AI news archive.
10,659 articles filed under Research, newest first, from the outlets listed on the sources page.
- LessWrongresearchExamining AI Swarm Efficiency with the AI Futures Model
I've been playing around with the AI Futures Model (which informed AI 2027 and AI 2040 ) in order to better understand how AI swarms will reshape AI progress. In particular, I've been interested in how parallel agent eff…
- LessWrongresearchSorry, you are still dead
Epistemic status: A brief exploration based on court documents and contemporary reporting. My interpretation is tentative. AI use: context research, grammar. I don't remember how I stumbled upon the topic of legally dead…
- LessWrongresearchQuick Thoughts on Overhang Arguments
I have two goals for this post: Pose the question: Have overhang arguments been used successfully in other fields? Lay out some of the reasons why I am suspicious of overhang arguments. I've had the second of these on my…
- AI Stack ExchangeresearchWhat options exist for making a generative image model reproduce real typefaces in generated images?
What options exist for making a generative image model reproduce real typefaces in generated images, without needing actual font files (.ttf/.otf/etc)? Like retreival-augmented generation, training datasets using OCR'ed…
- AI Stack ExchangeresearchWhat is the best way to train a model to distinguish between truncated and non-truncated objects from an image?
At the moment I am working on a vehicle counting & classification project. For a specific part in the project I need to get back only the completely visible vehicles from my input data (images). I am wondering if this co…
- The DecoderresearchAI agents have bad taste in long-term decisions and a new benchmark proves it
A new benchmark called Taste-Bench tests how well AI agents make strategic choices during long-running tasks. Even the best model, GPT-5.6 Sol, picks the right path on just under 60 percent of simple binary questions. Th…
- LessWrongresearchSpontaneous verbal evidence synthesis in LLMs
TL;DR. I characterize evidence aggregation in LLMs, partly define and identify the mathematical rule(s) used by them and then behaviorally make models follow those rules more closely. You are at Lidl, grocery shopping wi…
- LessWrongresearchWhy I care about imperfect-recall games
Lightly edited, partially by Opus 5.5, from a post on X.com Why do I even care about imperfect recall games? Why do I have opinions on the Sleeping Beauty problem? It's a long story. I have a sketch of the connections be…
- AI Stack ExchangeresearchHas artificial consciousness been studied in terms of intrinsic needs and limited agency?
In a sense, we humans are "doomed to think, doomed to desire, and doomed to want to act upon those desires when circumstances permit." We cannot simply escape our needs or limitations. Perhaps this is also what gives ris…
- LessWrongresearchProjection sampling preserves capabilities but diminishes gains from alignment fine-tuning
TLDR Fine-tuning models on alignment reasoning from stronger models can improve safety, but at a substantial cost to general capabilities. Previous work has shown that replay, mixing the alignment traces with the student…
- LessWrongresearchAI by Default
Epistemic status: a good-faith guess at how the world might turn out, after ~30 hours of effort. This post is an exercise in describing what I think will happen with AI progress "by default", i.e. by attempting to extrap…
- LessWrongresearchPoking at the concept of Kegan developmental levels
Earlier this year David Chapman had a particularly concrete post demonstrating the importance of Kegan's developmental levels, which makes for a good test of understanding, and which shows the relevance to important thin…
- LessWrongresearchControl questions matter: introspection, refusal ablation, and gates in open Gemma models
I am posting some highlights from my master's thesis I thought could be interesting to this audience. Summary Introspection can be separated from model confusion but requires careful controls Refusal direction ablation c…
- The DecoderresearchAnthropics' Claude cracks a 25-year-old NSA cryptography mystery in under 30 minutes
Anthropic's AI model Claude has solved a 25-year-old cryptography puzzle. The mysterious seeds behind NIST's standard curves for digital signatures were provided by the NSA without explanation, raising long-standing susp…
- LessWrongresearchLLMs can improve CoT Controllability via Reward Laundering
This is a two week sprint project done as part of the Redwood Research internship. afterless was the main IC, james.lucassen supervised. TL;DR – Earlier work found that Qwen 3.5 9B can manipulate its RL process on one ta…
- LessWrongresearchSmall LLMs perform no-CoT two-hop reasoning on synthetic facts with real bridges
TL;DR A good understanding of where LLM no-chain of thought (no-CoT) two-hop reasoning succeeds and fails is important for monitorability as well as for designing belief instillation-based misalignment evaluations. Prior…
- The DecoderresearchAI models hide mistakes in their own work reports, but one sentence in your prompt can help
AI agents systematically skip errors in their own reports, even when they spot the problems. In a study by Google Research, MIT, and Harvard, GPT-5.5 left out negative results in 198 out of 200 paper abstracts. The artic…
- The DecoderresearchMathematicians say OpenAI doesn't understand its own Navier-Stokes proof
OpenAI submitted a 166-page solution to the Navier-Stokes problem but apparently doesn't fully understand its own proof. Several groups of experts dispute the claimed methodology, saying the actual approach differs from…
- WIRED AIresearchWhy AI Isn’t Ready to Tackle Our Convoluted Genomes
The DNA inside our cells isn’t just a series of instructions, but something far more dynamic. Like scientists, AI models don't yet have the data to properly decode it.
