Today, the Michigan Supreme Court issued a split opinion in People v Prude (Docket 165664)—a case that was decided without oral argument.
Because users can now attach files with their “prompts” when using the free versions of the generative AI tools Claude.ai and ChatGPT, I decided to take each for a spin to outline the Prude decision.
The file/opinion
For this experiment, I deleted the three-page Syllabus from the .pdf file. This 21-page file was uploaded to the AI tools.
The prompt
The same prompt was given to each AI tool.
Draft an opinion outline of the attached court opinion which follows this format and uses these labels.
1.0 Case name(s):
1.1 Court name:
1.2 Docket (or file) number(s):
1.3 Argument (argued) date (if any):
1.4 Decision date:2.0 Statement of facts:
2.1 Relationship or status of parties (include parties’ full names):
2.2 Legally relevant facts (facts tending to prove or disprove an issue before the court):
2.3 Procedurally significant facts (set out the (a) cause of action, (b) the requested relief, and (c) any raised defenses):
2.4 Acronym(s) defined:3.0 Procedural history (the case disposition in the lower courts):
3.1 The lower court(s) decisions:
3.2 Any awarded damages:
3.3 Who appealed and why:4.0 Issues
4.1 Substantive issue(s)
4.1.1 The disputed point(s) of law:
4.1.2 The key case facts related to the disputed point(s) of law (legally relevant facts):4.2 Procedural issue(s)
4.2.1 What is the appealing party claiming that the lower court(s) did wrong:5.0 Judgment: (The court’s final decision about the parties’ rights, and the court’s response to a party’s request for relief. The appellate court will generally either affirm, reverse, or reverse with instructions.)
6.0 Holding: (The statement of law that the court’s answer to the issue.)
6.1 Opinion author:
6.2 Any case law overruled (with citation(s):
6.3 Any provision of law declared constitutional or unconstitutional:
6.4 Any newly announced legal standard(s):
6.5 List the (a) sentiment, (b) emotion, and (c) tone for the opinion.7.0 Rule of law or applied legal principle: (The rule(s) of law that the court applied to determine the parties’ substantive rights. The rule(s) of law could be based on constitution, statute, case law, regulation, or court rule. The rule(s) or legal principle(s) may be expressly stated in the opinion or it may be implied.)
8.0 Reasoning:
8.1 Explain the way the court applied the rules/legal principles to the particular facts to reach its decision. This includes syllogistic application of rules and policy arguments the court used to justify its holding (why the decision was socially desirable).
8.2 List the opinion’s headings and subheadings in text. Use sentence-case capitalization.
8.3 Note any image(s), figure(s), table(s), or charts in the majority opinion, and (a) repeat any text-based caption, and (b) describe each image, figure, table, or chart as they appear.
8.4 Note any dictionary references and (a) list each dictionary name, and (b) repeat any text-based term and the cited dictionary definition.9.0 Non-majority opinion(s)
9.1 Name of each concurring (only) judge. Explain their reason(s) for writing separate from the majority opinion:
9.1.1 List the headings and subheadings used in each concurring (only) opinion in text. Use sentence-case capitalization.
9.1.2 List the (a) sentiment, (b) emotion, and (c) tone for each concurring (only) judge’s opinion.
9.1.3 Note any image(s), figure(s), table(s), or charts in each concurring opinion, and (a) repeat any text-based caption, and (b) describe each image, figure, table, or chart as they appear.
9.1.4 Note any dictionary references and (a) list each dictionary name, and (b) repeat any text-based term and the cited dictionary definition.9.2 Name of each dissenting (only) judge. Explain their reason(s) for refusing to join in the majority opinion:
9.2.1 List the headings and subheadings used in each dissenting (only) opinion in text. Use sentence-case capitalization.
9.2.2 List the (a) sentiment, (b) emotion, and (c) tone for each dissenting (only) judge’s opinion.
9.2.3 Quote the sentences the judge used to express their dissent (such as “I respectfully dissent” or “I dissent”).
9.2.4 Note any image(s), figure(s), table(s), or charts in each dissenting opinion, and (a) repeat any text-based caption, and (b) describe each image, figure, table, or chart as they appear.
9.2.5 Note any dictionary references and (a) list each dictionary name, and (b) repeat any text-based term and the cited dictionary definition.9.3 Name each judge who wrote an opinion that both concurred in part and dissented in part from the majority opinion. Explain the reason(s) for it:
9.3.1 List the headings and subheadings used in each opinion that both concurred in part and dissented in part in text. Use sentence-case capitalization.
9.3.2 List the (a) sentiment, (b) emotion, and (c) tone for each opinion that both concurred and dissented.
9.3.3 Note any image(s), figure(s), table(s), or charts in each opinion that both concurred and dissented, and (a) repeat any text-based caption, and (b) describe each image, figure, table, or chart as they appear.
9.3.4 Note any dictionary references and (a) list each dictionary name, and (b) repeat any text-based term and the cited dictionary definition.10.0 Nonparticipating jurists
10.1 Name of each judge who did not participate in the decision and outline their reason(s) why:11.0 Additional comments or impressions: (Is the reasoning sound and, if so, how? Is the reasoning contradictory and, if so, how? What could be the decision’s political, economic, or social effects, if any?)
Now bold the labels.
Claude.ai generated an accurate outline
Claude did a great job. I am very satisfied with this output.
ChatGPT hallucinated, twice
I first thought that I must have uploaded the wrong file when reading ChatGPT’s response. ChatGPT’s response was completely made-up garbage. The response starts on page 4.
So I created a new chat and repeated the prompt. The second round of output was still made-up garbage, but now with a whole new set of “details”. The response begins on page 4.
Thoughts
Earlier today, ChatGPT impressed me for catching the most spelling and punctuation errors on a test sentence. For that type of task, ChatGPT did better than Claude.ai.
But it seems like it’s a different ballgame if the task involves working from a supplied attachment.
In this space for this experiment, Claude.ai did a really good job.
ChatGPT’s responses were made-up disasters.