3 ms·

Highlights from this fantastic acceptance of judicial reality (Roberts') [1]: >In 1905, Justice David Brewer somewhat ungenerously referred to his law clerk as

Highlights from this fantastic acceptance of judicial reality (Roberts') [1]: >In 1905, Justice David Brewer somewhat ungenerously referred to his law clerk as "a typewriter, a fountain pen, used by the judge to facilitate his work." ... >Law professors report with both awe and angst that AI apparently can earn Bs on law school assignments and even pass the bar exam. LEGAL RESEARCH MAY SOON BE UNIMAGINABLE WITHOUT IT. [Emphasis MINE]. I contend that it is already unethical for doctors/lawyers to NOT be utilizing LLM-like systems] >..."hallucinations," which caused the lawyers using the application to submit briefs with citations to non-existent cases (always a bad idea). HA HA HA HA — I love a witty justice, even one without humility. >MACHINES CANNOT FULLY REPLACE KEY ACTORS IN COURT [Emphasis MINE]. Not YET, Chief. >I predict that judicial work — particularly at the trial level — will be significantly affected by AI. MF'n Duh! >Best wishes to all in the New Year ...cause corporate profits will rise, and junior-level clerkships might become less-existant. ¢¢ [1] PDF: https://www.supremecourt.gov/publicinfo/year-end/2023year-endreport.pdf https://www.supremecourt.gov/publicinfo/year-end/2023year-en...

anon373839 3y ago
> I contend that it is already unethical for doctors/lawyers to NOT be utilizing LLM-like systems] Bah. I was working on an appellate matter recently that involved some nuanced applications of well-known precedents. I tried to use GPT-4 to assist in exploring the issues, but I found it useless. It was unable to engage in a sophisticated discussion and instead preferred to speak in canned generalities. For what GPT-4 was offering, a quick trip to a bookmarked treatise on Westlaw would have been just as easy, and trustworthy.
- ProllyInfamous 3y ago
  Since early 2023, GPT-4 has been heavily nerfed (..."as a large language model"...), and is not niche-specific (in this case: legal training). Roberts is hinting at the future, one where trials and discovery are dominated with industry-specific datasets. >I tried to use GPT-4 to assist in exploring the issues, but I found it useless. These systems are mirrors, and you have yet to figure out the correct prompting. Once "PACER-AGI" becomes commercially-viable (IMHO, the major hold-up is liability from hallucinations), you will see an even-more-alarming rate of judicial redundancy.
  - pests 3y ago
    > you have yet to figure out correctly promoting Nope, you're holding the phone wrong.
  - anon373839 3y ago
    I actually work with LLMs extensively, and have been for several years now. I suspect domain adaptation could help here, but I am convinced that the issue I experienced is a more fundamental limitation: I don’t think this technology is inherently suited to the task of innovative logical analysis over novel issues. The same problem can be observed with programming tasks: LLMs can crank out accurate code to solve common problems, but they fail at truly original tasks. I suspect interpolation in the latent space just isn’t sufficient to do the job.
    - ProllyInfamous 3y ago
      My absolute agreement (with Roberts) is that legal research will be forever-changed. Most of the "innovative logical analysis over novel issues" that I have personally witnessed/directed/felt... has to do with comparing dead/living authors' POVs, between their fictional universes. Of course, this is a reflection of humanity, having elsewhere made such comparisons... but the ability to harness the patterns/connnections of just SO MANY unique POVs — is absolutely BEYOND HUMAN. -- Thanks for your informative perspective, and for sharing =D

anon373839 3y ago

> I contend that it is already unethical for doctors/lawyers to NOT be utilizing LLM-like systems] Bah. I was working on an appellate matter recently that involved some nuanced applications of well-known precedents. I tried to use GPT-4 to assist in exploring the issues, but I found it useless. It was unable to engage in a sophisticated discussion and instead preferred to speak in canned generalities. For what GPT-4 was offering, a quick trip to a bookmarked treatise on Westlaw would have been just as easy, and trustworthy.

ProllyInfamous 3y ago
Since early 2023, GPT-4 has been heavily nerfed (..."as a large language model"...), and is not niche-specific (in this case: legal training). Roberts is hinting at the future, one where trials and discovery are dominated with industry-specific datasets. >I tried to use GPT-4 to assist in exploring the issues, but I found it useless. These systems are mirrors, and you have yet to figure out the correct prompting. Once "PACER-AGI" becomes commercially-viable (IMHO, the major hold-up is liability from hallucinations), you will see an even-more-alarming rate of judicial redundancy.
- pests 3y ago
  > you have yet to figure out correctly promoting Nope, you're holding the phone wrong.
- anon373839 3y ago
  I actually work with LLMs extensively, and have been for several years now. I suspect domain adaptation could help here, but I am convinced that the issue I experienced is a more fundamental limitation: I don’t think this technology is inherently suited to the task of innovative logical analysis over novel issues. The same problem can be observed with programming tasks: LLMs can crank out accurate code to solve common problems, but they fail at truly original tasks. I suspect interpolation in the latent space just isn’t sufficient to do the job.
  - ProllyInfamous 3y ago
    My absolute agreement (with Roberts) is that legal research will be forever-changed. Most of the "innovative logical analysis over novel issues" that I have personally witnessed/directed/felt... has to do with comparing dead/living authors' POVs, between their fictional universes. Of course, this is a reflection of humanity, having elsewhere made such comparisons... but the ability to harness the patterns/connnections of just SO MANY unique POVs — is absolutely BEYOND HUMAN. -- Thanks for your informative perspective, and for sharing =D