The news desk
Recent developments in artificial intelligence, with the original sources one click away.
Researchers propose methods to maintain LLM safety during fine-tuning, enable inverse reasoning for safer outputs, and improve agent training through environment tuning.
Sources · arXiv AI (cs.AI)
Recent arXiv papers explore emergent misalignment in fine-tuned models, safety probing in reward systems, and intellectual property protection for LLMs.
Sources · arXiv AI (cs.AI)
Jensen Huang called reports of stalled collaboration with OpenAI 'nonsense,' pushing back on claims of tensions between the companies.
Sources · TechCrunch AI
AWS launches Nova LLM-as-a-Judge evaluation tool on SageMaker AI to assess generative model outputs beyond traditional statistical metrics.
Sources · Amazon AWS AI
Anthropic adds plug-ins to Cowork, allowing teams to customize Claude's workflows, tools, and commands for consistent outcomes.
Sources · TechCrunch AI
Amazon Bedrock now supports global cross-region inference for Claude 4.5 models, enabling AI deployment in South Africa with automated request routing.
Sources · Amazon AWS AI
WIRED's hands-on testing of Google's Auto Browse feature reveals the AI agent can handle some tasks but falls short of seamless automation.
Sources · WIRED AI
District Judge Sidney H. Stein upheld an order requiring OpenAI to turn over 20 million anonymized ChatGPT conversation logs to news organization plaintiffs.
Sources · news.bloomberglaw.com
Grok faces criticism for generating inappropriate content while Claude Code demonstrates strong technical capabilities across diverse tasks.
Sources · MIT Technology Review
The AI agent formerly known as MoltBot and Clawdbot has rebranded to OpenClaw, marking its third name change amidst viral adoption and escalating security concerns.
Sources · Forbes / DEV Community +12
OpenAI has built an in-house AI data agent that leverages GPT-5, Codex, and memory capabilities to analyze large datasets and generate insights quickly.
Sources · OpenAI
The AI assistant, initially Clawdbot, then Moltbot, has rebranded to OpenClaw, emphasizing open-source infrastructure and enhanced security.
Sources · Clawd to Moltbot to OpenClaw: one week, three names, zero chill - Medium / From Moltbot to OpenClaw: When the Dust Settles, the Project Survived - DEV Community +7
Recent arXiv preprints examine proactive questioning in reasoning models, cost-effective inference strategies, and LLM evaluation for homelessness policy.
Sources · arXiv AI (cs.AI)
Recent arXiv papers explore latent reasoning, computational efficiency, and explainability challenges in LLM chain-of-thought prompting.
Sources · arXiv AI (cs.AI)
San Francisco startup Logical Intelligence takes a different path to AGI while tech giants invest heavily in large language models.
Sources · WIRED AI
Google launches Chrome feature that uses generative AI to autonomously navigate and browse websites on behalf of users.
Sources · WIRED AI
Three arXiv papers investigate how multiple AI models work together, revealing challenges with current benchmarks and collaboration methods.
Sources · arXiv AI (cs.AI)
OpenAI announces plans to test advertising in ChatGPT while dealing with significant financial burn rate, according to Last Week in AI podcast.
Sources · Last Week in AI
AI agents increasingly targeted through prompt injection attacks, with incidents involving Gemini and Claude systems cited as examples.
Sources · MIT Technology Review
Researchers publish papers on demonstration-free robot control, multi-agent memory systems, and improved reasoning model training methods.
Sources · arXiv AI (cs.AI)