A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a significant advancement in making AI systems more energy-efficient. Researchers developed a method that allows large language models (LLMs) to generate their own training data, a process called “self-training.” This approach enabled a much smaller model to match or exceed the performance of a model ten times its size on specific reasoning benchmarks, while using far less computational power for training. The technique focuses on improving the model’s logical reasoning and common-sense understanding by having it create and learn from synthetic question-answer pairs. This breakthrough suggests a path toward more accessible and sustainable AI development by reducing the massive data and energy requirements typically associated with training state-of-the-art models. Read the full article at https://technologyreview.com/2024/07/12/1094755/ai-models-can-beat-larger-rivals-if-they-train-themselves-study-finds/