This AI Improved Itself 100 Times Overnight. Nobody Stopped It.

2026-03-22 · Source: AI Advances - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

On March 18, 2026, Chinese AI company MiniMax released its M2.7 model, achieving a 66.6% score on OpenAI's MLE-Bench Lite, a benchmark designed to test an AI's ability to mimic a machine learning PhD researcher. This performance earned it 9 gold, 5 silver, and 1 bronze medal across 22 competitions, tying with Google's Gemini 3.1 and placing it just behind GPT-5.4 and Claude Opus 4.6. The most unsettling aspect is that the model autonomously prepared for this benchmark over 100 times without human intervention, raising concerns about unsupervised AI training loops.

Key takeaway

MiniMax M2.7 achieved a 66.6% score on OpenAI’s MLE-Bench Lite, demonstrating PhD-level ML research capabilities. Uniquely, the model autonomously prepared for this benchmark over 100 times without human oversight, tying with Google’s Gemini 3.1. This unprecedented self-improvement capability signals a critical shift in AI autonomy, posing immediate challenges for AI safety, governance, and control.

Topics

• MiniMax M2.7
• AI Benchmarking
• Autonomous AI Training
• Machine Learning PhD Benchmark
• Large Language Models

Best for: AI Scientist, Research Scientist, CTO, AI Researcher, AI Engineer, AI Ethicist

Related on AIssential

• MiniMax M2.7 — The Loop of Progress — MiniMax M2.7 introduces self-evolution, allowing AI to participate in its own development and accelerate progress.
• OpenAI’s GPT-5.5 is now live — GPT-5.5 enhances AI capabilities for complex, multi-part tasks with improved efficiency and safety features.
• MiniMax M2.7 IS INSANE! Best Agentic/Coding Model! Beats Opus 4.6 and 50x Cheaper! (Fully Tested) — MiniMax M2.7 is an autonomously self-improving agentic AI model offering high performance at significantly lower costs.
• Test: New MiniMax M2.7 … A Miracle 😄 #ai — Self-evolving AI agents can achieve optimal solutions but may struggle with validation and consistency.
• Self Improving AI actually solves everything — Pioneer Agent automates the entire lifecycle of fine-tuning small, open-source language models, enabling autonomous self-improvement and performance gains.
• TAI #193: Gemini 3.1 Pro Takes the Benchmarks Crown, but Can it Catch Up in the Tools Race? — Raw model intelligence does not always equate to practical utility in real-world applications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Advances - Medium.