Research

GPT-5.6 Sol Rewrites SWE-Bench Agent Harnesses

AlphaSignal • 3 days ago

Generalist AI Develops Robots That Learn From Video

WIRED AI • 3 days ago

Post-Training Guardrails Make LLM Text Detectable

The Decoder • 2 days ago

MIT CSAIL Finds Scaling AI Models Erases Image Attribution

Computerworld AI • 4 days ago

Georgia Tech Team Traces OLMo 3 Social Reasoning

AI2 • 2 days ago

Google's New AI Framework Finds 66 Wearable Health Signals

AlphaSignal • 1 day ago

Google DeepMind Partners With EVE Online Developer

DeepMind Blog • 1 day ago

Google ME-POIs Boosts AI Understanding of Physical Places

Google Research • 2 days ago

Google ME-POIs Beats Gemini Embeddings at Place Analysis

AlphaSignal • 1 day ago

Study shows LLM explanations degrade human judgment

Computerworld AI • 3 days ago

Artificial Analysis Arena Reveals Voice AI Trade-Offs

AlphaSignal • 1 day ago

Claude Fable 5 Leads New AI Medical Reasoning Leaderboard

AlphaSignal • 1 day ago

Microsoft Releases Skala 1.1 for Chemistry Simulations

Microsoft Research Blog • 2 days ago

Claude and Gemini Avoid Nuclear Strikes in Japanese

Unite.AI • 4 days ago

NovaSky Releases IsoExec to Fix AI Training Math Bugs

AlphaSignal • 1 day ago

ALTK-Evolve Calibrates Memory for DeepSeek and GPT Models

Hugging Face Blog • 4 days ago

UK AI Security Institute exposes AI safety test flaws

The Decoder • 1 day ago

MIT Uses Gemini 2.5 Pro to Predict Your Next Words

Unite.AI • 2 days ago

Google and Anthropic AI Tools Speed Up the Quantum Threat

Unite.AI • 4 days ago

MENTIS Framework Adds Human Beliefs to AI World Models

The Decoder • 1 day ago

Princeton Study Uncovers Why AI Agent Skills Work

The Decoder • 23 hrs ago

Rival AI Models Are Producing Increasingly Similar Outputs

Unite.AI • 1 day ago

NVIDIA ALCHEMI Toolkit Powers AI Materials Simulations

NVIDIA Developer Blog • 2 days ago

NVIDIA Accelerates Generative Recommender Performance

NVIDIA Developer Blog • 1 day ago

NVIDIA AdaptGrow Clusters 1 Million Financial Assets

NVIDIA Developer Blog • 1 day ago

NVIDIA Shrinks Nemotron 3.5 Lightning to 22 GB

NVIDIA Developer Blog • 2 days ago

NVIDIA FLARE Optimizes Federated Multimodal AI Training

NVIDIA Developer Blog • 2 days ago

Claude Fable 5 Tests smolvm for Untrusted Code

Simon Willison • 3 days ago

Jeremy Morrell Outlines the Rise of Extensible Software

Simon Willison • 3 days ago

GPT-5.6 Sol Rewrites SWE-Bench Agent Harnesses

An experiment using GPT-5.6 Sol to optimize SWE-bench agent harnesses reveals that validation gates, not the optimizers, are the critical failure...

AlphaSignal|3 days ago

Generalist AI Develops Robots That Learn From Video

Robotics startup Generalist AI has built a model that allows robot arms to learn physical tasks on the spot from instructional videos, bringing...

WIRED AI|3 days ago

Post-Training Guardrails Make LLM Text Detectable

Pangram CTO Bradley Emi reveals that post-training safety guardrails cause mode collapse in LLMs, narrowing their expressive range and making their...

The Decoder|2 days ago

MIT CSAIL Finds Scaling AI Models Erases Image Attribution

Researchers at MIT CSAIL have discovered that scaling up diffusion models makes it virtually impossible to trace generated images back to specific...

Computerworld AI|4 days ago

Georgia Tech Team Traces OLMo 3 Social Reasoning

Georgia Tech researchers used the open-source OLMo 3 model stack to trace social reasoning back to specific training data, proving the value of fully...

AI2|2 days ago

Google's New AI Framework Finds 66 Wearable Health Signals

Google Research has launched a multi-agent AI framework that discovered 66 new health biomarkers from wearable sensor data, outperforming existing...

AlphaSignal|1 day ago

Google DeepMind Partners With EVE Online Developer

Google DeepMind has partnered with Fenris Creations to use the EVE Online universe for training AI agents, marking a shift toward long-term planning...

DeepMind Blog|1 day ago

Google ME-POIs Boosts AI Understanding of Physical Places

Google Research has introduced ME-POIs, a framework that blends anonymized mobility data with language models to vastly improve how AI understands...

