Recently, I hosted a LinkedIn Live where we took a deep dive into OpenAI’s growing patent portfolio. I’m excited to share the key insights and takeaways from that session with you. Let’s explore OpenAI’s patent filing strategy, the technical specifics of their patents, and what might be coming next:
OpenAI’s Patent Filing Strategy
Until not too long ago, if you searched for patents by OpenAI, you would have come up empty-handed. But things have changed. Right now (23 May 2024), OpenAI has seven published patent families, including six freshly issued US patents and two more US applications that are just about to be approved:
The speed at which these patents have been granted is impressive, averaging only 11 months from filing to grant. This rapid pace is thanks to OpenAI’s savvy use of the USPTO’s Track One program, which speeds up the examination process for an additional fee. In the LinkedIn Live, I wrongly mentioned 1,500.- USD, but an attentative participant corrected me:
What’s really interesting is how OpenAI’s approach differs from the norm. Typically, US deep-tech companies start with US provisional applications followed by PCT applications to stretch out the patent prosecution timeline.
But OpenAI is doing things differently. They’re filing full US utility patent applications right off the bat and using the Track One program to get their patents granted quickly. So far, it’s all about the US market—no international filings are visible yet.
What’s Inside OpenAI’s Patents?
Let’s get into the meat of it—what exactly are these patents about? Here’s a quick rundown of some key patents:
- US 2024/020096 A1 & US 2024/020116 A1: These are about systems and methods for generating natural language using language models trained on computer code, likely related to code generation using large language models (LLMs).
- US 11,886,826 B1: This patent covers systems and methods for language model-based text insertion, enhancing text generation and editing capabilities.
- US 11,983 488 B1: Focuses on systems and methods for language model-based text editing.
- US 11,922,144 B1: Involves schema-based integration of external APIs with natural language applications, highlighting interoperability between LLMs and other software.
- US 11,922,550 B1: Covers systems and methods for hierarchical text-conditional image generation, which relates to advanced image generation techniques.
- US 11,887,367 B1: Utilizes machine learning to perform automatic interface actions based on video and input datasets, probably indicating steps towards autonomous AI agents.
- US 11,983,806 B1: This one’s about systems and methods for image generation with machine learning models, broadening the scope of AI-driven image creation.
From what a quick scan of the independent claims reveals, these patents don’t cover broad, fundamental breakthroughs but rather focus on specific use cases and workflows within the AI domain.
What’s Next for OpenAI’s Patent Portfolio?
Looking ahead, the potential for international filings remains speculative at the moment due to the typical 18-month publication lag. Assuming that OpenAI has filed PCT applications claiming the priority of the US first filings, the first such PCT application will be published only in September 2024:
Homemade Problems Down The Road?
Notably, all patents (except for one family) have been filed several months after the launch of ChatGPT in November 2022. The strategic timing of these patents—filed post-ChatGPT launch—raises some interesting questions.
Could OpenAI face challenges similar to the infamous “Steve Jobs patent mistake,” where pre-patent public disclosures led to invalidations in certain jurisdictions? The story goes like this:
During a keynote presentation, Steve Jobs demonstrated a feature known as the “rubber banding” or “overscroll bounce” effect, which is when a user scrolls to the end of a list on a touchscreen device, and the list bounces back. This public demonstration occurred before Apple filed patents covering this feature. While US patent law allowed for a one-year grace period for public disclosures by the inventor, other jurisdictions like Europe do not, requiring absolute novelty. Consequently, the public demonstration by Steve Jobs was considered prior art in Europe, ultimately leading to invalidation of the corresponding European patent when it was challenged in court.
OpenAI’s relatively late entry into the patent arena (most patents filed after March 2023) suggests a deliberate pivot towards securing intellectual property post-ChatGPT’s public release in November 2022. This approach ensures rapid protection in the US, but it will be interesting to see how their international strategy unfolds.
Conclusion
OpenAI’s rapid and targeted US patent filings signify a strategy to safeguard their innovations, primarily focusing on the US market with a streamlined, accelerated approach. As their patent portfolio evolves, I’ll be keeping a close eye on their moves for international filings to better understand their global IP strategy.
Here’s the full recording of the LinkedIn Live:
Stay tuned for further updates, especially around September and October, when we might see new international filings emerge. This ongoing journey into OpenAI’s patent landscape offers invaluable insights into the strategic maneuvers of a leading AI powerhouse.
If you enjoyed this deep dive, be sure to get on my email list for more updates and live events!
[…] of the most significant trends in OpenAI’s patent landscape is the expanding scope of the technologies being patented. Initially, OpenAI’s patents were […]