Introducing CopyrightCatcher, the first Copyright Detection API for LLMs; Patronus AI, March 6, 2024

Patronus AI; Introducing CopyrightCatcher, thefirst Copyright Detection API for LLMs

"Managing risks from unintended copyright infringement in LLM outputs should be a central focus for companies deploying LLMs in production.

• On an adversarial copyright test designed by Patronus AI researchers, we found that state-of-the-art LLMs generate copyrighted content at an alarmingly high rate 😱
• OpenAI’s GPT-4 produced copyrighted content on 44% of the prompts.
• Mistral’s Mixtral-8x7B-Instruct-v0.1 produced copyrighted content on 22% of the prompts.
• Anthropic’s Claude-2.1 produced copyrighted content on 8% of the prompts.
• Meta’s Llama-2-70b-chat produced copyrighted content on 10% of the prompts.
• Check out CopyrightCatcher, our solution to detect potential copyright violations in LLMs. Here’s the public demo, with open source model inference powered by Databricks Foundation Model APIs. 🔥

LLM training data often contains copyrighted works, and it is pretty easy to get an LLM to generate exact reproductions from these texts1. It is critical to catch these reproductions, since they pose significant legal and reputational risks for companies that build and use LLMs in production systems2. OpenAI, Anthropic, and Microsoft have all faced copyright lawsuits on LLM generations from authors3, music publishers4, and more recently, the New York Times5.

To check whether LLMs respond to your prompts with copyrighted text, you can use CopyrightCatcher. It detects when LLMs generate exact reproductions of content from text sources like books, and highlights any copyrighted text in LLM outputs. Check out our public CopyrightCatcher demo here!

Testing Ethical Boundaries. The New York Times Sues Microsoft And OpenAI On Copyright Concerns; Forbes, December 29, 2023

 Cindy Gordon, Forbes; Testing Ethical Boundaries. The New York Times Sues Microsoft And OpenAI On Copyright Concerns

"We have at least seen Apple announce an ethical approach to discussing upfront with the US Media giants their interest in partnering on AI generative AI training needs and finding new revenue sharing models.

Smart Move by Apple...

The court’s rulings here will be critical to advance ethical AI practices and guard rails on what is “fair” versus predatory.

We have too many leadership behaviors that encroach on others Intellectual Property (IP) and try to mask or muddy the authenticity of communication and sources of origination of ideas and content.

I for one will be following these cases closely and this also sends a wake -up call to all technology titans, and technology industry leaders that respect, integrity and transparency on operating practices need an ethical overhauling.

One of the important leadership behaviors is risk management and looking at all stakeholder views and appreciating the risks that can be incurred. I am keen to see how Apple approaches these dynamics to build a stronger ethical brand profile."

The Real Deal: Using Found Content ; Lexology, August 1, 2018

Frankfurt Kurnit Klein & Selz PC - Terri Seligman, Lexology; The Real Deal: Using Found Content

[Kip Currier: Informative article with tips on deciding how and when to use images found on the Net.

In my IP course I've shared the "teachable moment" story of a savvy business friend who was getting a start-up up and running about a decade ago and asked me "if it's OK to just scrape images from the Internet to use on the company's website?" You can anticipate my response, which always elicits a knowing laugh from the students--and reinforces the importance of considering potential copyright and risk management issues.]

"As reported by MediaPost, replacing the use of stock images with crowdsourced photos from real people is gaining popularity with major brands. The attraction is obvious: photos from real consumers can be more "authentic, local and real" than stock imagery.

But it's important to keep some rules of the road in mind to avoid the potential of liability for use of found content."

What Harley Davidson’s $19.2M Throttling Of Sunfrog REALLY Means… And It’s Not The Money; Above The Law, April 23, 2018

Tom Kulik, Above The Law; What Harley Davidson’s $19.2M Throttling Of Sunfrog REALLY Means… And It’s Not The Money

When it comes to intellectual property rights, companies ignoring their impact do so at their own risk.

"The point here is that rapid growth and success makes being proactive even more essential to the business.   Rather than follow-through with significant steps to stop the printing of infringing products, something got lost in the process and Sunfrog simply couldn’t get its arms around the scope of the problem.  In effect, Sunfrog’s failure to effectively address this problem  made Sunfrog a counterfeiter — it permitted the printing of infringing designs on T-shirts sold through its website, making Sunfrog a nice profit in the process. Of course, this was never Sunfrog’s intent — it set out to create a highly successful platform for printing custom T-shirts online, and in fact, succeeded in doing so.  That said, it also underestimated the extent to which a sizable part of its business model required intellectual property oversight — an oversight that is now costing them in both monetary and reputation damages.

Ultimately, the Sunfrog case is highly instructive on a number of levels, but the failure to appreciate the scope and extent of intellectual property oversight by Sunfrog is telling.  Whether your company or client is a startup or an already successful going concern, the use of intellectual property can never be taken for granted. When it comes to intellectual property rights, companies ignoring their impact do so at their own risk.  The good news is that warning signs usually present themselves at some point.  The bad news is that such signs can be ignored or otherwise under-appreciated.  That is the real point here, and a risk that your company (or client) shouldn’t take — just ask Sunfrog."