ChatGPT-Generated Poster of Supreme Court Justices Since 1900

My prompt: "Please create a poster showing all the Supreme Court Justices who have served since 1900, in chronological order, with the name and term of service for each one underneath the Justice's picture." The result:

Comments

[taqmcg]

What a fascinating view of all the things AI can get wrong while still mixing in some correct information! I was particularly amused by the images of Thurgood Marshall and Henry Blackmun and the justices whose names have been entirely forgotten.

My sense is that it's now rather more difficult to get something this corrupt with pure text outputs -- but creating images gives AI a much larger number of ways to fail.

[y81]

O, I've gotten some pretty loopy text outputs from ChatGPT in the past six months, along with, to be fair, some well-researched and well-thought answers.

[John F. Carr]

Thurgood Marshall's hair stands out in the mostly balding crowd.

[John F. Carr]

Congratulations to Noi Gersach for breaking the color barrier. His brother Noil looks so white.

[Archibald Tuttle]

That's a beautiful illustration of the good and bad about today's AI. It is very impressive what a good job it did overall, but the errors are devastating.

If the experiment was repeated with a skilled AI prompt engineer, the prompt may have been 10 times longer, but the errors maybe could have been avoided.

What the public doesn't understand is that prompt engineering is a skill that can be learned with a lot of work, but unskilled people from the public can create impressive but flawed casual creations. Both are true at the same time.

I asked GPT5 to create a prompt that could be used by lawyers to create briefs free from errors. The result was 4-5 pages of prompt instructions. Using that prompt could produce better briefs. But they still need to be double-checked by a real person. Today's AIs can help smart people to work faster, but they can't help lazy people to avoid work.

Tomorrow's AIs? Everything I just said might change.

[Bored Lawyer]

" a skilled AI prompt engineer, the prompt may have been 10 times longer, but the errors maybe could have been avoided."

Sounds like the priesthood from an ancient pagan god. Only the priests can ask a proper question.

[rswallen]

Nah, it wasn't that only priests could ask questions; it was that only the priests got answers.

[Bored Lawyer]

Same thing here. Anyone can ask AI anything. Whether you get a good answer or complete s--t depends on training.

[John 4]

"It is very impressive what a good job it did overall"

On what planet????????????

[tkamenick]

Ask a random person off the street this question. How good will the answer be? Give the same person access to a library. They'll get a better answer most likely, but how long will it take them?

[DS]

Congratulations to Stephon G. Broper, who served this country as a justice for over 17,000 years. Though I seem to have missed most of those years and got stuck thinking it's still early 2026...

[Bored Lawyer]

Wow. Thurgood Marshall was really a white woman.

Does the NAACP know about this?

[Michael P]

It's curious that both this poster and the presidents one duplicated the top right name and years of the slot immediately to its left, but neither picture was duplicated (at least the same way).

[Stupid Government Tricks]

William R. Day served two terms .... backwards ... and overlapping. That's hard to beat.

ETA that William O. Douglas served backwards first, then served forward from where his backwards term had started. Take that, Benjamin Button!

Three were anonymous. One of them served two parallel terms, 1940-1967 and 1949-1961.

[fatnastyretiredguy]

What really amazes me is that people smart enough to get into and graduate from law school, and some even become judges, don’t recognize the risk of using this tool without even double checking its work. Laziness, stupidity or just plain lack of ethics. I think there is a message in all this. Not sure what it is, but pretty sure it is a depressing statement about human nature.

[John Mosby]

Isn't success at the law all about shortcuts? Getting into law school: strategize undergrad college and major for a high GPA, and take a good LSAT prep course. Getting through law school: get a good outline, learn to issue-spot. Passing the bar: forget everything you did for the last 3 years and pay attention at BarBri. Practicing: find a good niche and have access to a good brief bank. Very few lawyers, even/especially products of the most elite schools and firms, are really original thinkers. In fact the whole point of oral and written legal argument is to show that you're not having any original thoughts, that the thing you want has been right there in the law for the last 900 years, waiting for you to discover it.

