Generative AI in a nutshell

K&L Gates fined for hallucinated citations. A fantastic explainer for busy laypeople

Estimated reading time: 2 minutes (and video: 18 minutes)

K&L Gates was fined recently (May 2025) for using hallucinated citations in its work. This prompted me to reiterate an important idea: the new generation of AI – the LLMs or Generative AI– are different from any software we’ve used before. That is why they give excellent responses, sometimes with confident-sounding mistakes. They are not deterministic — as a calculator is.

It’s a new concept that is important to understand. And so many legal professionals would be better off if they understood this from first principles.

So how do you learn without getting into too many Computer Science weeds?

I highly recommend this 18-minute explanation of how LLMs work. (credit: Henrik Kniberg)

• It provides a technical overview of LLMs' work, not just vague analogies.

• You will get an understanding of the computer science involved in LLMs.

• It is also very fun to watch – the 18 minutes will go fast!
  

  Thanks to one of our readers for sending me this one. I’ll be posting additional tutorial resources soon.

ps - Kenneth Rashbaum, Partner at Barton LLP, expressed concerns that the AI industry has thrown up it’s hands saying “Oh, well, AI will be AI. It's still in early stages.” and is doing very little to reduce hallucinations.

My person view is that those vendors are wrong. They shouldn’t just live with hallucinations and expect their users to live with it. There are essential and doable efforts to be undertaken: improving grounded-ness, algorithmic verification, and user interfaces for easier source verification.