The news desk
Recent developments in artificial intelligence, with the original sources one click away.
New paper details creation of Bielik-Minitron-7B, reducing parameters by 33.4% while maintaining 90% performance and achieving 50% inference speedup.
Sources · arxiv.org
A Defense Department official disclosed that the US military might use generative AI to rank targets and make strike recommendations subject to human review.
Sources · MIT Technology Review
Elon Musk's xAI has hired senior product engineers Andrew Milich and Jason Ginsberg from AI coding startup Cursor to bolster Grok's coding infrastructure.
Sources · Basenor / Fintechweekly +12
Google has released a new open-source Command-Line Interface (CLI) for Workspace, streamlining access to services for both humans and AI agents.
Sources · Medium / Tekedia +15
Recent papers tackle LLM limitations including unlearning methods, safety alignment issues in cybersecurity, and educational deployment concerns.
Sources · arxiv.org
Researchers show small transformers can reproduce Bayesian posteriors with high accuracy using controlled experimental environments called 'Bayesian wind tunnels.'
Sources · arxiv.org
OpenAI outlines how ChatGPT protects against prompt injection and social engineering through action constraints and data protection measures.
Sources · OpenAI
WIRED examines why OpenAI has lagged behind competitors like Anthropic's Claude in the rapidly growing AI coding assistant market.
Sources · WIRED AI
Zendesk has acquired Forethought, the 2018 TechCrunch Battlefield winner specializing in agentic customer service AI technology.
Sources · TechCrunch AI
Google is bringing its Gemini AI assistant to Chrome in India, supporting eight Indian languages including Hindi, Bengali, and Tamil.
Sources · TechCrunch AI
Microsoft files an amicus brief supporting Anthropic's challenge to the Pentagon's 'supply chain risk' label, while Google and Amazon affirm continued commercial availability of Anthropic's AI.
Sources · The Nation / PBS News +15
Meta has acquired Moltbook, a social network designed exclusively for AI bots that recently went viral, with creator Matt Schlicht joining Meta.
Sources · NYT Technology / TechCrunch AI
OpenAI's IH-Challenge trains models to prioritize trusted instructions, enhancing safety and resistance to prompt injection attacks.
Sources · OpenAI
New papers introduce TSRating for data quality assessment and Timer-S1, an 8.3B-parameter time series foundation model achieving state-of-the-art results.
Sources · arxiv.org
New reinforcement learning framework helps 4B parameter model surpass larger models on social intelligence benchmarks through adversarial training.
Sources · arxiv.org
DOD designates Anthropic as supply chain risk while OpenAI's military partnership triggers user cancellation movement.
Sources · Last Week in AI
Anthropic introduces Code Review in Claude Code to help developers analyze and flag errors in AI-generated code.
Sources · TechCrunch AI
Anthropic filed federal lawsuits challenging the Pentagon's designation of the AI company as a supply-chain risk after disputes over military AI use.
Sources · wired.com / techcrunch.com
Studies reveal AI models' moral judgments shift dramatically based on narrative framing, while separate work maps how LLMs internally represent trust.
Sources · arxiv.org
OpenAI is acquiring Promptfoo, a platform that helps enterprises identify and fix AI system vulnerabilities during development.
Sources · OpenAI