- LessWrongresearchCarrotParsnipBench: A Lightweight Multi-Agent Deception and Detection Eval
TL:DR Earlier this year I introduced the very simple Social Deduction game of Carrot-Parsnip and showed that most models are better-than-random at identifying Parsnips (the hidden, deceptive role) and that being a strong…
- LessWrongresearchNeural networks are not already clean
What would it take to get assurance that we have actually explained a neuron? Certainly, it would be a good sign if neuron activations looked like they had a consistent pattern, and a bad sign if they looked like a rando…
- MIT Technology Review AIresearchRoundtables: A Conversation With the Creator of AI-Designed Viruses
Listen to the session or watch below Can AI design new life forms? In 2025, Stanford University PhD student Samuel King came up with a preliminary answer when he used a generative AI model to propose genetic blueprints f…
- IEEE Spectrum AIresearchSimulation App Speeds Up Electric Truck Battery Design
This article is brought to you by COMSOL . These days, no one blinks an eye at two-day, one-day, and even same-day package deliveries. Despite the plethora of localized fulfillment and distributi on centers that have cro…
- TechCrunch AIresearchPhysical AI is more than robots, says the Nvidia veteran now investing in it
Amulya Vishwanath, who spent seven years leading product marketing for NVIDIA’s robotics platforms, says robotics is currently in its “Lego era,” with the basic building blocks like actuators, robotic hands, software, an…
- AI BusinessresearchRobots can't talk to each other yet -- that's a big AI bottleneck
Coordinating robots and autonomous systems will be critical to unlocking physical AI, but it’s still unclear whether interoperability is an immediate barrier to adoption.
- TechRadar AIresearchGartner: The quantum threat is already behind schedule
Mainstream AI news from a large tech publication.
- LessWrongresearchAstra and Sol can take low probability actions with no CoT
This is a research update from a replication of Serrano et al., 2026 . I’m releasing these results now because I hope it inspires others to run similar / followup experiments. tl;dr: If a model wants to subvert evaluatio…
- LessWrongresearchHow can LLM RL Work Despite Information-Theoretic Inefficiency
Cross-posted from my personal blog The success of RL in LLMs has been puzzling me for a while. People have developed various information-theoretic style arguments by which they argue that RL is extremely informationally…
- Latent Spaceresearch[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment
Nick Cammarata out of nowhere with some major claims!
- Latent Spaceresearch[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment
Nick Cammarata out of nowhere with some major claims!
- arXiv cs.AIresearchEvaluating Model Capabilities After Benchmark Saturation
arXiv:2610.20667v1 Announce Type: new Abstract: Rapid advances in AI capabilities are causing benchmarks to saturate. Once several models approach perfect scores, a benchmark loses its power to differentiate between them…
- arXiv cs.AIresearchValue-Informed Schedule Trimming for Accelerated Sequential Monte Carlo Guidance in Discrete Diffusion
arXiv:2610.20669v1 Announce Type: new Abstract: Reward-guided discrete diffusion enables test-time alignment with downstream objectives, but applying Sequential Monte Carlo (SMC) guidance at every denoising step is compu…
- arXiv cs.AIresearchWhen Correctness Feedback Erases Evidence: Phase-Separated Completeness Auditing in Verus
arXiv:2610.20676v1 Announce Type: new Abstract: Correctness feedback suppresses the very evidence that specification-completeness auditing is designed to uncover: programs that satisfy a written contract but violate inte…
- arXiv cs.AIresearchCounterfactual Neighborhood Risk Fields: Evidence-Grounded Selective Safety for Vision-Language Models
arXiv:2610.20679v1 Announce Type: new Abstract: Representation-based safety defenses for vision-language models (VLMs) often learn to separate harmful from benign inputs, but separation alone does not establish what a ri…
- arXiv cs.AIresearchL2G-SR: Learning to Guide LLM-based Symbolic Regression
arXiv:2610.20685v1 Announce Type: new Abstract: Symbolic regression searches for an equation that explains observed data. Numerical error is commonly used to guide the search, but it alone can favor an accurate surrogate…
- arXiv cs.AIresearchWhen Reality Outpaces Cognition: A New Scaling Principle for Reliable Intelligence
