In partnership with

Good morning. It’s Monday, May 12th.

On this day in tech history: 1997: A massive ransomware cyberattack, known as WannaCry, impacted over 200,000 computers across 140 countries. Originating from servers in China, it affected organizations like FedEx, the UK’s National Health Service, and Telefónica in Spain, highlighting vulnerabilities in global cybersecurity.

In today’s email:

• OpenAI Restructure Enables IPO

• Microsoft’s “Pages” AI

• New Google AI Features

• 4 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Microsoft, OpenAI Rewrite Partnership to Enable IPO and Secure Post-2030 AI Access

Microsoft and OpenAI are renegotiating the terms of their multibillion-dollar alliance as OpenAI seeks to convert its for-profit arm into a public benefit corporation and eventually go public. According to the Financial Times, a key sticking point is Microsoft’s equity stake in the restructured entity, which remains governed by OpenAI’s nonprofit board. Microsoft, having invested $13 billion, must approve the conversion. One proposal under discussion involves Microsoft trading a portion of its equity for extended access to OpenAI’s technologies beyond the current 2030 cutoff.

The talks also involve revisions to their 2019 contract. The balance of power has grown increasingly fraught as OpenAI’s enterprise offerings begin to encroach on Microsoft’s own AI product roadmap. Compounding matters is OpenAI’s Stargate initiative, a multibillion-dollar supercomputing project that signals an intent to scale infrastructure independently, potentially reducing reliance on Microsoft’s cloud stack. Read more.

Microsoft’s Pages brings ChatGPT-like collaboration without coding features

Microsoft has launched Pages, a new Copilot tool designed for collaborative writing and text refinement. Similar to ChatGPT’s Canvas, Pages allows users to directly edit AI-generated responses by selecting text and prompting modifications via an “Ask” button. While geared toward content iteration such as tweaking tone, expanding sections, or rewording ideas, it currently lacks coding capabilities.

Pages supports task-specific prompts like structured guides, goal-oriented schedules, or step-by-step planning, all formatted for checklists or time-based tracking. Read more.

Google's Creative Tools Get Smarter

At the Game Developers Conference, Google unveiled its Gemma 3 model, enabling local, multimodal AI integration in games. Developers can use it for dynamic NPC interactions, with a Unity plugin for real-time dialogue. Google also highlighted Gemini 2.0 for cloud-based AI coaching in larger games.

 New NotebookLM Features:

Google’s NotebookLM is adding a “Video Overviews” feature, powered by Veo 2, to convert notebook content into short video summaries. The update also includes a new “Editor’s Picks” section for community-driven content. These features will launch alongside the mobile version for iOS and Android.

Neuron Synchronization Powers Sakana AI’s New Thought Machine

Sakana AI has launched the Continuous Thought Machine (CTM), a neural network that integrates neuron synchronization and timing to enhance decision-making. Unlike traditional AI models, which focus on single neuron outputs, the CTM leverages the precise timing of neuron activity, allowing for complex, multi-step reasoning. The model’s performance is demonstrated in tasks like maze-solving and image classification, where it exhibits human-like behavior, such as path-following in mazes and focusing attention on relevant image features.

The CTM’s asynchronous nature allows it to reason about both static and sequential data effectively. The model’s behavior is emergent, based on neuron dynamics and synchronization, offering improved interpretability and efficiency in problem-solving tasks. This approach moves closer to replicating brain-like processes, marking a step forward in AI's integration with biological insights. Read more.

• Stanford Finds AI Learns Better from Successes

• Prompt Design Affects Hallucinations

• Alibaba’s Qwen Web Dev Automates Front-End Creation

• Alibaba Trains AI Search Without the Internet

• Tencent's Open-Source Tool Powers Multi-Subject Video Creation

• Bytedance’s Agent TARS Handles Real-Time, Complex MacOS Tasks

• Junior lawyers would ‘riot’ if AI tool taken away: PwC partner

• Deepseek-R1 triggers boom in reasoning-enabled language models

• Digital marketing used to be about clicks, but the rise of ChatGPT means it’s ‘now all about winning the mentions’

• Nearly 20 million jobs are on the chopping block to be replaced by AI, SHRM research shows

• Microsoft study breaks down tradeoffs between API and GUI agents

• China's Baidu looks to patent AI system to decipher animal sounds

• New Lego-building AI creates models that actually stand up in real life

• Pope Leo XIV lays out vision of papacy and identifies AI as a main challenge for humanity

• SoundCloud changes policies to allow AI training on user content

• Honor Phones Launch AI Video Tool Powered by Google’s Veo 2

• Amazon offers peek at new human jobs in an AI bot world

• Meta Is recruiting former Pentagon officials as it ramps up military ambitions

📄 G-FOCUS: Towards a Robust Method for Assessing UI Design Persuasiveness

📄 Perception, Reason, Think, and Plan: A Survey on Large Multimodal Reasoning Models  ​​

📄 Scalable Chain of Thoughts via Elastic Reasoning

📄 Sentient Agent as a Judge: Evaluating Higher-Order Social Cognition in Large Language Models

📄 FG-CLIP: Fine-Grained Visual and Textual Alignment

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!