More groups are trying to build AI that improves itself

In short: Researchers and smaller teams, not just the biggest AI labs, are running experiments where AI helps build better AI.

What's going on

Big AI companies are racing to create “self-improving” AI models. That means the AI is used to help make the next version of itself better, like a student who can also rewrite and upgrade their own textbook.

Some people think this could be a fast path to “superintelligence,” which is the idea of an AI that becomes smarter than humans. The concern is that if improvements feed into more improvements, progress could speed up in a loop.

A key point in WIRED’s report is that this work is not limited to the so-called frontier labs, meaning the best-funded groups with the biggest models. Experiments in “AI building AI” are spreading more widely. That suggests the future of self-improving AI could involve many players, including smaller labs and researchers.

What to watch

If more groups can run these kinds of experiments, progress could become harder to track and coordinate. It could also make it tougher to agree on safety rules, because there are more builders, more approaches, and more incentives to move quickly. Watch for clearer evidence of how well these self-improving methods actually work, and for new safety efforts that focus on the process of improvement, not just the final model.

Source: Wired