I recently had a client who used an LLM to summarize a series of video lectures I did, and then comment on them. The summaries were about 50% on the ball, 25% s

I recently had a client who used an LLM to summarize a series of video lectures I did, and then comment on them. The summaries were about 50% on the ball, 25% silent on important points, and 25% wrong and weird. Their commentary was a waste of their time and mine, as they got a partial and distorted notion of my work— but I couldn’t easily repair the damage. All I could really do is warn them that nothing they had read should be trusted.