Inside Mind Lab: The Team Teaching AI to Learn Like a Human

In the world of Artificial Intelligence, headlines are usually dominated by numbers. We hear about “Trillion Parameters,” “Thousands of GPUs,” and “Billion-dollar clusters.”

But at Macaron AI, we believe the most important element of intelligence isn’t the hardware. It’s the philosophy.

Behind the colorful interface of the Macaron app lies a powerhouse research division called Mind Lab. They are the ones who made it possible to generate an app in 2 minutes. They are the ones who figured out how to give an AI “Deep Memory.”

But who are they? And why do they believe the rest of the industry is looking at AI the wrong way?

We take you inside Mind Lab to meet the team that is teaching machines not just to compute, but to experience.

The “All-Star” Roster

Mind Lab isn’t a typical startup engineering team. It is a research squad with a pedigree that rivals the world’s biggest laboratories.

The 10-person core research group is composed of veterans from industry giants like OpenAI, DeepMind, and ByteDance’s Seed team. These are the engineers who helped build the foundational systems that launched the current AI wave.

Leading the charge are researchers with deep academic roots. The team includes alumni from Tsinghua University (Pony and Qihan), Cornell University (Alex), and is founded by Andrew, an MIT alumnus who currently serves as a Research Director at the Tsinghua Shenzhen Research Institute.

This isn’t a team that just met at a hackathon. Their collaboration began more than 10 years ago. Together, they have co-authored over 200 research papers with more than 30,000 citations across reinforcement learning and large-scale optimization.

They have “been there, done that” with pre-training. Now, they are looking for what comes next.

The Philosophy: Real Intelligence is Continual Learning

The prevailing dogma in Silicon Valley is “Scaling Laws”—the idea that if you just feed a model more data and buy more GPUs, it will get smarter.

Mind Lab disagrees. They align with the view recently expressed by Ilya Sutskever: “Real Intelligence is Continual Learning”.

Think about how a human learns. You didn’t learn to ride a bike by reading 10,000 books about bicycles (Pre-training). You learned by getting on the bike, falling off, and adjusting your balance (Reinforcement Learning from Experience).

This is Mind Lab’s ethos: “Real intelligence learns from real experience”.

They believe that simply throwing more static data at a model has hit diminishing returns. To make Macaron truly smart—smart enough to plan a complex trip or understand your mood—it needs to learn from feedback loops. It needs to try, receive feedback (from you, the user), and improve.

The Breakthrough: Efficiency is the New Scale

This philosophy led to Mind Lab’s most significant technical breakthrough: LoRA-Native Reinforcement Learning.

Training a massive AI model usually requires a budget that only Google or Microsoft can afford. But Mind Lab wanted to bring this power to a personal agent.

They became the first team in the world to achieve high-performance Reinforcement Learning (RL) on the largest open-source 1-trillion-parameter model using a technique called LoRA (Low-Rank Adaptation).

The result? They managed to train this massive model while consuming only 10% of the typical GPU budget.

This isn’t just a cost-saving measure; it’s a revolution in accessibility. It means Macaron can update its models faster, iterate on features like “Mini-app Generation” quicker, and offer a smarter agent without charging users enterprise-level fees.

Giving Back to the Community

Mind Lab knows that secret sauce shouldn’t stay secret forever.

In a move that surprised many, they open-sourced their core algorithms. They merged their efficiency techniques into foundational infrastructures including NVIDIA Megatron-LM and ByteDance Seed Verl.

Why? Because they believe that for AI to truly “Live Better” with humans, the infrastructure needs to be transparent. They want to set the standard for how models learn from experience, proving that you don’t need a walled garden to build world-class tech.

What This Means for You

When you use Macaron to generate a Fitness Tracker or chat with the Group Agent, you aren’t just using a clever app. You are using a product built on top of a frontier research stack.

You are benefiting from a decade of collaboration between top scientists who decided that the future of AI wasn’t about building a bigger brain in a jar—it was about building a brain that learns, adapts, and grows along with you.