Microsoft Research Podcast

AI Frontiers: The future of scale with Ahmed Awadallah and Ashley Llorens

Sep 14, 2023

AI scientists Ahmed Awadallah and Ashley Llorens discuss the future of scale in AI, including advancements in large-scale models like GPT-4 and their impact on reasoning and problem-solving. They explore the dynamics between model size and data, the use of large-scale models to improve smaller ones, and the need for better evaluation strategies. They also delve into topics such as spending compute budget on bigger models, the capabilities and limitations of AI models, the concept of post-training in language model training, and advancements in AI and adaptive alignment.

42:34

Creator website

Episode guests

Ahmed Awadallah

Ashley Llorens

AI Summary

AI Chapters

Episode notes

Podcast summary created with Snipd AI

Quick takeaways

The importance of data, especially high-quality and representative data, has been recognized as crucial in improving the performance of large-scale AI models.

The use of powerful models like GPT-4 to train smaller, more specialized models can enhance reasoning abilities and boost performance in specific domains and tasks.

Deep dives

The Importance of Data in AI Progress

The podcast episode discusses how the understanding of what drives progress in AI has evolved over the years. While scale was initially thought to be the main driver, the importance of data in training AI models has become increasingly clear. More data, especially high-quality and representative data, has proven to be crucial in improving the performance of large-scale models. The podcast also highlights the value of training models on diverse datasets, including text and code, which surprisingly enhances their performance in various tasks. Additionally, the episode explores the two stages of training models: pre-training and post-training, and how they contribute to further advancements in AI.

Introduction

5min

Spending Compute Budget on Bigger Models

24min

Capabilities, Limitations, and Safety Evaluation in AI Models

4min

Exploring the Concept of Post-Training in Language Model Training

4min

Advancements in AI and Adaptive Alignment

5min

Powerful large-scale AI models like GPT-4 are showing dramatic improvements in reasoning, problem-solving, and language capabilities. This marks a phase change for artificial intelligence—and a signal of accelerating progress to come.

In this Microsoft Research Podcast series, AI scientist and engineer Ashley Llorens hosts conversations with his collaborators and colleagues about what these models—and the models that will come next—mean for our approach to creating, understanding, and deploying AI, its applications in areas such as healthcare and education, and its potential to benefit humanity.

This episode features Senior Principal Research Manager Ahmed H. Awadallah, whose work improving the efficiency of large-scale AI models and efforts to help move advancements in the space from research to practice have put him at the forefront of this new era of AI. Awadallah discusses the shift in dynamics between model size and amount—and quality—of data when it comes to model training; the recently published paper “Orca: Progressive Learning from Complex Explanation Traces of GPT-4,” which further explores the use of large-scale AI models to improve the performance of smaller, less powerful ones; and the need for better evaluation strategies, particularly as we move into a future in which Awadallah hopes to see gains in these models’ ability to continually learn.

Learn more: