OpenAI has secured $40 billion, catapulting its valuation to $300 billion.
Reported as one of the largest private funding rounds in history, OpenAI aims to further advance AI research, scale its computing infrastructure, and enhance the capabilities of ChatGPT, which now serves over 500 million users weekly.
“We’re thrilled to collaborate with SoftBank Group, a company with unparalleled expertise in scaling groundbreaking technology,” OpenAI stated in its blog post. TechCrunch and CNBC notes that the new funding is also expected to support OpenAI’s ambitious Stargate project, which focuses on building a network of high-powered AI data centers across the U.S.
before getting too swept up in the victory lap, there’s another story brewing…
OpenAI Accused of Training GPT-4o on Paywalled Content Without Permission
A new study by the AI Disclosures Project, a nonprofit watchdog, claims OpenAI may have trained its GPT-4o model using paywalled books from O’Reilly Media without authorization, as reported by Indianexpress.
The research suggests GPT-4o exhibits a stronger recognition of non-public O’Reilly content compared to earlier models like GPT-3.5 Turbo, which showed a higher affinity for publicly available samples. The study found no licensing agreement between OpenAI and O’Reilly Media.
These allegations emerge as OpenAI faces multiple lawsuits over its data training practices, with critics arguing they amount to copyright infringement. Researchers used a technique called “membership inference attack” (DE-COP) to assess whether GPT-4o had prior knowledge of specific texts, testing 13,962 excerpts from 34 O’Reilly books. Findings indicated GPT-4o recognized more paywalled content than its predecessors, even after accounting for improved model capabilities.
However, the study acknowledges limitations, including the possibility that users may have submitted paywalled excerpts in ChatGPT prompts.
As OpenAI and Google continue lobbying for AI training on copyrighted content to be classified under “fair use,” OpenAI has also been striking post-hoc licensing deals with news publishers, social platforms, and stock media providers.