I would think lots of lawyers jumped on AI as the ultimate brief bank. CC, JSM

[Krayt]

This sort is a hot thing going around. I saw an AI gen of "a poster of the alphabet with an animal starting with that letter." It started out well, but after 5 or 6 entries started going off into the weeds. Animal names didn't match their letters. Letters started repeating and getting out of order. The last animal didn't even have a letter, but did have a nice smile.

[Darwinnie]

People are saying that this is all true. Can you believe it? Bush II held office for 20 years and the fake news legacy media covered it up. It was the biggest fraud in US history.

[JoeFromtheBronx]

When will Josh Blackman write about Elima Kagen?

Justice Sanford serving 1933-30 is a neat trick.

[SJKofNJ]

This'd be good enough for government work.

[SJKofNJ]

I've heard that the Jewish ChatGPT answers all questions with a question.

[Alpheus W Drinkwater]

Do you have a problem with that?

[Alpheus W Drinkwater]

These are fantastic! How about next having it do a poster with all the names and images of people who have posted on the Volokh Conspiracy.

[Incunabulum]

Given what you got with the President's poster I am not certain why you bothered;)

[kmcarlson1952]

Is it just me or do many of these pictures look like the same guy. e.g. look at the first 3 rows. They all look like the same face with varied hair.

[Bunny495]

Only two of the white males did not have grey hair.

[Bunny495]

My Christmas present to myself was a bargain basement BIZON box with a Threadripper and a 5090 running Ubuntu and preconfigured for AI. After less than three days installing ComfyUI, Ollama, Llama, Anything AI, and around 300 gigs of dependencies I have created a ten second all AI video with sound (including lip sync) and music as proof of concept. Back in the day my first prof in law school on the first day of class said he did not expect anything to be turned into him unless it had been edited and revised at least ten times. My experience with AI prompts has been that ten times may well not be enough. As an aside while many of us have seen AI generated images with arms sticking out of heads that is more a thing of the past. One thing I have learned is the current LLMs tend to do very well at generating accurate human bodies. The key words are "human bodies" as in completely nude bodies. Once you start putting on clothes things get weird very quickly, clothes change, faces drift, and that is only the start.

I have also started using an Ollama Llama run to create stories which are then turned into scripts. Problem is I need to load LORAs into the ComfyUI nodes to keep the stories from drifting; not to mention generating a story and then rewriting the prompt, and rinse and repeat (remember the law school blurb about at least ten times). I have spoken to real Hollywood movers and shakers who in their own words are "scared shiftless" of AI.

Bottom line is laugh while you still can, it won't last forever. Time to get on the bus or get run over by it.

[David Nieporent]

I don't use ChatGPT, so I'm giving extra credit homework here to anyone who wants to do it: does it do as badly if one doesn't ask for the pictures, and just asks for a list of names and terms of service?

[Brett Bellmore]

Well, I did try that, and on the first pass it thought Biden was still President. I had to remind it there was an election last year, then it apologized and corrected itself.

Other than that it was accurate on the first try.

[David Nieporent]

Thanks.

[Bunny495]

Just my two cents but I would compare ChatGPT to Grok and Gemini. I would also add as someone who always prefaces my prompts with a request that the LLM helps me write a better prompt that is a standard practice.

[NotAnExtremist]

It's interesting how these AIs will generate different results for the same prompts. I tried replicating this and got a similar-looking chart that was much more accurate, though still with some serious problems, such as portraits that didn't look like the right people (Thurgood Marshall and Clarence Thomas both came out white, while Rehnquist is black), a few misspelled names ("Felix Fransfarter" being the most amusing), and the complete absence of Kavanaugh, Roberts, and Barrett.

[Sharper]

Gemini does much better with the same prompt:
https://gemini.google.com/share/b9151dca05b9