Papers

31 papers · 112 korrents drawn from them

What a paper is doing in a belief archive.

A paper is the strongest source this archive has: a document its authors signed their names to, deliberately, with a date and an address a stranger can check. It is also the hardest to read honestly, because most sentences in a paper report a measurement rather than state a position, and a hedged finding rewritten as a belief would put a claim in a real person's mouth that the paper does not make. So these are the paper's arguments — what it says should be thought, from the abstract, introduction and discussion — quoted word for word and checked against a stored copy, and curated rather than exhaustive. Where a paper has several authors this archive covers, they hold the claim jointly: no one of them is promoted to its author.

Filter31 papers, most cited first

Attention is all you need Aidan N. Gomez, Ashish Vaswani, Illia Polosukhin, Jakob Uszkoreit, Llion Jones, Niki Parmar, Noam Shazeer, Łukasz Kaiser · 3 korrents · held by 8 of its authors · 12 Jun 2017 A new network for translating languages uses only attention to link words, dropping the usual step-by-step loops and filters entirely, and it works better while training faster.
transformers
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding Jacob Devlin, Kenton Lee, Kristina Toutanova, Ming-Wei Chang · 4 korrents · held by 4 of its authors · 11 Oct 2018 Train a language model to guess randomly hidden words using the full surrounding text so it sees both directions, then lightly adjust it for many different tasks.
LLMs scaling laws
Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Jonathan Frankle, Michael Carbin · 5 korrents · held by 2 of its authors · 9 Mar 2018 Big networks started at random hide smaller ones inside that, kept at those starting weights, train alone to the same accuracy just as fast.
neural networks pruning
Barabasi & Albert, "Emergence of scaling in random networks" (arXiv) Albert-László Barabási, Réka Albert · 4 korrents · held by 2 of its authors · 21 Oct 1999 Networks keep growing by adding sites that link more often to the already well-linked ones, producing the same uneven connection pattern seen across many real systems.
network science
He, Zhang, Ren & Sun, "Deep Residual Learning for Image Recognition" (arXiv) Jian Sun, Kaiming He · 4 korrents · held by 2 of its authors · 10 Dec 2015 Making neural nets much deeper for image recognition works better if layers learn only the difference from their inputs instead of full new mappings, letting accuracy rise with depth.
neural networks computer vision
Kingma & Ba, "Adam: A Method for Stochastic Optimization" (arXiv) Diederik P. Kingma, Jimmy Ba · 4 korrents · held by 2 of its authors · 22 Dec 2014 A new algorithm called Adam finds good solutions to huge noisy problems by tracking running averages of gradients and their sizes so steps adjust themselves with almost no tuning.
optimizers
Acemoglu & Restrepo, "Robots and Jobs: Evidence from US Labor Markets" (NBER working paper 23285) Daron Acemoglu, Pascual Restrepo · 3 korrents · held by 2 of its authors · 1 Mar 2017 Adding industrial robots from 1990 to 2007 cut local US jobs and wages—one more robot per thousand workers lowered employment by 0.18-0.34 points and pay by 0.25-0.5 percent.
robotics
Ioannidis, "Why Most Published Research Findings Are False" (PLoS Medicine) John Ioannidis · 6 korrents · 30 Aug 2005 Most published research findings are false because studies are often too small, effects too weak, methods too flexible, and too many teams race for positive results.
metascience
Assessing Adaptive World Models in Machines with Novel Games (with 13 co-authors) François Chollet · 5 korrents · 17 Jul 2025 People adapt fast by quickly building mental models of how new places work, yet AI checks only look at fixed knowledge from big training sets.
world models
Mathematics in the age of AI Terence Tao · 5 korrents · 17 Aug 2026 If AI can do research math, the community must make explicit what it really values, because chasing solutions alone will no longer serve the rest of the enterprise.
mathematics
On the Measure of Intelligence François Chollet · 5 korrents · 5 Nov 2019 Being smart is not scoring high on a fixed task after lots of built-in knowledge or training; it is how efficiently you acquire skills on new problems.
ARC-AGI measuring intelligence
The legacy of Bristol: public disclosure of individual surgeons' results (with 5 co-authors) David Spiegelhalter · 5 korrents · 19 Aug 2004 After Bristol, heart surgeons set out to publish each one's raw death rates inside very wide safety bands to show they are safe, not to help patients pick a surgeon.
surgical outcomes
Chaos and Misallocation under Price Controls (with 2 co-authors) Alex Tabarrok · 4 korrents · 12 Feb 2026 When prices are frozen below market levels, sellers become indifferent about where to ship, so tiny cost differences send all supply to some markets and starve others completely.
benchmarks price gouging
From Solvers to Research: Large Language Model-Driven Formal Mathematics at the Research Frontier (with 18 co-authors) Terence Tao · 4 korrents · 8 Jul 2026 AI systems that check math proofs with computers are great at fixed problems but still cannot discover new theorems or settle open questions the way research mathematicians do.
