-
Ilya Sutskever speech on receiving honorary degree from University of Toronto
Read more: Ilya Sutskever speech on receiving honorary degree from University of Toronto“At the University of Toronto when I was a student here, we were doing the best AI research [with Geoffrey Hinton] out of anywhere. The most revolutionary ideas, the most exciting work.” Ilya Sutskever. Here’s Sutskever talking about living in the most unusual time ever, with the advent of AI: The day will come that…
-
OpenAI proposes 217 hours of depositions for OpenAI witnesses. Plaintiffs want 255 more hours. Open says Plaintiffs asked irrelevant questions, including: What’s your least favorite Joyce novel?
Read more: OpenAI proposes 217 hours of depositions for OpenAI witnesses. Plaintiffs want 255 more hours. Open says Plaintiffs asked irrelevant questions, including: What’s your least favorite Joyce novel?Interesting disagreement in the MDL litigation, In re OpenAI Copyright Infringement Litigation, over the number of hours the plaintiffs should get for the depositions of OpenAI witnesses. One of the greatest efficiencies of MDL Litigation is the streamlining of depositions and avoidance of repetitive depositions of the same witness. Plaintiffs want 255 hours, not including…
-
LLMs learn more when given incorrect answers to MC questions
Read more: LLMs learn more when given incorrect answers to MC questionsShadi Hamdan and Deniz Yuret posted a preprint paper titled “How much do LLMs learn from negative examples?“ Abstract: Large language models (LLMs) undergo a three-phase training process: unsupervised pre-training, supervised fine-tuning (SFT), and learning from human feedback (RLHF/DPO). Notably, it is during the final phase that these models are exposed to negative examples —…
-
Apple researchers cast doubt on AI reasoning models of other companies
Read more: Apple researchers cast doubt on AI reasoning models of other companiesAs Apple struggles to integrate AI into Siri and its other software, Apple researchers posted a controversial paper provocatively titled: “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity.” The researchers are: Parshin Shojaee*†, Iman Mirzadeh*, Keivan Alizadeh, Maxwell Horton, Samy Bengio, and Mehrdad Farajtabar. First…
-
First copyright trial for AI company set for June 9 in UK High Court. Getty Images v. Stability AI.
Read more: First copyright trial for AI company set for June 9 in UK High Court. Getty Images v. Stability AI.The irony of ironies is that, despite the 41 copyright lawsuits against AI companies filed in the United States, the UK High Court will hold the first trial involving AI generators in Getty Images v. Stability AI. The trial starts today, June 9, before Justice Joanna Smith, and is slated for 3 weeks. From the…
-
Scholarship: How courts should treat AI models’ memorization of some content from training datasets
Read more: Scholarship: How courts should treat AI models’ memorization of some content from training datasetsAI researchers are studying the phenomenon that AI models unintentionally memorize some of the content from the training datasets. Unintentional Memorization of Some Training Content A recent paper “How much do language models memorize” by researchers from Meta, Google DeepMind, Cornell, and NVIDIA (John X. Morris, Chawin Sitawarin, Chuan Guo, Narine Kokhlikyan, G. Edward Suh,…
-
Sarah Andersen v. Stability AI parties dispute over Dr. Ben Zhao’s serving as expert, despite making tools to poison data for AI image models
Read more: Sarah Andersen v. Stability AI parties dispute over Dr. Ben Zhao’s serving as expert, despite making tools to poison data for AI image modelsBelow are the competing letters the parties in the Sarah Andersen v. Stability AI lawsuit filed. The controversy is that the Dr. Ben Zhao of University of Chicago is one of the researchers behind Nightshade, Glaze, and other adversarial tools used to “poison” the data for image generators so that they fail at creating images…
-
Hearing whether to disqualify Dr. Ben Zhao as plaintiffs’ expert in Andersen v. Stability postponed until June 10
Read more: Hearing whether to disqualify Dr. Ben Zhao as plaintiffs’ expert in Andersen v. Stability postponed until June 10No hearing today. Order: Notice is hereby given that the Discovery Hearing previously set for June 6, 2025 is CONTINUED to June 10, 2025 at 9:00 a.m. (PT) via Zoom Webinar before Magistrate Judge Lisa J. Cisneros. From Judge Cisneros’s website: Public Hearings Please click the link below to join webinars (public hearings). If you…
-
Scholarship: How should courts weigh AI companies’ use of shadow libraries of “pirated” books?
Read more: Scholarship: How should courts weigh AI companies’ use of shadow libraries of “pirated” books?One of the most controversial aspects in the copyright lawsuits against AI companies is how courts should weigh the companies’ use of controversial “shadow libraries” consisting of unauthorized or “pirated” copies of books others have compiled and made available online. These shadow libraries have provoked condemnation from book authors and a fair amount of media…