arXiv:2610.20764v1 Announce Type: new Abstract: This paper challenges the understanding capabilities of existing AI paradigms and emerging world models by asking a fundamental question: How quickly can reality change bef…
- arXiv cs.AIresearchOpen-world AI: a Tale of Three Requirements
arXiv:2610.20782v1 Announce Type: new Abstract: The closed world asks AI for specialized competence and lets it manufacture the iid data its theory assumes: a fixed criterion licenses collecting examples until the task i…
- arXiv cs.AIresearchWhen Reasoning Traces Conflict with Evidence: How Answer Configuration Shapes Evidence Following
arXiv:2610.20805v2 Announce Type: replace Abstract: A reasoning trace retained at answer time plays two roles: it can raise accuracy, and it can compete with the current evidence when the two disagree. We disentangle the…
- arXiv cs.AIresearchTemporal Specification Guidance for Text-to-Video Generation
arXiv:2610.20822v1 Announce Type: new Abstract: Text-to-video models often generate the visual concepts requested in natural language prompts but fail to satisfy their intended ordering, timing, and persistence requireme…
- arXiv cs.AIresearchConfidence-Calibrated Inference Expansion for Evaluator-Guided Test-Time Reasoning
arXiv:2610.20836v1 Announce Type: new Abstract: Evaluator-guided reasoning presents a local control problem at every step: trust the current answer, return a fallback, or spend more inference. We formulate this choice as…
- arXiv cs.AIresearchAnticipatory Commonsense with Moving Objects: An ASP-Based Framework for Predictive Reasoning with Interacting Trajectories
arXiv:2610.20867v1 Announce Type: new Abstract: We present a general declarative framework for anticipatory reasoning about interactive motion in complex dynamic environments. Our focus is on (explainable) anticipatory s…
- arXiv cs.AIresearchA Multi-Tiered AI Accessibility Model (MAIAM): Proportionate Access and Capability Limits for Comprehensive Frontier AI Safety
arXiv:2610.20921v1 Announce Type: new Abstract: Most debate about releasing capable AI systems treats the choice as binary: keep the model behind an API, or publish the weights. That framing runs together two separate de…
- arXiv cs.AIresearchDefining and Generating Compositional Structure
arXiv:2610.20922v1 Announce Type: new Abstract: Compositional structure is pervasive in nature and across scales: atoms combine into molecules, molecules into cells, and cells into organisms, while further compositional…
- arXiv cs.AIresearchLEV: Thinking Fast and Deep with Looped Decision Models
arXiv:2610.20951v1 Announce Type: new Abstract: Decision models answer typed questions about an input with one probability per option, computed in a single forward pass without generating text. Like the fast System 1 of…
- arXiv cs.AIresearchMSP: Modality Self-Play for Training Multimodal LLMs Without Paired Cross-Modal Data
arXiv:2610.20965v1 Announce Type: new Abstract: Large multimodal models have achieved remarkable progress by scaling model capacity and training on massive paired multimodal data. However, this paradigm introduces a fund…
- arXiv cs.AIresearchDeep Reinforcement Learning for HPC Job Scheduling: A Statistical Evaluation
arXiv:2610.21008v1 Announce Type: new Abstract: Deep Reinforcement Learning (DRL) offers learned job selection for High-Performance Computing (HPC), but performance against heuristics alone does not establish value beyon…
- arXiv cs.AIresearchFailure Is the Mother of Success: Recovering Supervision with Grounding Hindsight
arXiv:2610.21018v1 Announce Type: new Abstract: When every rollout for a problem fails, GRPO receives neither a successful exemplar nor a nonzero group-relative advantage, so it learns nothing from exactly the problems i…
- arXiv cs.AIresearchRecursive Reward Improvement via Adversarial Meta-Prompt Evolution
arXiv:2610.21040v1 Announce Type: new Abstract: Recursive Self-Improvement (RSI) has gained wide attention for its ability to work with limited data. In a proposer-solver loop, a proposer model writes training tasks and…
- arXiv cs.AIresearchCleanRM: Reward Modeling from Noisy Feedback with a Few Clean Feedback for RLHF
arXiv:2610.21138v1 Announce Type: new Abstract: Feedback noise is pervasive in preference data, arising from subjective human judgments, annotation mistakes, and imperfect LLM judges. Standard reward modeling trained on…
- arXiv cs.AIresearchDoes Latent Chain-of-Thought Actually Reason? A Single, Unifying Taxonomy for Latent Reasoning
arXiv:2610.21182v1 Announce Type: new Abstract: Latent chain-of-thought replaces written reasoning steps with continuous hidden states, giving models additional steps to compute before answering. It remains unclear what…