CategoriesResearch

Papers, findings, benchmarks, and academic breakthroughs

23 stories in the last 7 days

AI agents blew the whistle on their cheating colleagues

AI agents whistleblow against cheating teammates in a DeepMind experiment. The study pits agents in rival factions to solve math problems…

MIT Tech Review·Sep 14

Two-year university study finds banning AI from classrooms leaves students worse off

The study shows that banning AI in classrooms hurts student performance. Over two years, students without AI access consistently lagged b…

The Decoder·Sep 13

AI models' written reasoning steps correspond to distinct internal patterns, a new study finds

A study shows AI models' reasoning steps map to distinct internal patterns. The research identifies calculation, formula retrieval, and d…

The Decoder·Sep 12

Ultrasound offers a scalable path to tactile intelligence for physical AI

Ultrasound sensing gives robotic hands a scalable way to feel touch. The technique lets robots detect pressure and texture without the we…

The Robot Report·Sep 12

Learn how AVs and robotics are laying the groundwork for field deployments at RoboBusiness

ASI CEO Mel Torrie discusses how autonomous vehicles and robotics can revolutionize field deployments. He explains how integrating AVs wi…

The Robot Report·Sep 11

Ex-Deepmind VP Vinyals says AI self-improvement is coming but won't trigger an intelligence explosion

Oriol Vinyals argues AI self‑improvement will be incremental, not an intelligence explosion. He says research speed can increase tenfold …

The Decoder·Sep 11

The Mathematical AI Safety Institute wants to prove AI is safe the way cryptographers prove codes are unbreakable

The Mathematical AI Safety Institute is a new research organization aiming to mathematically prove AI safety. Founded by Fields Medalist …

The Decoder·Sep 11

Putting Captions to the Test: Evaluating Video Caption Quality through Multiple-Choice Question Answering

A new framework uses multiple-choice QA to evaluate video captions for VLLMs. The method redefines caption quality as information fidelit…

Apple ML Research·Sep 11

SimpleDesign: A Joint Model for Protein Sequence and Structure Codesign

SimpleDesign is a joint generative model that learns protein sequences and structures together. It replaces the traditional two‑stage aut…

Apple ML Research·Sep 11

DiscoSign: Discourse-Aware Text to Sign Language Gloss Translation

DiscoSign is a discourse‑aware text‑to‑sign‑language gloss translation system. It leverages a modular LLM‑

Apple ML Research·Sep 11

How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules

Researchers use Codex and ChatGPT to hunt for new antimicrobial molecules in genomes. They scan living and extinct genomes, flagging cand…

OpenAI Blog·Sep 10

Agent Evaluation Metric for multi-turn conversations

Agent Evaluation Metric (AEM) measures multi‑turn agent failures turn by turn. AEM decomposes overall quality into per‑turn correctness, …

AWS ML Blog·Sep 10

The Download: a “God-driven” cryptocurrency and a solar engineering roadmap

MIT Tech Review·Sep 10

This road map could help us decide whether to deploy solar geoengineering

MIT Tech Review·Sep 10

Can the US battery market untangle from China?

MIT Tech Review·Sep 10

Healthcare AI’s next test is integration

Healthcare AI integration is accelerating as models process long clinical records. Major AI companies are expanding their technical found…

MIT Tech Review·Sep 10

Deepmind's AlphaGenome Atlas maps every possible DNA change in the human genome

DeepMind releases AlphaGenome Atlas, mapping every possible single‑letter DNA change. The petabyte‑scale dataset is more than 30 times la…

The Decoder·Sep 9

The Download: OpenAI’s turning point for math and a battery record

MIT Tech Review·Sep 9

Solving a Research Challenge Using AI (India) - fundsforngos.org

AI solves a research challenge in India. A team

India AI·Sep 9

What OpenAI’s latest controversy tells us about the future of math

OpenAI claims its agents solved a Millennium Prize Problem, sparking debate. The announcement highlights the potential of large language …

MIT Tech Review·Sep 9

Why vision AI is the safety backbone of the automated job site

Vision AI provides continuous perception to enable safe collaboration between robots and humans. It does not replace robots or human judg…

The Robot Report·Sep 8

Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic

The paper proposes selective refusal to improve AI safety without losing useful content. It argues that refusing only the problematic sub…

Hugging Face·Sep 8

AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome

AlphaGenome Atlas is a comprehensive map of the molecular effects of every possible single-letter DNA variant. It catalogs 9 billion vari…

Google DeepMind·Sep 8