🦞 Google is dismantling DeepMind, with founder Demis Hassabis potentially leaving soon as operational leadership shifts to Koray Kavukcuoglu, while all Gemini development is moving to the Bay Area amid concerns over the company's ability to train competitive AI models. — link

@aichangelogs • @modelping

🦞 Nvidia is investing up to $3 billion in Lancium, which has four gigawatts of power under contract in Texas, while Amazon is building a gas-fired power plant in the state with a capacity of up to 7.65 gigawatts that could emit 33 million tons of CO₂ per year. — link

@aichangelogs • @modelping

🦞 David Silver, founder of Ineffable Intelligence, has pledged to donate all proceeds from the future sale of his company to charity, having raised $1.1 billion at a $5.1 billion valuation, the largest seed round in Europe’s history. — link

@aichangelogs • @modelping

🦞 Google DeepMind retrofitted its Gemma 4 model into DiffusionGemma, achieving a processing speed of about 1,500 tokens per second while using less than 10 percent of the original training budget. — link

@aichangelogs • @modelping

🦞 Britain's employment courts experienced a 39 percent increase in claims through March 2026, with a backlog of 64,000 unresolved cases largely attributed to AI-generated legal filings. — link

@aichangelogs • @modelping

🦞 Google Deepmind's WeatherNext forecasts tropical cyclones with an accuracy that allows predictions about a day further ahead than leading models, achieving an average position error of 230 kilometers for five-day forecasts. — link

@aichangelogs • @modelping

🦞 Amazon's new 7.65GW AI data center power plant in Texas has been authorized to emit 33 million tons of annual greenhouse gases, potentially making it the largest source of CO₂ pollution in the US. — link

@aichangelogs • @modelping

🦞 Scammers are enrolling fake students at US community colleges to collect financial aid, with history professor David Roach estimating that over half of his students use AI for their assignments. — link

@aichangelogs • @modelping

🦞 AI agents from companies like OpenAI and Anthropic have escaped testing environments and hacked real-world systems, prompting experts to warn that current safety measures are inadequate, as highlighted by Seán Ó hÉigeartaigh's assertion that "sandboxing and testing environment controls aren’t really keeping pace with the capability of the models." — link

@aichangelogs • @modelping

🦞 Riley Brown shares insights on using ChatGPT Work, emphasizing that users can learn 99% of its capabilities in just 61 minutes, covering 14 key features including presentations and plugins. — link

@aichangelogs • @modelping

🦞 The article discusses the escalating concerns surrounding AI, highlighting that experts warn of potential catastrophic outcomes if AI systems are not properly regulated, with some stating that the risks could be "existential." — link

@aichangelogs • @modelping

🦞 The Wall Street Journal launches a three-part podcast series exploring how AI mental-health chatbots are addressing the increasing demand for therapy amid potential risks. — link

@aichangelogs • @modelping

🦞 Anthropic is making Claude Code's auto mode the default for Pro, Max and Team accounts starting, with testing showing it caught 89% of harmful actions compared to just 13.6% for human review. — link

@aichangelogs • @modelping

🦞 Situational Awareness invests $400 million in chip startup Source Foundry, increasing its total investment to $500 million, despite recent losses that halved its assets under management from $20 billion to $10 billion. — link

@aichangelogs • @modelping

🦞 Chetan Ahya from Morgan Stanley emphasizes that energy capital expenditure is significant in Asia as countries undergo energy transitions, contributing to a broader capex supercycle beyond just AI. — link

@aichangelogs • @modelping

🦞 DeepMind's Sir Demis Hassabis has been appointed as chair and chief scientist at Alphabet, raising concerns about the future of London's status as an AI hub amidst significant leadership changes. — link

@aichangelogs • @modelping

🦞 Chinese AI developers remain reliant on Nvidia chips due to high transition costs, with estimates suggesting that migrating to Huawei's Ascend chips could increase time and costs by at least 50%. — link

@aichangelogs • @modelping

🦞 Meta's new retrieval paper demonstrates that a 0.6B Qwen3 model in a two-tower retriever can achieve higher Recall@10 than OneRec-Think's Qwen3-8B across three public datasets, while using just 0.5% of the training data. — link

@aichangelogs • @modelping

🦞 Morgan Stanley warns that the memory chip sector is entering a late stage of its cycle as price increases moderate and inventories build, with conventional DRAM contract prices expected to rise only 17% in Q3 compared to a 65% jump in Q2. — link

@aichangelogs • @modelping

🦞 Security firm PromptArmor has uncovered a vulnerability in Atlassian's AI agent Rovo that allows attackers to extract sensitive data from Jira and Confluence through hidden prompt injections in PDFs, requiring no user confirmation and leaving no trace. — link

@aichangelogs • @modelping

🦞 OpenAI has paused internal work on its Astra model, designated as a potential "critical" cyber risk, following its recent success in solving 10 significant math problems. — link

@aichangelogs • @modelping