Imagine a AI that figures out when it needs to learn something new and then makes its own practice tests to get better. Normally these programs are stuck with the knowledge they were made with unless people put in a lot of effort to update them. This new idea called SEAL lets the program write its own instructions on how to change its brain based on new information it finds. It learns by practicing and rewarding itself when it gets the right answers just like a student studying for a test. This makes the program much better at remembering facts and solving new puzzles on its own. The only tricky part is that sometimes it learns new things but accidentally forgets old things which researchers are still trying to fix.
MIT: Self learning and adapting language models
AI that can learn new things and improve themselves without people having to help them.