Meta unveils a new artificial intelligence model for self-evaluating and training other models.
Meta announced the launch of an innovative artificial intelligence model called “Self-supervised Learning Resident,” which aims to improve the evaluation and training capabilities of artificial intelligence models independently and without the need for human intervention. This step comes as part of the rapid expansion of artificial intelligence technologies, where major companies like Meta strive to develop advanced models to keep up with the ongoing challenges in this field.
Developing artificial intelligence models poses a significant challenge, requiring huge investments and resources to ensure continuous innovation and competitiveness. With the widespread adoption of artificial intelligence by many major technology companies, whether to develop their own models or integrate solutions from external parties, there is a growing need for technologies that facilitate and accelerate this process.
Until now, most artificial intelligence model training relies on human intervention through techniques like “reinforcement learning with human feedback,” which requires significant time and effort and leads to slow development.
However, Meta’s new model stands out for its ability to conduct evaluation and training processes autonomously without human intervention. This feature opens up new horizons to address the current challenges faced by developers and companies in terms of time and cost, enhancing the efficiency and speed of developing more accurate and effective models.
The “thought chain” technique, previously used by OpenAI in its models, is an important part of this new model. This technique involves breaking down large and complex problems into smaller logical steps, contributing to improving the accuracy of results in areas like programming, mathematics, and science.
Meta utilized the same technique in developing the “Self-supervised Learning Resident” model, training it on data generated by artificial intelligence itself, significantly improving the model’s accuracy and efficiency.
Jason Weston, the researcher on the project, explained that the goal is for artificial intelligence to outperform humans not only in performance but also in its ability to evaluate itself and improve its outputs, affirming that this model will be a fundamental cornerstone for the future development of artificial intelligence.
