François Chollet's New AGI Lab

On a stormy evening in Silicon Valley, a group of researchers huddled around a glowing screen, their excitement palpable. Each was driven by a shared vision: a world where AI not only assisted but truly partnered with humanity in unlocking the mysteries of the universe. At the heart of their mission was Ndea, a trailblazing lab daring to go beyond traditional deep learning. They envisioned an artificial general intelligence capable of intuitive leaps and formal reasoning, one that could revolutionize self-driving technology, drug discovery, and sustainable energy.

Meanwhile, across the tech landscape, OpenAI quietly rolled out a game-changing addition to ChatGPT: Tasks. It wasn’t just about smarter conversations anymore; this was personal AI at its most practical. Birthdays remembered, weather updates sent, and recurring reminders—all scheduled seamlessly. This was a step toward making AI not just an assistant but an indispensable life companion.

On the other side of the innovation spectrum, MiniMax was making waves of its own. Their MiniMax-01 series, featuring a groundbreaking Lightning Attention architecture, promised to redefine long-context understanding in AI. With the capacity to process millions of tokens and a staggering 456 billion parameters, MiniMax’s models weren’t just tools—they were a bold invitation to developers worldwide to dream bigger and build smarter.

The stage was set, the players in motion, and the possibilities endless. This was the dawn of a new chapter in AI, where boundaries blurred and imagination thrived.

Let’s dive in.

Ndea: Pioneering AGI for Unprecedented Scientific Advancement

Ndea is an AI research and science lab focused on developing advanced AI systems that integrate intuitive pattern recognition with formal reasoning. The organization believes that the future of scientific progress relies on artificial intelligence capable of independent invention and discovery, moving beyond the limitations of current deep learning technologies. Ndea advocates for a new paradigm called program synthesis, which seeks discrete programs that can explain observed data, enhancing generalization and learning efficiency. By combining program synthesis with deep learning, Ndea aims to create artificial general intelligence (AGI) that can drive unprecedented scientific advancements across various fields, including self-driving vehicles, drug discovery, and sustainable energy. The lab is committed to building a highly skilled team to tackle the challenges of AGI and invites collaboration to explore the unknown frontiers of science and technology.

ChatGPT Adds Task Scheduling and Reminder Features

OpenAI has introduced a new feature in ChatGPT called "Tasks," which allows users to schedule reminders and recurring tasks. This beta function is currently available to Plus, Pro, and Team subscribers. With Tasks, users can instruct ChatGPT to perform actions at specified times, whether it's a one-time reminder or a recurring task, such as daily weather updates or reminders for important dates like birthdays. To use this feature, subscribers need to select the "ChatGPT with scheduled tasks" option from the model selection menu. Users can create tasks by typing out their requests along with the desired timing. Notifications will inform users when tasks are completed, and they can manage their tasks through a dedicated section in the profile menu. However, there is a limit of 10 active tasks at any time, and the feature may still be in the fine-tuning phase as OpenAI collects user feedback.

Open-Source Models with Lightning Attention

MiniMax has announced the open-source release of its MiniMax-01 series, which includes two models: MiniMax-Text-01, a foundational language model, and MiniMax-VL-01, a visual multi-modal model. This series features an innovative Lightning Attention architecture, allowing for a remarkable 456 billion parameters and the capability to process up to 4 million tokens, significantly more than other leading models. The design aims to support the development of AI agents by facilitating sustained memory and extensive inter-agent communication. MiniMax is offering competitive pricing for its APIs, set at $0.2 per million input tokens and $1.1 per million output tokens. The models are available on GitHub and will continue to receive updates, with the goal of inspiring further research in long-context understanding and enhancing ongoing innovations in AI technology.

Hand Picked Video

In this video, we explore ChatGPT's new Tasks feature - a game-changing automation tool that lets you schedule AI-powered tasks to run automatically.

Top AI Products from this week

• Persana AI Sales and Voice Agents - Persana's AI-powered GTM and voice agents supercharge sales and marketing teams with 24/7 pipeline generation and enrichment. Access 700M+ contacts and 300M+ companies along with 75+ data provider and transform your revenue with digital workers.

• Browser Use - Let AI control your browser. Say what you need, and browser-use gets it done:

• Roundups - Use AI to research products and generate guides that influence buying decisions. Publish and share your Roundups and earn commissions when readers click through and make purchases.

• Doctor Droid - Doctor Droid simplifies the process of troubleshooting your Kubernetes (k8s) cluster. With just a message to the bot, you can quickly check pod statuses, restart pods, run custom commands, or diagnose problems you may find challenging to debug on your own.

• DryMerge - DryMerge’s AI agents connect to your apps and automate tasks. Describe your task in chat - e.g. “When I get an email invoice, add it to a sheet & send me a weekly summary of new expenses” and DryMerge builds a custom agent to automate it in the background.

• Google AI Studio - Google AI Studio is the fastest way to start building with Gemini, our next generation family of multimodal generative AI models.

• DeepSeek for iOS - Powered by the groundbreaking DeepSeek-V3 model with over 600B parameters, this state-of-the-art AI leads global standards and matches top-tier international models across multiple benchmarks.

This week in AI

• Seaweed-APT: One-Step Video Generation - Seaweed-APT introduces a model for real-time video generation, producing 1280x720 videos in one step, significantly faster than traditional methods.

• Microsoft 365 Copilot Chat - Microsoft introduces Copilot Chat for Microsoft 365, enhancing productivity with AI-driven assistance across apps, streamlining workflows and collaboration.

• Contextual AI Platform Launch - Contextual AI announces the general availability of its platform, enabling businesses to leverage advanced AI capabilities for enhanced decision-making and insights.

• Luma AI's Genie 1.0 and Series B - Luma AI launches Genie 1.0, a generative 3D model for creating objects in under 10 seconds, alongside a $70 million Series B funding round to enhance multimodal AI capabilities.