Good morning. It’s Monday, June 2nd.

On this day in tech history: 2014: Apple introduced Swift, a statically-typed, LLVM-compiled language for iOS/macOS apps, featuring type inference, null-safe optionals, and ARC. Up to 2.6x faster than Objective-C, its safe, readable syntax and 2015 open-source release (Apache 2.0) with SwiftUI support revolutionized scalable, cross-platform app development.

In today’s email:

• OpenAI’s internal doc leak

• Google releases an app that lets you run AI models locally

• Perplexity pushes into enterprise tools with Labs release

• 5 New AI Tools

• Latest AI Research Papers

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

OpenAI wants ChatGPT to be a ‘super assistant’ for every part of your life

OpenAI is set to release the o3-pro model soon, having already alerted select customers to its imminent release. This advancement forms a key part of a larger strategy detailed in a leaked internal document. It sketches out a future where ChatGPT, powered by the upcoming o3 model, becomes a full-spectrum operator: managing calendars, booking travel, navigating software, and even contacting professionals on your behalf.

It refers to the assistant’s capabilities as “T-shaped”: broad enough to manage daily tasks, yet deep enough to handle specialized work. A generative UI is paired with tools like “Computer Use,” giving the assistant hands-on control over devices and applications. OpenAI is also building its own search index, letting ChatGPT sidestep traditional engines and act more like a one-stop interface than a passive responder. Despite redactions, the memo further indicates that hardware integration remains central to this direction.

For now, monetization takes a back seat, with OpenAI prioritizing usage growth to trial new revenue models in the second half of 2025, including potential business models for free users. Read more.

Google quietly released an app that lets you download and run AI models locally

Google AI Edge Gallery enables Android users to download and run open-source AI models locally, using on-device processing to perform tasks such as image generation, code writing, and question answering without requiring an internet connection. Compatible with models like Gemma 3n and featuring a configurable “Prompt Lab” for single-turn prompts, the app’s performance depends on device hardware and model size, with an iOS version forthcoming.

As Veo 3 rapidly generated millions of AI videos across 71 countries, driving TPU infrastructure to capacity, Google emphasized responsible AI use by watermarking all Veo 3 videos, except for Ultra-tier users of the Flow tool, which supports specialized workflows in the Gemini app on Android and iOS. Planned updates will introduce image-to-video conversion, improved audio fidelity, faster rendering, and integration with Google Workspace, underscoring Google’s commitment to combining local AI capability with scalable cloud-based production tools. Read more.

Perplexity pushes into enterprise tools with Labs release

Perplexity has launched Perplexity Labs, a productivity tool for Pro subscribers ($20/month) that generates reports, spreadsheets, dashboards, and interactive web apps using AI. Available on web, iOS, and Android, with desktop apps to follow, Labs handles extended tasks, using features like code execution, chart creation, and file generation. Results, including images and documents, are organized in a dedicated workspace for download or review.

The launch signals Perplexity’s pivot beyond search, arriving the same day as Manus debuted its AI slide generator. Labs supports data structuring, formula application, and custom workflows, all powered by an AI system that typically takes 10 minutes or more to complete a task. The company is also testing a web browser (Comet) and recently acquired professional social platform

This broader product push aligns with Perplexity’s corporate focus, following the launch of an enterprise plan and reported efforts to raise up to $1B at an $18B valuation. Read more.

• AI meets game theory: How language models perform in human-like social scenarios

• As Revenue Triples, Anthropic Expands Voice Features and Floats AI Usage Tax

• AI Takes the Lead in Meta’s Risk Assessments

• Home robotics is going to work in 2026

• Dubai Attracts Global AI Talent with Tax Perks, High Pay, and Golden Visas

• Robot navigates high-speed parkour with autonomous movement planning

• Early AI investor Elad Gil finds his next big bet: AI-powered rollups

• QwenLong-L1 solves long-context reasoning challenge that stumps current LLMs

📄 Table-R1: Inference-Time Scaling for Table Reasoning

📄 Boosting MLLM Capabilities in Visual-based Spatial Intelligence ​​

📄 AlphaOne: Reasoning Models Thinking Slow and Fast at Test Time

📄 ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models

📄 HardTests: Synthesizing High-Quality Test Cases for LLM Coding 

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!