Pretty sure that, looking back from the rubble-strewn wasteland of the Machine Wars, this is the point humanity will identify as when we probably should've, you know, pulled the plug on this stuff.
Wired reports:
...A project from Tsinghua University, the Beijing Institute for General Artificial Intelligence (BIGAI), and Pennsylvania State University shows that AI can learn to reason in this way by playing with computer code.
The researchers devised a system called Absolute Zero Reasoner (AZR) that first uses a large language model to generate challenging but solvable Python coding problems. It then uses the same model to solve those problems before checking its work by trying to run the code. And finally, the AZR system uses successes and failures as a signal to refine the original model, augmenting its ability to both pose better problems and solve them.
In other words, it's...
The team was rewarded with its novel approach to AI: The new protocol "significantly improved the coding and reasoning skills of both 7 billion and 14 billion parameter versions of the open source language model Qwen," including those that had received "human-curated data."
This does feel like the next step for AI — the sort of thing that could take it to another level. Asking oneself questions, after all, feels significantly close to "metacognition," a hallmark of higher thinking.
It's the process of having thoughts about thoughts, what MIT describes as the process in which "learners use knowledge of the task at hand, knowledge of learning strategies, and knowledge of themselves to plan their learning, monitor their progress towards a learning goal, and then evaluate the outcome."
And indeed, that appears to be how researchers are thinking of this:
[Researcher Andrew] Zhao told me that the approach resembles the way human learning goes beyond rote memorization or imitation. "In the beginning you imitate your parents and do like your teachers, but then you basically have to ask your own questions," he said. "And eventually you can surpass those who taught you back in school."
In this case, of course, "those who taught you back in school" is...us. Human beings. We're the ones the AI is going to "surpass."
Not a good sign for those of us who are acutely worried about what AI means for the future of humanity. And really, the writing is on the wall already:
[A] project like Absolute Zero might lead to AI systems that are less like copycats and more like humans.
Bit nervous about this one folks!
Our latest video 👇
