OpenAI Goes to School; Microsoft IDs AI Layoffs

OpenAI's Study Mode Integration: OpenAI is pushing its ChatGPT 'Study Mode' into educational platforms like Canvas, aiming to become a core part of learning infrastructure rather than just a homework helper. This move signals OpenAI's ambition to 'own the rails' of AI-powered education.

Microsoft's AI Layoff Wave: Microsoft's research into Copilot chats reveals AI is encroaching on jobs involving research, summarizing, persuasion, and information organization. This suggests a shift in enterprise budgets towards 'prompt-enabled productivity' tools, putting white-collar jobs 'on borrowed time' if they can be fit into a prompt.

Meta's AI-Enabled Interviews: Meta is piloting interviews where candidates can use AI assistants during coding tests. This reflects a pragmatic approach to the reality of AI in the workplace, aiming to identify engineers who can effectively work with AI tools rather than banning them. It's a move that challenges traditional interview methods and filters for future-fit talent.

Quick Hits:

• Adobe enhances Photoshop with new AI features for image manipulation.
• Anthropic is reportedly seeking $3-5 billion in funding at a $170 billion valuation.
• Google's NotebookLM now offers video summaries, and Google's AI Mode in Search includes a 'Canvas' planning tool.
• Yelp is using AI to create video summaries of restaurants.
• New AI launches include Magic Patterns (AI design platform), PodClips (podcast content tool), RunLLM (AI support assistant), and Jotform (Gmail AI assistant).
• Other trending tools: Genspark Super Agent, Heardly, CopyOwl, and Flot AI.

Source

Stanford’s Virtual AI Lab: Researchers at Stanford and Chan Zuckerberg Biohub have developed a “virtual lab” of AI scientists capable of designing, debating, and testing biomedical discoveries. This AI team, featuring an “AI principal investigator” and specialized agents, can produce lab-ready results in days, significantly accelerating scientific discovery. The AI lab even provides full transcripts of the AI team’s reasoning for human review.

Meta Targets Mira Murati’s Startup Talent: Meta is aggressively pursuing employees from ex-OpenAI CTO Mira Murati’s Thinking Machines Lab, offering substantial compensation packages (some exceeding $1 billion). This move highlights Meta’s determination to acquire top AI talent for its superintelligence team and its strategy to undercut rivals like OpenAI with open-source models. Despite the lucrative offers, no employees have accepted, indicating strong confidence in Thinking Machines Lab’s own trajectory.=

Tooling:

• 🎥 Wan2.2: Alibaba’s new open MoE model for AI video generation.
• 🤖 GLM-4.5: Z.ai’s new SOTA open-source agentic AI family.
• 📊 Shortcut AI: An AI agent for Excel spreadsheets.
• Microsoft is reportedly close to a deal to retain access to OpenAI’s tech even after the company’s AGI milestone.
• xAI has opened the waitlist for its upcoming “Imagine” image and video generation model.

OpenAI Prepares for GPT-5 Launch: OpenAI is reportedly set to release GPT-5 as early as August. This new AI system will integrate distinct models for various functions, including the o3 model, aiming to unify its o-series and GPT-series models. This signifies a major step in OpenAI’s efforts to create more cohesive and powerful AI systems.

AI Budgeting for Startups: Startups are advised to allocate 10-15% of their R&D budget to AI, considering engineer salaries and AI tool costs. While AI adoption varies, AI-native startups may spend more. Companies should adapt their budgets as AI becomes increasingly integrated into operations, focusing on prompt-enabled productivity.

The Three Layers of ROI for AI Agents: A three-layer framework for AI agent ROI highlights: 1) Labor efficiency, which, while easy to explain, doesn’t always translate to immediate realized ROI; 2) Net-new revenue, generated from tasks businesses couldn’t do before AI; and 3) Optimization, where AI models enhance decision fluency and ML improves decision precision. This framework provides a comprehensive view of AI agent value.

AI Writing Red Flags & Prompt Tips: Wikipedia editors have compiled a master list of AI writing “red flags” (e.g., “serves as a testament,” “rich cultural heritage”). This list can be used to refine AI prompts and outputs, transforming negative phrases into active, positive statements. This emphasizes the importance of careful prompt engineering and editing for high-quality AI-generated content.

Other Key Developments:

• Cursor’s Bugbot offers automated code review, catching bugs and security vulnerabilities with high resolution rates.
• Google’s Web Guide uses Gemini to intelligently cluster search results, improving information organization.
• Kimi K2 shows competitive results against Claude 4 Sonnet for agentic coding, offering a cost-effective open-source alternative.
• Gemini 2.5 Pro demonstrated the ability to solve complex mathematical problems, even winning gold at the IMO 2025, showcasing the potential of LLMs for advanced reasoning.
• Memories.ai introduced an advanced memory system for video AI, enabling persistent video chat across archives.
• TimeScope is a new open-source benchmark for testing vision-language models on long videos, revealing current limitations in temporal understanding.
• A black market for Nvidia B200 chips has emerged, with over $1 billion worth smuggled to China despite export controls.
• Google’s AI Overviews now serve 2 billion monthly users, with AI Mode reaching 100 million users in the US and India.
• Prem Qu Nair (ex-Windsurf) joined Cognition to work on software engineering.
• Opal, Google Labs’ new tool, allows users to build and share AI mini-apps using natural language.
• Incogni: Erases sensitive personal data from the web.
• Qwen3 30B A3B 2507: Delivers GPT-4o-level offline conversations on laptops, offering enterprise-quality AI without internet or per-use costs.
• Qwen3-235B-A22B-Thinking: Alibaba’s thinking AI for complex reasoning with transparent thought processes.

Source