AI Roundup 047: NYT v. OpenAI
AI Roundup 047: NYT v. OpenAI
Friday, December 29.
NYT v. OpenAI
With less than five days left in the year, we have one last big piece of AI news: the New York Times sued OpenAI and Microsoft for copyright infringement.
Why it matters:
OpenAI is already fending off multiple IP lawsuits. This new one both adds the weight of the NYT, and surfaces new and more creative examples of GPT-4 spitting out exact copies of copyrighted content.
Unlike previous cases, the NYT is 1) listing Microsoft as a co-defendant, 2) including RAG (retrieval augmented generation) as an example of infringement, and 3) asking for GPT-4 and other infringing models to be entirely destroyed.
The key question in all of these lawsuits is whether training AI models constitutes "fair use" of copyrighted content - something that the courts will ultimately need to decide.
Where we're headed:
Another issue is that many AI companies don't disclose what datasets were even used as part of training, though a proposed bill from Congress could change that.
Yet the future likely involves paid partnerships with content owners, at least while there's a legal gray area. OpenAI has struck new agreements, and Apple reportedly discussed multiyear deals with NBC, Condé Nast, IAC, and other organizations.
But there is some potential collaboration between authors and AI makers - the American Journalism Project launched a new initiative this year with OpenAI to help apply emerging tech to local journalism.
Elsewhere in AI anxiety:
AI is routinely used to tell children bedtime stories, via apps like Bluey-GPT, Oscar, Once Upon a Bot, and Bedtimestory.ai.
Hollywood unions battle with AI recreations sets a precedent for pushback against automation - but will it matter in the long run?
And Microsoft's Image Creator has been caught generating realistic and violent images of world leaders.
Wartime AI
A new investigation details how China's Ministry of State Security (the country's main intelligence agency) built an AI system to track US spies in its effort to challenge the CIA.
Between the lines:
It is not surprising that governments are building AI-powered intelligence systems - it's surprising that we get to hear about them.
That said, most military use cases of AI are still predictive. In the coming years, we're going to see far more uses of generative AI, for good and bad: on-demand propaganda, real-time refugee translations, and more.
And until then, it's a free-for-all between AI contractors and defense startups - as companies like Rebellion Defense struggle, others like Shield AI are poised to raise at a $2.8B valuation.
Things happen
Researchers are increasingly using Harry Potter to test AI. EU VP says AI Act will "not harm innovation and research". The NHS used AI kettles and fridges to reduce hospital readmissions. Apple exec leaves to join Jony Ive and Sam Altman on AI hardware project. 2023: The Year of AI. OpenAI's Cookbooks. Dark Visitors - a list of known AI agents. Microsoft's next Surface laptops are "true AI PCs". Flood monitoring companies are using AI to minimize damage. Anthropic projects $850M in ARR next year. Microsoft Copilot is now available on Android. Son uses AI to resurrect his late father.
