Michael Veale
Dutch judge uses chatGPT to answer a *factual question* about the lifetime of solar panels which was used in determining the extent of compensation. Really. Really. https://www.nu.nl/tweakers/6323110/rechter-gebruikt-chatbot-chatgpt-in-een-uitspraak.html @alex
Rechter gebruikt chatbot ChatGPT in een uitspraak

Een kantonrechter heeft antwoorden van ChatGPT, de AI-chatbot van OpenAI, gebruikt om tot een uitspraak te komen. Voor zover bekend, is dit nog niet eerder gebeurd. Het is echter ook mogelijk dat dit nu pas voor het eerst in het vonnis wordt benoemd.

NU
Aug 2, 2024 at 2:44pmWeb
Show thread[David Nathanson has moved]Aug 2, 2024
@mikarv @alex I quit.
Show threadTom Bellin Aug 2, 2024
@mikarv @alex Inevitable.
Show threadWim🧮Aug 2, 2024

@tob @mikarv @alex

As ChatGPT has no notion of truth or correctness, I imagine this could lead to an appeal on procedural grounds? The judge should have consulted an expert.

Show threadTom Bellin Aug 2, 2024

@wim_v12e @mikarv @alex The judge thinks they have consulted an expert.

Hundreds of millions of dollars has been spent to convince regular people (like this judge) that chatbots are "AI."

Show threadAlex HannaAug 2, 2024
@mikarv Could we *not*.
Show threadᴚ uɐᗡAug 2, 2024
@mikarv @alex oh jesus christ no
Show threadNiccoAug 2, 2024
@mikarv @alex oh boy. I’m officially ashamed to be Dutch right now. Let’s hope this judge will be kicked out asap
Show thread🍉 –Hans Bot– 🙃Aug 2, 2024
@nicovanmourik @mikarv @alex Boy. There’s so much to be ashamed of when you’re Dutch. From imperialism through slavery to German collaboration, Srebrenica, and now the extreme right in power. A continual, deep embarrassment 😬
There are things to be proud of, I think, erm, wait, I’ll come up with something
Show threadFrederik BorgesiusAug 2, 2024

@mikarv

!?!?!?

Show threadFrederik BorgesiusAug 2, 2024
@mikarv Oh no, @dingemansemark - did you see this?
Show threadlightspeedAug 2, 2024

@Frederik_Borgesius @mikarv @dingemansemark Completely unacceptable. The fact that it was added in the judgment itself makes clear this subdistrict court judge thinks LLMs are a search machine.

Makes you wonder how often LLMs were used without anybody knowing in a judgment. The judiciary should urgently issue a guideline for judges that prohibits the use of LLMs.

cc: @ilyaz

Show threadCybarbieAug 2, 2024

@lightspeed @Frederik_Borgesius @mikarv @dingemansemark @ilyaz It does not matter whether you use a search engine or an augmented LLM as long as you are citing the actual source it finds such as government statistics or some industry report.

The point is the reference should be the official source not 'ChatGPT' or 'Google'.

There isn't a single lifespan for solar panels - some last longer than others - newer ones are expected to last 50 years.

Ask stupid questions get stupid answers.

Show threadWinfried is Anti Fascism 🏳️‍🌈Aug 2, 2024
@nf3xn @lightspeed @Frederik_Borgesius @mikarv @dingemansemark @ilyaz when ask ChatGTP for references chances are high it hallucinates them.
Show threadsynAug 3, 2024
@winfriedtilanus you obviously shouldn't just ask for references but also check that the reference actually says what is claimed that it says
Show threadWinfried is Anti Fascism 🏳️‍🌈Aug 3, 2024
@syn First check if the references exist. I've seen a list of 5 really catchy titles of scientific articles. None of them existed.
Show threadsynAug 3, 2024
@winfriedtilanus well yes obviously you can't check the content if the source doesn't exist in the first place
Show threadMark DingemanseAug 2, 2024
@lightspeed @Frederik_Borgesius @mikarv @ilyaz what a nightmare! Do judges not even read the news, e.g. of practitioners in the US getting disbarred for relying on LLM-generated bullshit?
Show threadWinfried is Anti Fascism 🏳️‍🌈Aug 2, 2024
@dingemansemark @lightspeed @Frederik_Borgesius @mikarv @ilyaz the biggest problem of the LLMs is that their output looks like higher quality then it in reality is. That is a trap many more people will fall into.
Show threadFrederik BorgesiusAug 3, 2024
@dingemansemark Shocking.
Show threadAnna MiriamsdochterAug 2, 2024
@mikarv @alex o nee 🫢
Show threadMichael EkstrandAug 2, 2024
@mikarv @alex wtf?
Show threadalastair87 (old account)Aug 2, 2024
@mikarv @alex The simplest way for the judge to understand the problem might be to be told that it's akin to asking another human to answer the question from memory and not looking up the detailed documentation. Hopefully they wouldn't do that so they shouldn't do this either.
Show threadSebastien 🌍 🇵🇸🇺🇦 Aug 3, 2024

@mikarv why is this so worrying? Yes he could have consulted an expert but those are humans and humans are also prone to making mistakes like this human judge proves with this case.

Many people have relied on ths judge's expertise thinking he is an expert. They could just have used ChatGPT instead 😈 @alex

Show threadMichael VealeAug 3, 2024
@SebastienK because you (a plaintiff, a higher court) can appraise the informational quality of eg a study or table on solar panel life, you cannot appraise ‘what chatgpt happened to say with that random seed and that prompt’, because there is no method.
Show threadSebastien 🌍 🇵🇸🇺🇦 Aug 3, 2024
@mikarv I get that but what I wonder now and did not read in the article is if ChatGPT was correct? What does an expert say?
Show threadMichael VealeAug 3, 2024
@SebastienK Like asking an expert if the roll of the dice were correct in hindsight. The problem is the process.
Show threadJohn PanzerAug 6, 2024
@mikarv @alex wtf
Show threadxi timpinAug 6, 2024
@mikarv @alex Sounds like the judge thinks chatGPT is some kind of expert witness.
I don't know anything about the Netherlands legal system but this must surely be overturned if an appeal is lodged.