Are LLMs Good at Causal Reasoning? with Robert Osazuwa Ness - #638

In this episode of the TwiML AI podcast, host Sam Charrington interviews Robert Ness, a senior researcher at Microsoft Research, professor at Northeastern University, and founder of AltDeep.ai, about causal reasoning and large language models (LLMs). They discuss a recent paper co-authored by Ness that explores the capabilities of LLMs in causal analysis. The conversation covers the basics of causal analysis, examples of causal reasoning in LLMs, and the potential of LLMs to enhance causal analyses. While LLMs show promise in tasks like pairwise causal discovery and creating causal graphs, concerns remain about their reliability, potential memorization of benchmarks, and the need for human oversight. The discussion also touches on future research directions, such as using Reinforcement Learning from Human Feedback (RLHF) to improve causal reasoning in models and the importance of understanding how models arrive at causal conclusions.