AI1 min read
Researchers used Claude Opus 5 to hack OpenAI
Hacktron AI used a new version of Claude to breach OpenAI's security.
From TechCrunch AI
Loading
90 days of the enterprise, pro and team plans free, no card needed. See the plans
Blog
Daily notes on new models, LLM releases, agent frameworks and AI research, written from the sources we follow and delivered as a newsletter every day.
Get the morning and evening issues
Every new post of the day, in one email. Confirmation required.
AI1 min read
Hacktron AI used a new version of Claude to breach OpenAI's security.
From TechCrunch AI
AI1 min read
PrismML launched Bonsai 2, a 5.9 GB model based on Qwen3.8. It compresses the original by nine to ten times while keeping most performance.
From TechCrunch AI
Photo: Yan Krukau
AI labs use more AI to track agent errors. Some companies prefer human logs instead.
From TechCrunch AI
AI1 min read
Base Labs joined Hugging Face and Goodfire AI to build safety tools for open models. They aim to make safety transparent and part of the training process.
From TechCrunch AI
AI1 min read
Nunchux AI launched VC-Attention to speed up video diffusion models without retraining. It uses low-bit quantization and replaces the slow softmax stage.
From MarkTechPost
AI1 min read
Knowledgator released GLiFormer, an encoder that extracts nested JSON without generating tokens.
From MarkTechPost
Agents2 min read
Researchers released 38 open-source skills to improve AI reasoning in healthcare. Agents using these skills win 70% to 86% of tests against those without them.
From AWS machine learning blog
AI1 min read
Anthropic CEO Dario Amodei wants outside groups to audit AI safety. Experts say fixing network security basics is more effective.
From TechCrunch AI
Models1 min read
Mozilla Firefox now uses Mistral models for its Smart Window assistant. Mistral will serve users in France, North America, the UK, and Germany.
From Mistral AI news
AI1 min read
Nums AI launched Causilo, a tabular foundation model that ranks first among single models on TabArena for both classification and regression. It features an in-context learning architecture with Apache-2.0 code but requires a separate license for production use.
From MarkTechPost
LLMs2 min read
A new study finds few-shot prompting effects vary by task and are driven by demonstration content rather than prompt length. Replacing demonstrations with random text isolates the impact of representation shifts caused by content.
From arXiv cs.CL
LLMs2 min read
Researchers propose inference networks that route queries to lower-cost models first, activating expensive experts only when confidence drops below a specific threshold.
From arXiv cs.CL
LLMs1 min read
Together AI released a playbook to help companies migrate from closed to open source models.
From Together AI blog
AI1 min read
Sakana AI researchers introduced PC-ALM, a layer-local training method for deep networks that achieves performance comparable to backpropagation up to 1000 layers on MNIST. The research provides a JAX implementation for experimentation and benchmarking.
From MarkTechPost
Research1 min read
MEPN improves few-shot sensor fault diagnosis by aggregating prototypes from multiple episodes, achieving higher accuracy than single-episode baselines.
From arXiv cs.AI
Research1 min read
Researchers released Occamy-1.0, a 35 billion parameter model trained on a Qwen3.6-35B checkpoint. Evaluations across co-work benchmarks show it consistently performs among the strongest models of its size, achieving a competitive position on a cost-performance Pareto frontier.
From arXiv cs.AI
LLMs1 min read
The ORQA framework provides a method for testing large language model knowledge across 116 occupations using data sourced from trusted occupation-specific websites. Testing of 15 models revealed performance variations, with Claude Opus and GPT-5.4 achieving approximately 58-62% accuracy, while open-weight models showed 33-41% accuracy.
From arXiv cs.CL
Research1 min read
Research indicates that large language models are inadequate for consequential quantitative decisions like risk pricing or network intrusion containment. The paper proposes Large Quantitative Models (LQMs) that address key limitations regarding reproducibility, lineage, and calibrated uncertainty.
From arXiv cs.AI
LLMs1 min read
The research introduces Chopthin-Consensus Power Sampling (CCPS), a method for LLM decoding that preserves reasoning diversity and improves accuracy. Evaluation across open-weight models and benchmarks shows CCPS matches or exceeds the Power-SMC baseline in 14 of 15 settings, achieving gains of up to 10.6 percentage points.
From arXiv cs.CL
Posts are drafted from public feeds by models OpenSmartRoute routes to - the same router, skill and metering customers use - and always link to the original source. Corrections: support.
Archive (100)
