My TEDx Talk is out! I discuss:
1) #music #copyright #lawsuits
2) how making (#bruteforce) all possible #melodies helps innocent © defendants
3) how can improve.
Here's to helping #songwriters like & create!
Pinned Tweet
Show this thread
Follow
Damien Riehl — @damienriehl@law.builders
@damienriehl
Lawyer, technologist, educator, musician, speaker. Litigator for ~15 years, coding 1985+, web 1995+, also cybersecurity. Now AI-backed tools .
Damien Riehl — @damienriehl@law.builders’s Tweets
If you want to test the waters of the *other* network, this tool makes it easy — and on Tuesday, it'll go away.
Takes a few minutes. It'll make your life better.
fedifinder.glitch.me
2
A new AI lawsuit, brought by Getty images (not yet docketed). Getty is suing for trademark infringement, not just for the AI using the Getty watermark, but it’s use with “bizarre” and “grotesque” AI output: twitter.com/michaelkasdan/
This Tweet is unavailable.
1
2
4
Show this thread
A *Code+Law+LLM* Research Agenda-
LLM classifies laws into:
1. mainly rigid rule-based (tax, permitting, etc)
2. mainly standards-based (fiduciary, "reasonable", etc)
-For 1: LLM generates computer code that humans validate.
-For 2: LLM is fine-tuned to better exhibit standard.
3
13
71
Show this thread
Another, focused on legal verbosity:
PROMPT: Below is a verbose sentence. Simplify its structure, keeping the same legal meaning.
1
1
Show this thread
Another good example:
1
1
Show this thread
Good piece from on the state of automated tools (e.g., #AI) finding #cybersecurity vulnerabilities:
schneier.com/blog/archives/
Show this thread
A nicely articulated framework on how to think about UX for mitigating hallucinations in domains where accuracy is important.
Quote Tweet
I've been thinking a lot about this — #GPT and #LLMs hallucinating both (1) propositions and (2) citations — and to combat it, I think there are (at least) three options:
1. Bullshitter
2. Searcher
3. Researcher
Details in this thread...
1/n twitter.com/samuelharden/s…
Show this thread
1
2
An #AI model has been generating a episode on since Dec. 14 — using #GPT (#GPT3) — which raises #copyright questions:
1. INPUT: Under 2d Cir. "Author's Guild" case, #fairuse?
2. OUTPUT: "Style of" ? Copyrightable?
1
4
Which to choose?
#1 BULLSHITTER is a nonstarter.
#2 SEARCHER seems obvious. But many rabbit holes.
e.g., sentence = BS hallucinated
e.g., sentence recites bad law (e.g., Plessy, Roe v. Wade)
#3 RESEARCHER is harder to build. But most trustworthy. Built atop ground truth
5/n
1
3
Show this thread
3. RESEARCHER. Atomize the <PROPOSITION> + <CITATION> graph. Give each <PROPOSITION> and <CITATION> unique identifiers. User's query builds most-common ground truth (non-hallucinated).
4/n
1
2
Show this thread
2. SEARCHER. LLM generates text. Run queries to substantiate (or debunk) the text.
This is like a senior partner saying "I'm pretty sure there's a case out there that says X. Find it!"
Good luck.
3/n
1
Show this thread
1. BULLSHITER. LLM hallucinations gone wild. Unchecked chaos.
2/n
2
Show this thread
Quote Tweet
Can confirm that ChatGPT hallucinates case names and citations. Two big problems:
1. These *look* correct - the format is right, the courts exist, etc.
2. It does this confidently and without any context clues that it's fabricating them twitter.com/damienriehl/st…
Show this thread
1
3
6
Show this thread
Docket Alarm Now Uses GPT-3 To Show You Summaries Of PDF Litigation Filings As You Review Docket Sheets
buff.ly/3DsFcL6
AI summaries help users triage documents to review
"Users have to balance the utility of the function against its level of accuracy, "
2
1
Not good when standard tools can brute-force 1/6 of your workforce:
"Within the first 90 minutes, the watchdog was able to recover nearly 14,000 employee passwords, or about 16% of all department accounts, including passwords like ‘Polar_bear65’ and ‘Nationalparks2014!’."
1
Show this thread
A reminder to use strong passwords — unlike some at the U.S. Dept. of :
"More than a fifth of the passwords protecting network accounts at the US Department of the Interior—including Password1234, Password1234!, and ChangeItN0w!"
#Security
schneier.com/blog/archives/
1
1
Show this thread
Here’s a great example from team showing another useful way GPT can be used to improve a key legal use case. Kudos to the crew for getting this out to market quickly! twitter.com/speedplane/sta
This Tweet is unavailable.
2
5
I know the headlines these days are all about ChatGPT replacing lawyers but all that overshadows the bigger story about the changing nature of law work. And it’s happening right now in corporate legal.
12
23
147
Show this thread
New feature from uses GPT-3 to show you summaries of PDF litigation filings as you review court docket sheets. I get a demo from and . #legaltech
6
12
GPT (large language model Davinci) now available in by via w HT
"The feature contains strong warnings around relying on its AI-generated summaries."
Nonetheless, an exciting development that users can try out a LLM twitter.com/speedplane/sta
This Tweet is unavailable.
2
8
Fascinating real-world integration of ChatGPT into an existing #legaltech product. One of the things it’s REALLY good at is summarizing - so this is smart. Nice work, ! twitter.com/speedplane/sta
This Tweet is unavailable.
1
2
12
Show this thread
The incomparable does some testing of #ChatGPT as #Legal Brief writing Tool.
linkedin.com/pulse/chatgpt-
2
3
Quote Tweet
Today Docket Alarm is releasing its first (and perhaps the legal tech industry’s first) integration with GPT3. All litigation filings on Docket Alarm (e.g., PACER, state courts, SCOTUS, IPRs, etc.) can now be auto-summarized into 3 easy bullets by GPT3.
Show this thread
0:44
1.8K views
2
1
13
Oh, and #ChatGPT will provide exemplar facts:
PROMPT: For this factual claim — "OpenAI's actions were the direct cause of Plaintiffs' injuries" — provide factual examples of how a large-language model on training text would cause an author of that training text to lose money.
2
Show this thread
Adventures in #ChatGPT for #law:
Craft #legal #arguments responding to 's Motion to Dismiss the #CoPilot litigation.
docketalarm.com/cases/Californ
1. Use Table of Contents to set a baseline — then create counterarguments:
1/n
3
2
14
Show this thread
Study by from — where #ChatGPT passed all 4 law-school exams.
But #LLM did score low:
- CON LAW: B (36th/40 students)
- EMPLOYEE BENEFITS: B- (18th/19)
- TAX: C- (66th/67)
- TORTS: C- (75th/75).
Complementary to .
papers.ssrn.com/sol3/papers.cf
1
This Tweet is unavailable.
1
1
16
Hype machine is certainly alive an active, but where are the features and use-cases? Will be doing a small reveal at the IRL event tonight.
1
1
3
This simple thing has taken YEARS and a pandemic. Kudos to Fix the Court for pushing on this issue over and over and over and over and over again. LIVE STREAMED ORAL ARGUMENTS FOR EVERYBODY! twitter.com/FixTheCourt/st
This Tweet is unavailable.
1
9
27
Show this thread