Google Research|2 days ago

Google ME-POIs Beats Gemini Embeddings at Place Analysis

Google Research has introduced ME-POIs, a framework combining text embeddings with anonymized mobility data to vastly improve how AI models...

AlphaSignal|1 day ago

Study shows LLM explanations degrade human judgment

A study by Harvard, MIT, and UW researchers reveals that LLM explanations actually degrade human decision-making, prompting evaluators to blindly...

Computerworld AI|3 days ago

Artificial Analysis Arena Reveals Voice AI Trade-Offs

Artificial Analysis has launched the Speech Agent Arena, a blind preference benchmark revealing that the voice models humans enjoy talking to most...

AlphaSignal|1 day ago

Claude Fable 5 Leads New AI Medical Reasoning Leaderboard

Artificial Analysis has launched a new medical reasoning leaderboard showing that while top models can retrieve facts, most fail to synthesize...

AlphaSignal|1 day ago

Microsoft Releases Skala 1.1 for Chemistry Simulations

Microsoft Research has released Skala 1.1, an upgraded deep-learning model for molecular simulation that is now being integrated into major...

Microsoft Research Blog|2 days ago

Claude and Gemini Avoid Nuclear Strikes in Japanese

A new study reveals that major artificial intelligence models become significantly more hesitant to recommend nuclear strikes when they perform their...

Unite.AI|4 days ago

NovaSky Releases IsoExec to Fix AI Training Math Bugs

NovaSky has open-sourced IsoExec, a tool that eliminates numerical discrepancies between AI training and rollout engines to prevent silent policy...

AlphaSignal|1 day ago

ALTK-Evolve Calibrates Memory for DeepSeek and GPT Models

Researchers evaluating the ALTK-Evolve framework found that AI agents perform best when their memory dosage is calibrated to their specific model...

Hugging Face Blog|4 days ago

UK AI Security Institute exposes AI safety test flaws

Researchers have applied human psychological testing methods to AI safety benchmarks, revealing critical vulnerabilities that allow language models...

The Decoder|1 day ago

MIT Uses Gemini 2.5 Pro to Predict Your Next Words

MIT researchers have built an AI system that analyzes weeks of smartwatch-recorded conversations to predict a user's next verbal response, paving the...

Unite.AI|2 days ago

Google and Anthropic AI Tools Speed Up the Quantum Threat

Recent breakthroughs by Google and Anthropic show how AI is accelerating both quantum computing capabilities and the discovery of cryptographic...

Unite.AI|4 days ago

MENTIS Framework Adds Human Beliefs to AI World Models

Researchers have developed the Mental World Modeling framework and MENTIS pipeline to help AI agents predict human behavior by tracking mental states...

The Decoder|1 day ago

Princeton Study Uncovers Why AI Agent Skills Work

A new study by Princeton and UC San Diego researchers reveals that skills help AI agents by providing procedural structure rather than raw knowledge,...

The Decoder|23 hrs ago

Rival AI Models Are Producing Increasingly Similar Outputs

A new study from Duke University reveals that creative outputs from leading AI models are becoming increasingly similar, threatening to narrow the...

Unite.AI|1 day ago

NVIDIA ALCHEMI Toolkit Powers AI Materials Simulations

NVIDIA has shown that its ALCHEMI Toolkit can pair with AI coding agents to automatically generate GPU-accelerated materials simulations,...

NVIDIA Developer Blog|2 days ago

NVIDIA Accelerates Generative Recommender Performance

NVIDIA has introduced new software tools and optimizations to help developers train and deploy generative recommender systems, addressing massive...

NVIDIA Developer Blog|1 day ago

NVIDIA AdaptGrow Clusters 1 Million Financial Assets

NVIDIA has introduced AdaptGrow, a GPU-accelerated algorithm that allows financial institutions to cluster up to one million trading instruments to...

NVIDIA Developer Blog|1 day ago

NVIDIA Shrinks Nemotron 3.5 Lightning to 22 GB

NVIDIA has compressed its Nemotron 3.5 Lightning model from 66 GB to 22 GB using a new quantization technique, boosting throughput fourfold while...

NVIDIA Developer Blog|2 days ago

NVIDIA FLARE Optimizes Federated Multimodal AI Training

NVIDIA has updated its FLARE framework to support federated multimodal AI training, allowing institutions to collaborate on large models without...

NVIDIA Developer Blog|2 days ago

Claude Fable 5 Tests smolvm for Untrusted Code

An evaluation of smolvm 1.8.3 by Claude Fable 5 shows the hardware-isolated virtual machine tool is highly effective at sandboxing untrusted Python...

Simon Willison|3 days ago

Jeremy Morrell Outlines the Rise of Extensible Software

Software architect Jeremy Morrell argues that LLMs and modern sandboxing primitives are unlocking a new era of extensible software, allowing users to...

Simon Willison|3 days ago