[ad_1]
Meta, the parent company of Facebook, announced the release of new AI models from its research division, including a “Self-Taught Evaluator” that could lead to less human involvement in AI development. This tool, introduced in an August paper, utilizes a technique similar to OpenAI’s o1 models to make accurate judgments about AI responses by breaking down complex problems into logical steps.

The Self-Taught Evaluator model was trained using AI-generated data without human input, showcasing a potential pathway towards autonomous AI agents that can learn from their mistakes. This advancement could eliminate the need for human input in the evaluation process, opening up possibilities for digital assistants capable of performing tasks without human intervention.

Meta researchers aim for AI to become better than the average human at self-evaluation, ultimately reaching super-human levels of intelligence. This concept is crucial for advancing AI capabilities and reducing the reliance on human involvement in the AI development process.

Additionally, Meta released updates to other AI tools, such as the image-identification Segment Anything model and tools to aid in the discovery of new inorganic materials. Companies like Google and Anthropic have also explored the concept of Reinforcement Learning from AI Feedback, although Meta is a leader in releasing their models for public use.

[ad_2]
SOURCE