The dangers of ChatGPT/AI - another hallucination case.

[atticus]

This post is a reminder to anyone using artificial intelligence "large language models" to conduct legal research that these systems are prone to 'hallucination'. They make things up.

There was a well-publicised case in the States a few months ago in which a lawyer had used an AI model to help prepare submissions to the court. The model cited case law - but the cases did not exist. They were fabrications. Now this has happened in an English tribunal. Read paragraphs 1 to 24 of Harber v Commissioners of HMRC, it won't take long.

If you must use an AI model, check everything that it produces for you.

[PallasAthena]

Very interesting, thank you.

[R0b]

Another relevant article in which a Court of Appeal judge admits to using AI assistance.

https://www.lexology.com/library/det...7-98d0c6c429e5

[R0b]

Alternative link if you don’t have an account with lexology.

https://www.ashfords.co.uk/insights/...n-his-judgment

[atticus]

I have used ChatGPT to help prepare notes for students. It can remove a lot of drudgery in preparing a first draft. But it needs to be checked, and I have found mistakes in its output. But it saved time in preparing those notes, even if you include my time spent checking and amplifying. I am sure lawyers will be using AI regularly.

The most important thing is that, as Birss LJ said, you have to take personal responsibility for your work.

[atticus]

test - 2 earlier replies to post #4 appear to have been sent to the spam holding cell.

[EXC]

test - 2 earlier replies to post #4 appear to have been sent to the spam holding cell.

I've approved your post. Blame AI.