In a groundbreaking move, Meta has announced the release of a new suite of AI models from its research division, featuring the innovative \"Self-Taught Evaluator.\" This advanced model aims to reduce human involvement in the AI development process by autonomously assessing the performance of other AI systems.
The Self-Taught Evaluator leverages the \"chain of thought\" technique, a strategy also employed by OpenAI's recent models, to break down complex problems into manageable steps. This approach enhances the accuracy of responses in challenging areas such as science, coding, and mathematics.
What sets Meta's evaluator apart is its training process, which relies entirely on AI-generated data, eliminating the need for human input during this stage. According to Meta researchers, this advancement paves the way for autonomous AI agents capable of learning from their own mistakes, potentially revolutionizing digital assistants by making them more intelligent and self-sufficient.
Jason Weston, one of the lead researchers, emphasized the significance of self-evaluating AI, stating, \"We hope, as AI becomes more and more superhuman, that it will get better and better at checking its work so that it will actually be better than the average human.\" This self-improvement capability could streamline the AI development process by removing the need for Reinforcement Learning from Human Feedback, a method currently reliant on specialized human annotators.
Meta's release also includes updates to its image-identification Segment Anything model, a tool designed to accelerate large language model response times, and new datasets aimed at aiding the discovery of novel inorganic materials. These tools underscore Meta's commitment to advancing AI technology and supporting diverse applications across various industries.
While other tech giants like Google and Anthropic are exploring similar concepts under the banner of Reinforcement Learning from AI Feedback (RLAIF), Meta distinguishes itself by making these models available for public use, fostering greater collaboration and innovation in the AI community.
Reference(s):
cgtn.com
