🤖🧠 Anthropic Tests AI That Improves Other AI Models

Anthropic has shared new research about AI systems that can improve other AI models. An automated researcher reads scientific papers, chooses a training method, and tests it for 30 minutes at a time.

In the study, the system worked on ten AI safety tests. It improved performance in every test without making the model’s overall results worse. The strongest methods were kept, and the weak ones were removed. Anthropic says the automated system costs about $4 per hour, compared with $150 per hour for human researchers.

However, the project still has important limitations. It can only work well when the safety tests and research information are reliable. The results are early, but they show how AI may develop better training methods in the future.

✅ Level: Pre-Intermediate
🔗 Source: [TechCrunch]

(https://techcrunch.com/2026/08/28/an-anthropic-researcher-just-gave-us-a-peek-at-self-improving-ai/)

Would you trust AI to improve other AI models? 👇

#News