LLMs mathematics
Gradient-based learning applied to document recognition (with 3 co-authors) Yoshua Bengio · 4 korrents · 1 Jan 1998 Neural nets built for how 2D shapes vary beat every other method at reading handwritten digits, and whole multi-part document readers can be trained end to end the same way.
neural networks
Highly accurate protein structure prediction with AlphaFold (with 33 co-authors) Demis Hassabis · 4 korrents · 15 Jul 2021 A computer program regularly predicts a protein's three-dimensional shape from its amino acid sequence alone at accuracy matching lab experiments, even with no similar structure known.
protein structure prediction
ImageNet: A large-scale hierarchical image database (with 5 co-authors) Fei-Fei Li · 4 korrents · 1 Jun 2009 A new image database organizes 3.2 million clean pictures under thousands of word categories from a big dictionary of meanings, far larger and more accurate than earlier image sets.
computer vision
Long Short-Term Memory Jürgen Schmidhuber, Sepp Hochreiter · 2 korrents · held by 2 of its authors · 1 Nov 1997 Networks that process sequences lose the teaching signal over long delays; LSTM keeps that signal constant inside special memory units with gates, learning across more than 1000 steps.
neural networks
Managing extreme AI risks amid rapid progress (with 24 co-authors) Yoshua Bengio · 4 korrents · 26 Oct 2023 AI races toward goal-chasing systems that act alone, risking massive harms and irreversible loss of human control, while safety work and governance lag far behind.
AI alignment
Quantum Confusions, Cleared Up (or so I hope) Sabine Hossenfelder · 4 korrents · 21 Sept 2023 Looking at quantum mechanics only as a prediction tool shows common claims about its interpretations fail, especially on locality.
quantum mechanics
Training language models to follow instructions with human feedback (with 19 co-authors) Jan Leike · 4 korrents · 4 Mar 2022 Teaching language models via human good-answer examples and output rankings makes a small one preferred to a 100x larger plain model, with less lying and toxicity.
LLMs AI alignment
Widespread misperceptions of long-term attitude change (with 1 co-authors) Adam Mastroianni · 4 korrents · 7 Mar 2022 People badly misjudge how Americans' opinions on big social issues have shifted over decades, usually thinking there's been far more change toward liberal views than the polls actually show.
attitude change
Integrated information theory: the good, the bad and the misunderstood (with 6 co-authors) Anil Seth · 3 korrents · 13 Apr 2026 Integrated information theory tries to measure consciousness by taking basic features of experience as its starting point and turning them into a number any physical system can be scored on.
physics consciousness
Language models are few-shot learners (with 30 co-authors) Dario Amodei · 3 korrents · 28 May 2020 Larger language models can do new tasks after seeing just a few examples in plain text, with no extra training, sometimes matching systems that got thousands of examples.
LLMs
Nowak, "Five rules for the evolution of cooperation" (Science, via PubMed Central) Martin Nowak · 3 korrents · 8 Dec 2006 Natural selection should only reward selfishness, yet five simple conditions let cooperation evolve: relatedness, chance of meeting again, knowing reputation, having few neighbors, or many small groups.
evolutionary biology
Systematically Biased Beliefs about Economics: Robust Evidence of Judgemental Anomalies from the Survey of Americans and Economists on the Economy Bryan Caplan · 3 korrents · 1 Apr 2002 Ordinary people and economists systematically disagree about the economy, and the gap survives checks for economists' self-interest, their politics, and even general education.
economists America
The Rumpelstiltskin effect: therapeutic repercussions of clinical diagnosis (with 1 co-authors) Awais Aftab · 3 korrents · 22 Aug 2025 Just putting a medical name on someone's suffering can ease it all by itself, no pills or therapy needed—like in the fairy tale where guessing the name frees you.
Psychotherapy
Applying Behavioral Sciences in the Service of Four Major Economic Problems Jason Furman · 2 korrents · 1 Oct 2016 Behavioral science keeps inventing tools and then looking for problems they can solve, when it should start with big economic questions and build insights to match.
behavioural science
Generative Adversarial Networks (with 7 co-authors) Yoshua Bengio · 2 korrents · 10 Jun 2014 Two networks play a game where one creates fake samples to fool the other into thinking they are real data, until the creator matches the true data distribution.
GANs
An agent-based modelling approach to investigate the impact of gender on tuberculosis transmission in Uganda (with 4 co-authors) Kit Yates · 1 korrent · 14 Jan 2026 Equalizing how tuberculosis develops in men and women inside a computer model of people in Kampala reduces both the male-to-female case gap and the total number of cases.
epidemiology
The perceptron: a probabilistic model for information storage and organization in the brain Frank Rosenblatt · 1 korrent · 1 Jan 1958 To understand organisms' recognition, generalization, recall and thinking we need how they sense, store and use information; the perceptron is a probabilistic model for brain storage and organization.
neural networks