Maybe there's room in legal tech for both #determinism and #probabilism.

Link to Thinking Machines' paper:
thinkingmachines.ai/blog/defeati...

The best lawyers I know are creative. They eschew the "most likely next token" — favoring out-of-distribution legal approaches and arguments. Some legal tasks have the "right answer"; many tasks, though, have "an answer so creative that it just might work."

Perplexity extrapolation of Claude data → percentages in Claude's Job Explorer:
www.perplexity.ai/search/how-m...

Claude's "Explore by Job" data:
www.anthropic.com/economic-ind...

Some notes and questions:
- That's just one model (Claude)
- How many "legal profession" queries in (1) Gemini/ Google, (2) ChatGPT, (3) Grok, (4) CoPilot?

#LegalConsumer queries? E.g., #Client queries? #WouldBeClients didn't hire lawyers because LLM output = "good enough"

Does an AI disclosure rule help any more than Rule 11 (or its state equivalents)? "Everything above my signature is true and correct" — whether created by AI, my first year associate, or my paralegal. AI "pinky swears" don't change behavior.

"It looks like you're filing a motion to dismiss. Would you like some help?"

Exactly: 🎯 . Exciting times! 🚀

Interesting — I had a nice lunch with someone from the Center for Humane Technology, which helped with this case. Thanks for passing this along!

...or on your favorite podcatcher:
kite.link/damien_riehl

Dark days, here in Minnesota. 🙁 These were Democrats: How will POTUS respond? 🤷