank John Preskill for suggesting the phrase βOne Sum To Rule Them All,β which inspired the title of this paper.β
βThe βJVG algorithmβ is crapβ
βCan coding agents relicense open source through a βclean roomβ implementation of code?β
βAnthropicβs Responsible Scaling Policy: Version 3.0β
βIn this paper, we address what's wrong with our conception of AGI, and why, even in its most coherent formulation, it is a flawed concept to describe the future of AI. We explore whether the most wideβ¦β
βBeyond Language Modeling: An Exploration of Multimodal Pretraining _ArXiv paper co-authored by Yann LeCun_β
βThe International AI Safety Report 2026 synthesises the current scientific evidence on the capabilities, emerging risks, and safety of general-purpose AI systems.β
βTo that end, we introduce the Geodesic Hypothesis, positing that token sequences trace geodesics on a smooth semantic manifold and are therefore locally linear. Building on this principle, we propose β¦β
βMotivated by the recent twisted MoTe$_2$ experiment [arXiv:2601.18508], we develop a disordered interacting edge theory of a fractional topological insulator at $Ξ½_{\text{tot}}=4/3$, consisting of twoβ¦β
βStatement from Dario Amodei on our discussions with the Department of Warβ
βThis vision paper lays out a 10-year roadmap for AI+HW co-design and co-development, spanning algorithms, architectures, systems, and sustainability. We articulate key insights that redefine scaling aβ¦β
βI am upset by WIRED cover story about the gay mafia supposedly running Silicon Valley.β
βRails testing on autopilot: Building an agent that writes what developers won'tβ
βOur Investment in Accrual _Blog post by General Catalyst_β
βWe study two recurring phenomena in Transformer language models: massive activations, in which a small number of tokens exhibit extreme outliers in a few channels, and attention sinks, in which certaiβ¦β
βThis high-performance QFC connecting ultraviolet and telecom bands satisfies the stringent requirements for long-lived remote ion-ion entanglement in scalable quantum networks [W.-Z. Liu et al., Naturβ¦β
β# FlashAttention-4: Algorithm and Kernel Pipelining Co-Design for Asymmetric Hardware Scalingβ
βhttps://anthropic.com/news/mozilla-firefox-securityβ
βOur work establishes Rydberg tweezer arrays as a powerful platform for spectroscopic studies of strongly correlated models, offering microscopic control and direct real-space access to emergent quasipβ¦β
βIn this thoughtful essay, Garnet Chan reflects on recent progress using classical heuristics in computational quantum chemistryβand what it means for quantum computing. The lessons he draws can help tβ¦β
βhttps://vitalik.eth.limo/general/2020/11/06/pos2020.htmlβ
β_ArXiv paper co-authored by Yann LeCun_β
βyou know what e reminds me of Kum have you read John Perry Bar's cyberspace essay from back in the day uh like in the 90s way before your talk probably okay but you have basine um do do you remember iβ¦β
βA new fast confirmation rule mechanism lets you get a hard guarantee that Ethereum will not revert after one slot (12 seconds)β
βMachine learning approaches to spatiotemporal physical systems have primarily focused on next-frame prediction, with the goal of learning an accurate emulator for the system's evolution in time.β
βI'm thrilled that Charlie Bennett and Gilles Brassard have received the A. M. Turing Award for βtheir essential role in establishing the foundations of quantum information science and transforming secβ¦β
βWe present V-JEPA 2.1, a family of self-supervised models that learn dense, high-quality visual representations for both images and videos while retaining strong global scene understanding.β
βMamba-3β
βYou know, one of the stats that people talk about a lot is this METR report which basically says for um each different model that comes out roughly how much human work can it do uh in an automated fasβ¦β
βMore here, including SWE-bench score of 93.9% (!) and a new model behavior known as "answer-thrashing" https://www-cdn.anthropic.com/53566bf5440a10affd749724787c8913a2ae0841.pdfβ
βYou can read a detailed technical report on the software vulnerabilities and exploits discovered by Claude Mythos Preview here: https://red.anthropic.com/2026/mythos-previewβ
βhttps://garryslist.org/posts/sfo-is-trying-to-bury-waymo-4395d272β
βIn this work, we introduce LeWorldModel (LeWM), the first JEPA that trains stably end-to-end from raw pixels using only two loss terms: a next-embedding prediction loss and a regularizer enforcing Gauβ¦β
β@nicoleyh11 offers guidance on handling faculty offers. http://quantumfrontiers.com/2026/03/11/nicoles-guide-to-navigating-faculty-position-offers/β
βBuilding superconducting and neutral atom quantum computersβ
βWeβre sharing new research into how AI might be misused to exploit emotions or manipulate people into making harmful choices.β
βRead more: https://www.anthropic.com/engineering/harness-design-long-running-appsβ
βhttps://garryslist.org/posts/richmond-just-voted-to-reinstate-their-flock-cameras-after-crime-spikedβ
βThis is a very good postβ
βBuilding on earlier work (Meirom, Mor, Weinstein Arxiv 2505.12441), we utilize a benchmarking strategy applicable for comparing these two architectures by measuring "quantumness" directly on optimal sβ¦β
βI'm giving his post to my agent and will ask it to help me talk to it better.β
βRead the full paper: https://transformer-circuits.pub/2026/emotions/index.htmlβ
βIn this paper, we present the AnIML Ontology, an OWL 2 ontology that formalises the semantics of AnIML and aligns it with the Allotrope Data Format to support future cross-system and cross-lab interopβ¦β
βTo read more about standalone functions, check [FUNCTIONS.md](./FUNCTIONS.md).β
βNew report from us: Can you prompt inject your way to an βAβ? ... We tested whether it works. It does on older & smaller models, but not on most frontier AI: https://gail.wharton.upenn.edu/research-anβ¦β
βBuilding and Running Rockefeller Capital Managementβ
βAnd how resilient are you? https://www.mercatus.org/resilient-societies#230548828-4153906252β
βThat's a study by Grid Lab that looks at if Wisconsin hypers scale data centers run on renewables versus gas, what are the cost impacts to all customers?β
βSee Mr. Chatterbox is a (weak) Victorian-era ethically trained model you can run on your own computer for background on this project.β
βRead the paper on arXiv: https://arxiv.org/abs/2603.12230β
β_Blog post by Hartmut Neven_ https://research.google/blog/safeguarding-cryptocurrency-by-disclosing-quantum-vulnerabilities-responsiblyβ
β# Wherefore Art Thou? Provenance-Guided Automatic Online Debugging with Lumosβ
βMy piece in the Economist where I argue against de-weirding AI. It is a strange technology with both risks & opportunities that need to be discovered. Pretending AI works like normal IT automation canβ¦β
βhttps://marginalrevolution.com/marginalrevolution/2026/04/ev-arts-patronage-tranche.html and a hearty thanks to @patrickc for enabling this.β
βI hadn't yet come across Trip's own writeup of the project, which provides way more detail about how he trained the model including SFT (supervised fine tuning) against synthetic chat examples createdβ¦β
βThis paper introduces IDEA2, a novel, semi-automated workflow that integrates Large Language Models (LLMs) within a collaborative, expert-in-the-loop process to address this challenge.β
βThe Hermes Deep Dive. (The new hot AI agent harness).β
βBut you know there are actually demos of of what's called Bzon sampling which was a proposal for using optical uh Ψ§Ω Ψ§Ω uh quantum computing to just demonstrate quantum advantage on a contrived benchmβ¦β
βHere's Anthropic's Mythos system card. The model demonstrated a striking leap in cyber capabilities relative to prior models.β