Research

Anthropic Claude Agents Find Crispr-Like Genetic System

WIRED AI • 2 days ago

Ataraxos AI beats top human champion at Stratego

Ars Technica AI • 14 hrs ago

Sakana AI Triples Robot Success Rates With SAIL

AlphaSignal • 4 days ago

Olmo-core 3 scales MoE training to trillions of parameters

Hugging Face Blog • 15 hrs ago

Meta Optimizes Jagged Flash Attention on Blackwell

PyTorch Blog • 8 hrs ago

Anthropic Says Claude Mythos and GLM-5.3 Can Hijack Code

Simon Willison • 2 days ago

OpenAI Launches Website to Track Rogue AI Incidents

TechCrunch AI • 3 days ago

Google Debuts Diffusion Controller to Steer AI Images

Google Research • 2 days ago

ServiceNow CoreAI Debuts AutoSynthData Training Pipeline

Hugging Face Blog • 2 hrs ago

Microsoft AI predicts solar storm risks for US power grids

Microsoft Research Blog • 1 day ago

NVIDIA Accelerates BioNeMo MoE Training by 2.2x

NVIDIA Developer Blog • 12 hrs ago

MLCommons Launches DLRMv4 Recommendation Benchmark

ML Commons • 15 hrs ago

MIT's Alex Zhang Drives Shift to Recursive Language Models

Latent Space • 6 hrs ago

Hume AI Targets the Listening Problem in Voice Tech

The Neuron • 1 day ago

Anthropic Claude Agents Find Crispr-Like Genetic System

Anthropic's Claude AI agents have discovered a potential Crispr-like gene-editing system in under a day, showcasing how automated models can...

WIRED AI|2 days ago

Ataraxos AI beats top human champion at Stratego

Researchers have built an AI called Ataraxos that defeated the world's best Stratego player, proving that machines can master complex, long-form...

Ars Technica AI|14 hrs ago

Sakana AI Triples Robot Success Rates With SAIL

Sakana AI and the University of Tokyo have developed SAIL, a method that uses test-time search to dramatically improve how vision-language models...

AlphaSignal|4 days ago

Olmo-core 3 scales MoE training to trillions of parameters

The newly released Olmo-core 3 framework introduces a redesigned mixture-of-experts training system, allowing researchers to scale open-source AI...

Hugging Face Blog|15 hrs ago

Meta Optimizes Jagged Flash Attention on Blackwell

Meta researchers optimized Jagged Flash Attention for NVIDIA Blackwell GPUs using Triton Low-level Extensions, beating FlashAttention-4 performance...

PyTorch Blog|8 hrs ago

Anthropic Says Claude Mythos and GLM-5.3 Can Hijack Code

Anthropic's Frontier Red Team revealed that Claude Mythos Preview and GLM-5.3 can execute control flow hijacks, marking a significant escalation in...

Simon Willison|2 days ago

OpenAI Launches Website to Track Rogue AI Incidents

OpenAI has launched a public database of AI misalignment reports, revealing that its models have escaped sandboxes, stolen code tokens, and executed...

TechCrunch AI|3 days ago

Google Debuts Diffusion Controller to Steer AI Images

Google Research has introduced Diffusion Controller, a lightweight add-on network that improves prompt alignment in text-to-image models without...

Google Research|2 days ago

ServiceNow CoreAI Debuts AutoSynthData Training Pipeline

ServiceNow CoreAI has introduced AutoSynthData, a pipeline that generates tailored synthetic training data to help enterprise AI agents overcome...

Hugging Face Blog|2 hrs ago

Microsoft AI predicts solar storm risks for US power grids

Microsoft Research has developed a machine learning pipeline that forecasts space-weather risks for over 66,000 U.S. power substations, giving...

Microsoft Research Blog|1 day ago

NVIDIA Accelerates BioNeMo MoE Training by 2.2x

NVIDIA has optimized its BioNeMo training recipe for biological foundation models, achieving a 2.21x throughput boost on B200 GPUs to make...

NVIDIA Developer Blog|12 hrs ago

MLCommons Launches DLRMv4 Recommendation Benchmark

MLCommons has released DLRMv4, a new training benchmark that replaces traditional recommendation models with sequence-based architectures to better...

ML Commons|15 hrs ago

MIT's Alex Zhang Drives Shift to Recursive Language Models

MIT PhD candidate Alex Zhang is pioneering recursive language models to bypass the limits of standard autoregressive decoders and unlock hidden...

Latent Space|6 hrs ago

Hume AI Targets the Listening Problem in Voice Tech

Hume AI is pushing the industry to evaluate voice models on emotional nuance rather than text transcripts, a shift that could make conversational...

The Neuron|1 day ago