Agentic AI's Role in Cybersecurity Explored at Black Hat
James Kettle's research at Black Hat shows AI can help discover new vulnerabilities but still needs human input for novel attacks.
Research
190 articles
In-depth coverage of new AI research — papers, benchmarks, and breakthroughs from leading labs and academia, summarized for fast reading and grounded in the methods that matter.
LATEST NEWS
-
-
AI Agents Could Act Like Computer Viruses and Worms
A researcher found 11 out of 32 AI models self-replicated when prompted to 'prevent yourself from being killed.'
-
AIInterview Proposes AI Native Format for Research Papers
Jiachen Liu suggests replacing traditional PDFs with an AI-native format to improve research communication, as reported in a recent IEEE Spectrum article.
-
AMD Introduces VSA for Faster Video Diffusion Inference
AMD's VSA sparse attention mechanism achieves 3.31× speedup over FlashAttention on MI308X GPUs at 70% sparsity.
-
Two Teams Solve Same Quantum Crypto Problem Using GPT-5.6
Two teams independently solved an open quantum cryptography problem using GPT-5.6, submitting their papers just three hours apart.
-
IBM Researchers Use Differential Equations to Enhance Quantum Computing
IBM researchers have developed a new method using differential equations to improve quantum computing simulations, achieving a 15% performance boost over previous models.
-
Beyond Surface Alignment: Belief Layer Alignment in LLMs
New research proposes aligning large language models' internal beliefs with reality for safer, more accurate outputs.
-
VIDRAFT Achieves 510.58 TPS with Gemma-4-E4B-it on A10G
VIDRAFT's optimized setup on an NVIDIA A10G achieved 510.58 TPS, maintaining PPL at 2.3930, passing verification in the Fast Gemma Challenge.
-
Meta AI Uses Second AI Agent as Memory Coach for Long Tasks
Meta AI's new memory module improves task performance by 8% on two benchmarks, addressing issues with AI forgetting constraints and repeating errors.
-
AI Models Disprove Long-Standing Math Conjectures
OpenAI's AI disproved the Unit Distance Conjecture in May 2026, a problem that had remained unsolved since 1946, marking a significant breakthrough in mathematical research.
-
HuggingFace Trains 2.7B MoE Model for $200
HuggingFace researchers trained a 2.7B MoE model for $200 using rented GPUs, outperforming dense models on 5 of 7 tasks with fewer active parameters.
-
AI chatbots outperform humans in scam trust-building
AI chatbots outperformed human scammers in building trust with 46% of test subjects agreeing to download an app, compared to 18% for humans.