Meta launches self-assessment AI model to reduce human involvement and enhance development efficiency
Meta announced the launch of a series of new artificial intelligence models, including a self-assessment model aimed at reducing human involvement and improving development efficiency. This model utilizes the "thought chain" technology similar to OpenAI's o1 model, which enhances accuracy in answering scientific, coding, and mathematical questions. Meta's researchers trained the model using data generated entirely by artificial intelligence, marking a significant advancement in AI self-learning and self-assessment capabilities. In addition, Meta also introduced other AI tools, further demonstrating its commitment to advancing artificial intelligence technology
According to the financial news app Zhitong Finance, Meta (META.US), the parent company of Facebook, announced that it will release a series of new artificial intelligence models, including a self-assessment model that may reduce human involvement in the AI development process. This release comes after Meta introduced the tool in August, detailing in a paper how the tool uses the same "pathway chaining" technology as OpenAI's o1 model to improve accuracy in answering challenging questions in science, coding, mathematics, and more. Meta's researchers trained the evaluation model using data generated entirely by artificial intelligence, eliminating the need for human input.
This move by Meta signifies a significant step for the company in the self-learning and self-assessment capabilities of artificial intelligence. This self-assessment technology not only improves the accuracy of models but also potentially reduces costs and inefficiencies in the reinforcement learning process that currently relies on human feedback. Meta's researchers state that using AI to assess its own capabilities is crucial in creating intelligent agents that can learn from mistakes and improve autonomously. These intelligent agents are envisioned to be digital assistants capable of performing a wide range of tasks without human intervention.
Jason Weston, a researcher at Meta, emphasizes that as AI advances, it will become increasingly adept at self-checking. The ability for self-learning and self-assessment is crucial for artificial intelligence. Apart from Meta, other companies such as Google and Anthropic are also researching concepts of reinforcement learning based on AI feedback, but they typically do not publicly release their models.
In this release, Meta also introduced other AI tools, including updates to the image recognition model Segment Anything, which can speed up response generation time for large language models (LLMs), and a dataset that can help discover new inorganic materials. The launch of these tools further demonstrates Meta's efforts and commitment to advancing artificial intelligence technology