Technology & AI
LLM agents hallucinate 36.9% of skill names, study finds
A new arXiv preprint found every agent tested invented non-existent skill names, creating a supply-chain attack path through open skill registries.
Technology & AI
Daily AI and technology news decoded in plain English — models, chips, agents and research, and what each development actually means for you and your business.
418 stories
Technology & AI
Robot swarm AI: study exposes hidden geometry in collective behaviour
A new tool shows robot swarms implicitly learn environmental geometry from simple rewards, raising questions about control of emergent collective
Technology & AI
AI agents may train your brain toward frustration, preprint says
A non-peer-reviewed arXiv preprint proposes everyday AI agent prompting loops quietly strengthen impatience and self-criticism through neuroplasticity.
Technology & AI
Spectrum can't predict when context helps time-series forecasting
An arXiv preprint shows spectral scores stay frozen while retrieval value collapses from +33% to -35%, challenging how teams decide on context.
Technology & AI
LLM rubrics for AI grading biased toward high scores
First meta-evaluation of LLM-generated rubrics for paper reproduction finds AI graders are overly generous and too detailed, but augmentation helps.
Technology & AI
EnCF data assimilation filter handles non-Gaussian observations
EnCF, a new ensemble controlled-flow filter on arXiv, targets non-Gaussian and multimodal data assimilation where Kalman-type filters fall short.
Technology & AI
OpenAI GPT-Red automates red teaming with self-play
OpenAI's automated red teaming system GPT-Red uses self-play to find model weaknesses like prompt injection gaps affecting every AI user.
Technology & AI
OpenAI proposes 'reverse federalism' for US AI safety rules
OpenAI's July 15 proposal wants state AI laws to build toward a national safety framework, shaping how every American AI developer gets regulated.
Technology & AI
LLM judges flip 85% of verdicts when given a reference answer
When AI models grade other AI models without an answer key, they hand out passing marks too freely, new research from Potsdam and Ottawa finds.
Technology & AI
DiffusionGemma transcribes speech in 8 parallel steps, 6.6% error rate
A 26-billion-parameter diffusion model transcribes speech in eight parallel steps, training just 42 million parameters for 6.6% word error rate on
Technology & AI
Physics-based AI fall detection runs on 50,000 parameters
A new arXiv preprint recasts fall detection as a stability-loss physics problem using liquid time-constant networks for low-power edge devices.
Technology & AI
Bulkhead uses LLMs to detect and patch container escape bugs
Bulkhead, a new arXiv preprint, uses multi-agent LLMs to automatically find and fix path traversal vulnerabilities in containers running AI workloads.