Latest developments in artificial intelligence, including large language models (LLMs), generative AI, multimodal AI, AI agents, robotics, and AI regulation and policy.
The autonomous AI penetration testing tool "Strix" has been unveiled, featuring dynamic code execution for provable vulnerability detection and CI/CD integration, gaining attention as an alternative to manual penetration testing.
Google expands Gemini's personalized AI image generation feature to free users in the US, allowing images to be created using user interests and linked Google account data.
Microsoft withdraws Claude Code license amid skyrocketing token costs and underwhelming output. A wake-up call for AI cost management in businesses.
A comprehensive test of Google Gemini's extensions reveals that while smart home control might justify switching from Google Assistant, most integrations lack practicality.
Ford rehires 350 veteran engineers following disappointing results from automated quality systems, aiming to cut $1 billion in costs.
Zhipu AI’s GLM-5.2 demonstrates performance on par with Anthropic’s Mythos in cybersecurity tasks, highlighting a narrowing US-China AI gap.
With advanced mathematical reasoning capabilities, AI challenges the role of mathematicians and the meaning of mathematics itself. UCLA's Terence Tao advocates for a transformative collaboration between humans and AI in what he calls "Grand Mathematics."
Four key insights for successful AI adoption in enterprises, drawn from a discussion with Chinese AI experts: upgrading executive awareness, selecting appropriate implementation scenarios, organizing business processes with ontology, and moving beyond hierarchical structures.
The JaiLIP method developed by Florida International University researchers bypasses multimodal AI safety through seemingly harmless image manipulation. Verification with BLIP-2 showed nearly double the harmful outputs.
SoftBank Group CEO Masayoshi Son has expressed doubts about Elon Musk's proposed orbital data center concept. As demand for AI computation resources surges, discussions on the feasibility and cost-effectiveness of space infrastructure are intensifying.
Researchers from Queen Mary University of London analyzed over 380,000 song lyrics from 1960 to 2023, finding a decline in words reflecting moral virtues and an increase in vocabulary associated with harm, corruption, and subversion.
OpenAI has unveiled the GPT-5.6 suite, featuring three models: flagship Sol, mid-tier Terra, and fast, cost-efficient Luna. Priced at half of competitors, the models emphasize enhanced safety and were released shortly after a regulatory request from the Trump administration.
Major tech companies like OpenAI, Google, Apple, and SpaceX are diving into in-house AI chip development. Analyzing the shift from reliance on single suppliers and its impact on the industry, including OpenAI’s collaboration with Broadcom on the "Jalapeño."
Google has unveiled a new policy document advocating for a balanced approach to AI regulation, proposing the establishment of an industry-funded independent body called FARO. The proposal highlights contradictions with past demands for stringent oversight.
A 10-year-old in Shanghai created a Pac-Man-style game using the AI platform "Linqiu" and won the gold prize for creativity in an AI contest.
A Tesla Model 3 using FSD crashed into a house in Texas, killing a 76-year-old woman. The lawsuit against Tesla could set a legal precedent for autonomous driving liability.
Patronus AI, a company specializing in evaluating the reliability of AI agents, has raised $50 million. Their "Digital World Models" technology, which detects shortcuts and flaws in agents, is gaining attention.
AI Berkshire, now available on GitHub, uses Claude Code to replicate the value investing methods of Warren Buffett and three other experts. Featuring parallel analysis and anti-bias mechanisms, the AI reportedly achieved an annual return exceeding 66% in live operations.
DeepSeek secures over 50 billion yuan in its first external funding round. Most of the funds will be invested in its AI computing centers as it aims to rival Anthropic in AI coding and Agent strategies.
The rapidly growing "Anthropic-Cybersecurity-Skills" repository on GitHub maps 762 practical security skills to six industry frameworks, enabling AI agents to achieve senior analyst-level capabilities.