MemRL outperforms RAG on complex agent benchmarks without fine-tuning

MemRL separates stable reasoning from dynamic memory, giving AI agents continual learning abilities without model fine-tuning ...

The 11 runtime attacks breaking AI security — and how CISOs are stopping them

CrowdStrike's 2025 data shows attackers breach AI systems in 51 seconds. Field CISOs reveal how inference security platforms defend against prompt injection, model extraction, and 9 other runtime ...
Microsoft

RetroInfer: A Vector-Storage Approach for Scalable Long-Context LLM Inference

The growing context lengths of large language models (LLMs) pose significant challenges for efficient inference, primarily due to GPU memory and bandwidth constraints. We present RetroInfer, a novel ...
Wall Street Journal

Meta Is Developing a New AI Image and Video Model Code-Named ‘Mango’

AI tools like Google’s Veo 3 and Runway can now create strikingly realistic video. WSJ’s Joanna Stern and Jarrard Cole put them to the test in a film made almost entirely with AI. Watch the film and ...
Morningstar

KIOXIA AiSAQ Technology Integrated into Milvus Vector Database

Kioxia America, Inc. today announced that its AiSAQ™ approximate nearest neighbor search (ANNS) software technology has been integrated into Milvus (starting with version 2.6.4), among the world’s ...
PC Gamer

Anthropic reveals that as few as '250 malicious documents' are all it takes to poison an LLM's training data, regardless of model size

AI Microsoft CEO warns that we must 'do something useful' with AI or they'll lose 'social permission' to burn electricity on it AI Steam updates AI disclosure form to specify that it's focused on ...
Microsoft

Quality Over Quantity? LLM-Based Curation for a Data-Efficient Audio-Video Foundation Model

Integrating audio and visual data for training multimodal foundational models remains a challenge. The Audio-Video Vector Alignment (AVVA) framework addresses this by considering AV scene alignment ...
MIT Technology Review

OpenAI has trained its LLM to confess to bad behavior

Large language models often lie and cheat. We can’t stop that—but we can make them own up. OpenAI is testing another new way to expose the complicated processes at work inside large language models.
Computer Weekly

AWS recalibrates data economics further with S3 Vectors, batch & Intelligent-Tiering

AI needs data, AI needs inter (and intra) data repository contextual linking and AI needs all of that with a whole menu of search types spanning contextual search, semantic search, SQL search and now ...
TechCrunch

Gemini 3 refused to believe it was 2025, and hilarity ensued

Every time you hear a billionaire (or even a millionaire) CEO describe how LLM-based agents are coming for all the human jobs, remember this funny but telling incident about AI’s limitations: Famed AI ...
Searchenginejournal.com

New Data Finds Gap Between Google Rankings And LLM Citations

A new report compares Google rankings with citations from ChatGPT, Gemini, and Perplexity, showing different overlap patterns. Perplexity’s live retrieval makes its citations look more like Google’s ...