Lawyer Ex Machina #35: Happy Stochastic Parrots Day
AI
The Copyright Office has issued new guidance on "works containing material generated by artificial intelligence."
Bloomberg Tax [($)]: PricewaterhouseCoopers has introduced an AI-enhanced chatbot developed by Harvey.ai to be used by its in-house lawyers to "speed up work from due diligence or regulatory compliance," among other tasks. [Reuters]
The Cybersecurity, Information Technology, and Government Innovation subcommittee held a Congressional hearing on "Advances in AI: Are We Ready For a Tech Revolution?"
From NBC News: “A deepfake app advertised itself on Meta platforms using faces of actors [Emma] Watson and Scarlett Johansson.”
At least 20 sports stadiums around the country have adopted facial recognition systems, for alcohol concessions, restricting access to training facilities, and general gate entry, among other uses.
Microsoft, a major investor in OpenAI who plans to integrate generative AI technology into a number of products, has laid off every employee involved in the company's Ethics and Society Team.
Here is a deep-dive explainer on how social media recommendations algorithms work.
GPT-4, the next iteration of OpenAI's large language model, is now public.
One of the benchmarks of GPT-4's improvement over 3.5/ChatGPT? It now scores within the top 10% on a Uniform Bar Exam.
How do you make GenAI products produce legal materials? Legal prompt engineering may help.
One reaction to GPT-4 and 2 follow-ups:
- https://mobile.twitter.com/jbrowder1/status/1635720431091974157
- https://mobile.twitter.com/DKThomp/status/1635728028281044993
- https://mobile.twitter.com/hamandcheese/status/1635734186718887947
Data Privacy
From Politico: "The privacy loophole in your doorbell"
FBI Director Christopher Wray admitted last week that the agency had previously purchased location data on Americans without a warrant.
A Catholic lay group bought mobile app tracking data to find priests who used dating and hooking apps, and report them to bishops.
Companies are increasingly concerned about employees feeding sensitive/confidential data into large language models like ChatGPT, which could be regurgitated verbatim, given the right prompts.
Miscellaneous
The Free Law Project has bundled a number of tools for accessing and working with federal docket information into The RECAP Suite.
From 3 Geeks and a Law Blog: "Next-Gen Bar Exam Must Tackle Google Schools and the Digital Native Myth by Testing Basic Tech Skills for Practice"
Events
Friday, March 17th, 8 am - noon PDT: Stochastic Parrots Day : an online conference for the 2 year anniversary of the paper "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big." Panelists include Emily M. Bender, Timnet Gebru, Safiya Noble and more.