An unimpressed Manhattan judge ordered a corporate representative for OpenAI to undergo a second deposition after finding he failed to answer “even the simplest questions” the first time around about what the company has described as efforts to limit chatbots from stealing writers’ work.
Magistrate Judge Ona Wang, in a sharply-worded 11-page order Tuesday, said OpenAI had been put on notice that the company’s purported expert on plagiarism John Vincent “Vinnie” Monaco was woefully underprepared for his January deposition, ordering him to submit to 3.5 more hours of questioning that took place Wednesday.
In granting a motion from the New York Times, Daily News and other news outlets suing OpenAI to compel the additional testimony, Wang deferred ruling on a request for sanctions, saying it would depend on how Monaco fared in his do-over. She said she may issue fines or recommend some of his answers be deemed as admissions.
OpenAI has previously said that Monaco has more knowledge than any of its engineers about Project Giraffe, an internal operation which the company claims is designed to develop ways to limit its learning language models, or LLMs, from inadvertently regurgitating copyrighted works — the issue at the core of the ongoing Manhattan Federal Court lawsuit.
The news organizations allege OpenAI is stealing and distorting their copyrighted works.
Wang said her review of the record found Monaco didn’t display any special knowledge, bringing nothing to his deposition but his own “hazy recollections.”
“Mr. Monaco based his testimony — which should bind the corporation — on his own hazy recollections working on Project Giraffe, a review of source code that he and others wrote (but which he does not remember), and a few other documents,” the judge wrote Tuesday.
Wang added, “He spoke to no other individuals to prepare for his deposition. But when asked even the simplest questions relevant to Plaintiffs’ output claims, Mr. Monaco failed to answer or provided incomplete or evasive answers,” and noted OpenAI’s attorney had similarly “impeded and frustrated” the deposition by objecting at least 200 times.
In a February filing, lawyers for the newspapers said they gleaned from Monaco’s Jan. 27 and 28 deposition that the tech company had been searching for copyrighted works in its training data and responses to user queries for some time.
They alleged the company had been engaged in a practice of “covering up its infringement by adding a filter to ChatGPT to block the regurgitation in real-time” and that OpenAI should have produced the information years ago.
OpenAI has countered that Project Giraffe runs tests and evaluations but denies it has the ability to search all of its responses to users for training data.
The complex lawsuit filed by the Times, the News and affiliated outlets in Tribune Publishing and MediaNews Group accuses OpenAI and Microsoft of stealing and distorting the work of human journalists, often giving its users — more and more of whom are turning to AI to get their news — an incomplete or inaccurate rehashing.
The Authors Guild and a litany of best-selling writers are also parties to the litigation.
The News reached out to lawyers for both sides for comment.