ChatGPT can now interact with your code, the US government proposes AI framework
what to know for now
🚀 ChatGPT expands Mac app integration. The new “Work with Apps” feature enables ChatGPT to read code from developer tools like VS Code, Xcode, and Terminal via the macOS accessibility API, automating context sharing without copy-paste. Currently limited to reading code, it doesn’t write directly into apps, but OpenAI aims to develop agentic systems and extend support to other text-based applications. Read more
🐦 AI criticizes Elon Musk. The AI identifies Musk as a leading source of misinformation on X, citing his high visibility and follower count as factors that amplify his posts. The analysis covers various topics, including politics, health issues, and conspiracy theories, and acknowledges the subjective nature of misinformation definitions. Read more
🧬 AlphaFold3 opens up. DeepMind has released the AlphaFold3 software code for non-commercial use, allowing scientists to model protein interactions with DNA and other molecules. Access to training weights is currently limited to academic researchers upon request, facilitating advancements in drug discovery and protein research. Read more
📈 Palantir transfers to Nasdaq. Palantir will begin trading on Nasdaq Global Select Market on November 26, following a successful transfer from NYSE. The company's stock has surged over 196% in the past year, driven by strong AI demand and robust quarterly results. Read more
🧪 AI Research of the Week
Large Language Models Can Self-Improve in Long-context Reasoning
From The Chinese University of Hong Kong and TencentJake’s Take: This paper presents SEALONG, a method to enable models to self-improve their long-context reasoning capabilities without relying on human or advanced model annotations. Experiments show significant improvements (especially on Llama models) suggesting self-supervision can effectively enhance long-context performance.
This could help improve efficiency in AI training, reducing dependency on human intervention, which could help accelerate future model advancements.
what to know for later
💻 OpenAI prepares 'Operator' AI agent. Scheduled for January release, it will be available as a research preview and via API for developers. "Operator" can execute tasks on users' computers, including writing code or booking travel. Read more
🧠 AGI predicted soon by CEOs. Sam Altman forecasts AGI within five years with minimal societal disruption using current hardware, while Dario Amodei targets a 2026-2027 breakthrough based on AI model advancements. OpenAI faces financial challenges despite significant investments, highlighting data and funding shortages in AI development. Read more
🛡️ DHS unveils AI framework. The framework targets AI-driven attacks, system vulnerabilities, and design flaws, offering guidelines for cloud providers, developers, and infrastructure operators. It promotes international standard harmonization while emphasizing domestic implementation across critical sectors. Read more
🏛️ OpenAI outlines gov’t AI partnership. The blueprint proposes AI economic zones and a National Transmission Highway Act to enhance energy infrastructure. It advocates for a North American AI Compact to compete with China and leverages naval nuclear resources for reindustrialization. Read more