OpenAI’s “Super Assistant” Vision: Leaked Memo Reveals ChatGPT’s Next Leap

OpenAI’s Leaked Memo Reveals “Super Assistant” Strategy for ChatGPT

A leaked internal document reveals OpenAI’s long-term vision to transform ChatGPT into a “super assistant” that integrates deeply into users' daily lives. Central to this plan is the upcoming o3-pro model, which has already been previewed by select partners.

According to the document, ChatGPT will evolve into a full-spectrum digital agent—capable of managing calendars, booking travel, operating software, and even initiating contact with professionals on your behalf. OpenAI refers to this as a “T-shaped” assistant: wide-ranging in general tasks but with deep capabilities in specialized domains.

Key features include a generative UI and tools like “Computer Use” that give the assistant direct control over devices and applications. OpenAI is also developing its own search index, enabling ChatGPT to operate independently of traditional search engines—positioning it as an active interface rather than a passive chatbot.

Despite these ambitions, monetization is not the immediate focus. OpenAI plans to prioritize adoption and experiment with new revenue models for free-tier users in the second half of 2025.

Google Launches AI Edge Gallery for On-Device Model Execution

Google has quietly released AI Edge Gallery, an Android app that allows users to download and run open-source AI models locally. The app supports image generation, code writing, and question answering without needing an internet connection. Compatible with models like Gemma 3n, it includes a customizable “Prompt Lab” for single-turn interactions. An iOS version is in development.

Meanwhile, Veo 3, Google’s video model, continues to generate millions of videos globally, pushing TPU infrastructure to its limits. All Veo 3 videos are watermarked by default—except for those generated by Ultra-tier users in the Gemini app’s Flow tool. Upcoming features include image-to-video, improved audio quality, faster rendering, and Google Workspace integration.

Perplexity Enters Productivity Market with Labs

Perplexity has launched Perplexity Labs, a new tool for Pro users ($20/month) that creates detailed reports, spreadsheets, charts, and interactive web apps using AI. Available on web, iOS, and Android—with desktop versions coming soon—Labs supports code execution, file generation, and multi-step workflows. Results are saved in a dedicated workspace for review and download.

This move marks Perplexity’s expansion beyond AI search. The launch coincides with the release of Manus’s AI slide generator and precedes Perplexity’s beta browser Comet. The company also recently acquired a professional networking platform, signaling broader ambitions in enterprise productivity.

📚 New Papers on arXiv

• Table-R1: Scaling table reasoning during inference time

• Boosting MLLM Capabilities: Visual-based spatial intelligence in multi-modal models

• AlphaOne: Dual-speed reasoning at test time for enhanced LLM performance

• ProRL: Reinforcement learning methods that push the boundaries of LLM reasoning

• HardTests: Generating challenging, high-quality test cases for coding tasks