Today I’m reacting to a July 2024 interview that Prof. Subbarao Kambhampati did on Machine Learning Street Talk.

Rao is a Professor of Computer Science at Arizona State University, and one of the foremost voices making the claim that while LLMs can generate creative ideas, they can’t truly reason.

The episode covers a range of topics including planning, creativity, the limits of LLMs, and why Rao thinks LLMs are essentially advanced N-gram models.

00:00 Introduction

02:54 Essentially N-Gram Models?

10:31 The Manhole Cover Question

20:54 Reasoning vs. Approximate Retrieval

47:03 Explaining Jokes

53:21 Caesar Cipher Performance

01:10:44 Creativity vs. Reasoning

01:33:37 Reasoning By Analogy

01:48:49 Synthetic Data

01:53:54 The ARC Challenge

02:11:47 Correctness vs. Style

02:17:55 AIs Becoming More Robust

02:20:11 Block Stacking Problems

02:48:12 PlanBench and Future Predictions

02:58:59 Final Thoughts

Show Notes

Rao’s interview on Machine Learning Street Talk: https://www.youtube.com/watch?v=y1WnHpedi2A

Rao’s Twitter: https://x.com/rao2z

PauseAI Website: https://pauseai.info

PauseAI Discord: https://discord.gg/2XXWXvErfA

Watch the Lethal Intelligence video and check out LethalIntelligence.ai! It’s an AWESOME new animated intro to AI risk.

Doom Debates’ Mission is to raise mainstream awareness of imminent extinction from AGI and build the social infrastructure for high-quality debate.

Support the mission by subscribing to my Substack at DoomDebates.com and to youtube.com/@DoomDebates. Thanks for watching.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit lironshapira.substack.com