Meta drops AI bombshell: Multi-token prediction models now open for research

Meta’s Novel Multi-Token Prediction Approach

• Released pre-trained models leveraging this new technique.
• Changes the traditional method of training LLMs (predicting the next word).

Details of the New Technique

• Forecasts multiple future words simultaneously.
• Promises enhanced performance and drastically reduced training times.

Origin and Impact

• Technique first outlined in a Meta research paper in April.
• Potentially transformative for the development and deployment of large language models.