Researchers from Tsinghua University and BIGAl have developed Absolute Zero, a groundbreaking self-learning Al that requires no human data. Instead, it creates its own tasks, solves them using diverse reasoning strategies, and improves through self-play-essentially teaching itself from the ground up.
This Al, known as the Absolute Zero Reasoner (AZR), has achieved top-tier performance in coding and math, outperforming models trained with expert guidance. It marks a major turning point in Al development: moving from systems that rely on human input to ones that can grow and evolve entirely on their own.
via : Technology
www.linkedin.com
www.marktechpost.com
This Al, known as the Absolute Zero Reasoner (AZR), has achieved top-tier performance in coding and math, outperforming models trained with expert guidance. It marks a major turning point in Al development: moving from systems that rely on human input to ones that can grow and evolve entirely on their own.
via : Technology
Absolute Zero: Self-Learning Language Models Without Any Human Data
What if a model could teach itself to think better—without ever being shown how? The researchers at Tsinghua University, BIGAI, and Penn State are exploring this very frontier with their breakthrough framework: Absolute Zero. 🚫 Zero Supervision, 🔁 Infinite Growth Traditional language model trainin
AI That Teaches Itself: Tsinghua University’s 'Absolute Zero' Trains LLMs With Zero External Data
AI That Teaches Itself: Tsinghua’s Absolute Zero Trains LLMs With Zero External Data

Last edited: