LLM Prediction Capabilities Match Human Accuracy
-A crowd of 12 LLMs vs a crowd of 925 human forecasters on a 3-month forecasting tournament
-LLM crowd is statistically equivalent to the human crowd
-Replicates the "wisdom of the crowd" effect for LLMs
https://t.co/8VIfooJhGX
The Vanderbilt Intellectual Property Program is excited to announce a major event in music, copyright, and AI on April 4th at Vanderbilt LawSchool. (Free) registration required.
#CLE available. @vanderbiltlaw@Vanderbilt_IP
Check out the details here https://t.co/D1UnYW8Dgs
We are very pleased to share that the website for the ATRIP 2024 Congress in Rome (30.6-3.7.2024), is now online: https://t.co/M8B03qSQpW
The website includes the call for papers, as well as the registration options. We look forward to welcoming you in Rome soon!
@ednewtonrex I find it helpful to separate change from progress and limit progress to positive change. Change is a difference between points A and B on a timeline. Progress is an improvement at point B measured not in $ but in human progress.
Are you in London? If so, don’t miss the world premiere of a play based on my legal sci-fi book “Forever” at Bentham House (UCL) on Tuesday 13 February at 1315, part of a great program on privacy. See details here: https://t.co/5NskgcGCa2.
(Almost a) final call for young IP scholars - our Essay Competition is still open for additional submissions until February 15th, 2024. See the full details and guidelines here: https://t.co/8TQ0LtFyEi
@OrinKerr A great question. In 2021 https://t.co/D5Ld7XQx3r I wrote “I expect an AI written law review article to be accepted for publication (as a test) within two years.” The comments suggested I was crazy. Maybe I wasn’t after all, but that does *not* make me happy.
The EU AI Act full text was leaked, and it reinforces several points:
- General-purpose AI models will have to share summaries of training data used
- These will have to be detailed enough for rights-holders to understand when their work has been trained on
- Open models won't be exempt
- Downstream changes e.g. fine-tuning will also need to share training data used
- This all applies to any model made available on the EU market, wherever it was trained
- Enforcement will be more lenient for SMEs, but they must still respect copyright law
- The existing right of rights holders to opt out of AI training is unchanged
https://t.co/jXc1rFQDmZ
Can artificial intelligence run an LLC by itself? Daniel Gervais (@DanielGervaisIP) joins @NPR's @AcademicMinute and examines the laws of the land to find out.
https://t.co/adTfTHTNJR
Join us for the US book launch of Forever, a legal sci-fi story, part one of the Coexistence Trilogy tomorrow 11 January at @vanderbiltlaw in Flynn Auditorium at 3:30 PM.
AI Timelines Shorter
-2,778 AI researchers surveyed
-By 2028, 50% chance of AI autonomously building payment processing site from scratch, and downloading & fine-tuning LLMs
-50% chance of unaided AI outperforming humans in every possible task by 2047 (13 yrs earlier than last)