Meta’s New AI Model Can Check Other AI Models’ Work -Researcher
Meta, Facebook’s parent corporation, has launched a new artificial intelligence model called the “Self-Taught Evaluator,” which aims to dramatically reduce human involvement in AI training and evaluation.
Jason Weston, an AI researcher, expressed optimism that as AI advances, it would increase its ability to check its work, potentially outperforming humans in some areas.
He emphasised that the ability to learn and evaluate itself is critical for achieving a greater level of AI capabilities.
The release comes after Meta introduced the tool in an August paper, which outlined how it uses the same “chain of thought” technique as OpenAI’s recently published o1 models to generate trustworthy judgements on model answers.
This strategy includes breaking down large problems into smaller logical steps, and it appears to increase the correctness of responses to difficult problems in science, coding, and mathematics.
According to a Reuters article, the development, which was announced on Friday, demonstrates Meta’s dedication to creating autonomous AI systems.
Unlike traditional models, which rely on human-generated data for training, the Self-Taught Evaluator was totally trained with AI-generated data.
This breakthrough approach eliminates the need for human interaction during the training phase, paving the way for completely autonomous AI agents capable of self-learning and correction.
We earlier reported that Meta is launching its AI chatbot Meta AI in six nations including Brazil, the United Kingdom, the Philippines, Bolivia, Guatemala, and Paraguay, Mark Zuckerberg announced on his WhatsApp channel.
