Enterprises looking to develop custom large language models (LLMs) may soon have a new tool at their disposal, thanks to Meta’s AI research team. The team is working on creating a Self-Taught Evaluator that could revolutionize how LLMs are developed.
In a recent publication by Meta’s AI research team, known as Meta FAIR, the technology behind the Self-Taught Evaluator was introduced. This innovative tool aims to reduce the time and resources needed to develop custom LLMs by allowing the model to create its own training data for evaluation purposes.
Traditionally, LLM evaluators require large amounts of annotated data, a process that is not only costly but also results in data that becomes outdated as the model improves. However, with the Self-Taught Evaluator, LLMs can generate their own synthetic data for evaluation, eliminating the need for costly human annotation.
The implications of this technology are vast, as it could significantly streamline the development process for custom LLMs. By reducing the time and resources required, enterprises can more efficiently develop cutting-edge models without the constant need for human intervention.