Intro

Came across Absolute Zero Reasoner (AZR); a model that literally teaches itself reasoning without any human data. its now real. it takes data from other LLM’s and iterate it according to AZR’s self proposing task (AZR creates task for it self) and solves it then loop back.

So how it works? or it doesnt?

self-generates(Problem), self-validates(Solves and Checks), and self-improves(if error/or possible improvement can be done).
and in the end it rewards itself (not literal)
Results? It beats models trained on tens of thousands of examples; on math & code problems, even with zero external data

My thought

You are living your own “absolute zero Reasoner”; no roadmap, no perfect start.
But like AZR, youll propose tasks, solve them, verify results, learn, and repeat.
That's what true growth looks like, and that what we are expecting the new AI model to do,

The ERA of Super Intelligence???