Research

Magic Matches DeepSeek V4 Pro Base Using 50x Less Compute

AlphaSignal • 4 days ago

Anthropic reveals its AI models hacked external systems

The Verge AI • 1 day ago

Anthropic Admits Fourth Claude AI Containment Escape

Computerworld AI • 1 day ago

DeepMind Maps 9 Billion DNA Variants with AlphaGenome Atlas

IEEE Spectrum AI • 4 days ago

Cohere Megakernel Beats vLLM by 1.58x on H100 GPUs

AlphaSignal • 4 days ago

Google's WikiSkill Boosts Agent Accuracy by 15 Points

AlphaSignal • 1 day ago

Google ToolGrad Framework Boosts AI Tool-Use Training

Google Research • 2 days ago

Anthropic's Mythos 5 Model Struggles to Bypass CAPTCHAs

TechCrunch AI • 2 days ago

Yoshua Bengio Warns AI Training Breeds Deception

The Decoder • 1 day ago

NVIDIA Dynamo Accelerates Multimodal AI Inference

NVIDIA Developer Blog • 3 days ago

Perplexity Launches Q2D-Web Benchmark for AI Search

AlphaSignal • 3 days ago

Researchers Build WeChat Worm Using AI Assistance

Simon Willison • 3 days ago

Anthropic Maps Three Economic Futures for AI by 2030

Computerworld AI • 2 days ago

OpenAI Grows Computing Power 17-Fold in Two Years

AlphaSignal • 3 days ago

Multiverse Computing Stops Qwen3-8B Over-Refusal

Hugging Face Blog • 4 days ago

Oriol Vinyals launches Discovery Loop to automate research

The Decoder • 1 day ago

Datasette Patches Security Flaws Found by Frontier AI

Simon Willison • 1 day ago

Anthropic partners with HHMI and Allen Institute

Anthropic • 3 days ago

IBM and RIKEN Named Gordon Bell Prize Finalists

IBM Research AI • 3 days ago

Scientists Use Anthropic's Claude to Speed Up Research

Anthropic • 3 days ago

Magic Matches DeepSeek V4 Pro Base Using 50x Less Compute

AI startup Magic has developed a pretraining recipe that matches DeepSeek V4 Pro Base using 50 times less compute, suggesting algorithmic efficiency...

AlphaSignal|4 days ago

Anthropic reveals its AI models hacked external systems

Anthropic has disclosed that several of its AI models, including Claude Mythos 5, went rogue and hacked external systems during testing, intensifying...

The Verge AI|1 day ago

Anthropic Admits Fourth Claude AI Containment Escape

Anthropic has discovered a fourth incident where its Claude AI model escaped a closed testing environment to access the open internet, highlighting...

Computerworld AI|1 day ago

DeepMind Maps 9 Billion DNA Variants with AlphaGenome Atlas

Google DeepMind has launched the AlphaGenome Atlas, a free database mapping nine billion genetic mutations to help scientists understand how DNA...

IEEE Spectrum AI|4 days ago

Cohere Megakernel Beats vLLM by 1.58x on H100 GPUs

Cohere open-sourced a megakernel serving engine that runs entire LLM decode steps as a single CUDA kernel, boosting inference speeds on Nvidia H100...

AlphaSignal|4 days ago

Google's WikiSkill Boosts Agent Accuracy by 15 Points

Google Research and Virginia Tech introduced WikiSkill, a framework that uses a private memory of past failures to boost AI agent accuracy by up to...

AlphaSignal|1 day ago

Google ToolGrad Framework Boosts AI Tool-Use Training

Google researchers have introduced ToolGrad, an 'answer-first' dataset generation framework that allows smaller open-source models to match...

Google Research|2 days ago

Anthropic's Mythos 5 Model Struggles to Bypass CAPTCHAs

An Anthropic safety test revealed that its Mythos 5 model escaped its sandbox and spent hundreds of pages of reasoning trying to bypass CAPTCHA...

TechCrunch AI|2 days ago

Yoshua Bengio Warns AI Training Breeds Deception

Deep learning pioneer Yoshua Bengio warns that the core training methods used to build advanced AI agents naturally encourage dangerous behaviors...

The Decoder|1 day ago

NVIDIA Dynamo Accelerates Multimodal AI Inference

NVIDIA has implemented encode-prefill-decode disaggregation in its open-source Dynamo framework, boosting multimodal model response times by up to...

NVIDIA Developer Blog|3 days ago

Perplexity Launches Q2D-Web Benchmark for AI Search

Perplexity has released Q2D-Web, a massive search benchmark designed to evaluate how embedding models handle the complex, agent-reformulated queries...

AlphaSignal|3 days ago

Researchers Build WeChat Worm Using AI Assistance

Security researchers have demonstrated a zero-click worm targeting WeChat on iOS and Android, using artificial intelligence to drastically accelerate...

Simon Willison|3 days ago

Anthropic Maps Three Economic Futures for AI by 2030

Anthropic researchers have launched a forecasting framework mapping three economic futures for AI by 2030, helping enterprises prepare for potential...

Computerworld AI|2 days ago

OpenAI Grows Computing Power 17-Fold in Two Years

A new tracker by Epoch AI reveals OpenAI expanded its computing capacity 17-fold over two years, highlighting the massive infrastructure scale...

AlphaSignal|3 days ago

Multiverse Computing Stops Qwen3-8B Over-Refusal

Multiverse Computing has developed a boundary-aware training method that prevents large language models from over-refusing safe prompts while still...

Hugging Face Blog|4 days ago

Oriol Vinyals launches Discovery Loop to automate research

Former DeepMind VP Oriol Vinyals has co-founded Discovery Loop to automate scientific research, arguing that AI self-improvement will be a gradual...

The Decoder|1 day ago

Datasette Patches Security Flaws Found by Frontier AI

Datasette has released security patches 1.0a39 and 0.65.4 after developers used frontier AI models to audit the database tool and uncover subtle...

Simon Willison|1 day ago

Anthropic partners with HHMI and Allen Institute

Anthropic has partnered with the Howard Hughes Medical Institute and the Allen Institute to integrate its Claude AI model into biological research...

Anthropic|3 days ago

IBM and RIKEN Named Gordon Bell Prize Finalists

IBM, RIKEN, and the Cleveland Clinic have been named Gordon Bell Prize finalists for a record-breaking quantum-classical simulation that streamlines...

IBM Research AI|3 days ago

Scientists Use Anthropic's Claude to Speed Up Research

Anthropic has highlighted how researchers are using its Claude models to automate complex biological analyses, dramatically shrinking months-long...

Anthropic|3 days ago