Meta, the owner of Facebook, announced that it is launching several new AI models from its research division. One of these models is called the “Self-Taught Evaluator,” which could help reduce the need for human involvement in developing AI.
This announcement comes after Meta shared details about the tool in an August research paper. The Self-Taught Evaluator uses a method called “chain of thought,” similar to what OpenAI used in its new o1 models. This method helps the AI break down complex problems into smaller, simpler steps, making it better at answering difficult questions in areas like science, coding, and math.
Meta’s researchers trained the evaluator model using data created entirely by AI, meaning there was no human involvement during this process. This approach of having AI assess other AI models opens up exciting possibilities for creating autonomous AI agents that can learn from their own errors. Many experts in the AI field see these agents as advanced digital assistants capable of performing a wide range of tasks independently, without requiring human help.
Self-improving models could eliminate the need for the costly and inefficient process known as Reinforcement Learning from Human Feedback. This current method relies on human experts to accurately label data and ensure the correctness of answers to complex math and writing queries. Jason Weston, one of the researchers, expressed optimism that as AI continues to advance beyond human capabilities, it will improve its ability to check its own work, potentially surpassing the average human’s performance. He emphasized that the concept of being self-taught and capable of self-evaluation is essential for achieving this super-human level of AI.
Other companies, like Google and Anthropic, have also researched RLAIF, which stands for Reinforcement Learning from AI Feedback. However, these companies tend not to make their models available to the public, in contrast to Meta.
Along with the Self-Taught Evaluator, Meta introduced several other AI tools. These include an update to the Segment Anything image identification model, a tool that speeds up response times in large language models, and datasets to support in the discovery of new inorganic materials.
- Code Smarter, Not Harder: 5 Free AI Tools You Need to Know! - January 22, 2025
- Meta Takes on CapCut with Its New Video Editing App ‘Edits’ - January 20, 2025
- NovaSky Debuts Affordable Open-Source AI Model for Advanced Reasoning - January 14, 2025
