MIT unveils SEAL, a self-improving AI model

SEAL, a self-improving AI model developed by MIT, creates its own training data to improve its performance without human intervention.
SEAL, a self-improving AI model from MIT, autonomously generates training data and fine-tuning instructions to improve its performance.

Researchers at the Massachusetts Institute of Technology (MIT) have unveiled SEAL, a new AI model capable of improving its own performance without human intervention. The framework allows the model to generate its own training data and fine-tuning instructions, enabling it to learn new tasks autonomously.

The model employs reinforcement learning, a method in which it tests different strategies, evaluates their effectiveness, and adjusts its internal processes accordingly. This allows SEAL to refine its capabilities and increase accuracy over time.

In trials, SEAL outperformed GPT-4.1 by learning from the data it generated independently. The results demonstrate the potential of self-improving AI systems to reduce reliance on manually curated datasets and human-led fine-tuning.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!