OpenAI has started offering PDF scanning and summarization to its customers. While that isn’t remarkable, it shows that OpenAI lacks a clear strategy for what it wants to accomplish as a company.
Let’s remember that OpenAI made the central part of their profit by providing API access to their GPT family of Large Language Models (LLM). That left others to build the products companies and end customers used to interact with AI. The most lucrative of these customers were the ones that did document summarization. Thus, OpenAI started competing with their most lucrative customers.
That leaves the question of whether OpenAI is changing its business strategy to offer more end products. It also makes other companies worried that their API supplier will become a competitor. Let’s have a look.
How to punish Businesses
Running a large language model seems simple on the surface. You ask a question, and it answers. At least on the surface, that’s what OpenAI looked like to most consumers.
However, the chat feature is a marketing toy to show the capabilities. The actual capabilities lie in the API, which allows less restricted LLM access. The API access allowed domain-specific responses, uploading data for summarization, or performing multi-step actions in a single command.
Consequently, others used the API to build productive tools with their custom interfaces. That way, the tool could present a streamlined interface with the full power of OpenAI behind it. Content summarization and support chatbots are two of the simplest examples. Others have built more complex integrations.
Thus, the implicit agreement was that OpenAI builds the backend and language processing, and others build the consumer tools. With the Document summarization tool, OpenAI has broken that agreement by releasing a DTC tool.
Have Businesses Done That Before?
When looking at the case, the antitrust complaints against Amazon immediately came to mind. It alleges that Amazon used the data from its third-party sellers to see what sells well. Once they saw what sold well, they mimicked the product as part of their Private Lable series, like Goodthreads, and started selling the products.
Consequently, they reduced the risk of failing. If true, it also would mean they are screwing over their business partners.
Whether or not OpenAI is doing the same remains to be seen.
Further Risks for Businesses
However, whether OpenAI is copying ideas isn’t the only problem. With the strategy in flux, it is unclear whether OpenAI will change its stand on providing API access to its competitors and whether it will do so at a price that enables anyone to build a competitive product.
There is no indication that OpenAI will change its stand on API access. Yet, the risk of losing the leading supplier will remain the sword of Damocles hanging over any AI-based startup.
The Open-Source Alternative
Resolving it will take some time and careful risk analysis for many companies. What seems to become a part of this is that you cannot build an AI business on someone else’s APIs. This insight will remain especially true for companies that haven’t mapped out their path to profitability and supplying a core component of a business.
Staying in control of the central innovations within a company should be one of the lessons learned. With the increasing numbers of open-source AIs, running different AI models is possible. Taking them into IaaS environments or private clouds can be especially significant for growing companies with high utilization. In these cases, it might be not just a survival but also a cost question.
No Business on Simple Ideas
The other lesson learned is that simple ideas must be a stepping stone, not the complete business. For many of the companies that are now on the brink, PDF summarization was their calling. They thought they found a golden nugget on the surface and could make hundreds of gold bars.
Yet, even without OpenAI offering a competitive project, it should be clear that it was only a question of time until one of the big players with full office suits integrated the capability into their offerings. If any company had taken the initial windfall and continued to enhance their product by providing further analytical functions or the ability to generate more than a summary, they could have stayed ahead.
The more complex the product, the more difficult it would have been to copy. Likewise, the more utility a product has (within Reason), the more likely it would have been that the customers return.
Return customers might have made it an attractive takeover target for office suit providers looking to enhance their customer base and capabilities.
Business is Innovation and Risk Management
This situation should be a warning call for founders and businesses building on AI. Just because you created your idea on AI doesn’t mean you can neglect innovation and risk management. An easy-to-copy idea built on someone else’s back invites copycats, including those carrying a significant part of the weight.
It also shows that a company searching for a business model isn’t the best partner. The question might not have arisen if OpenAI had a profitable history of selling API access.
Whether or not OpenAI destroyed their prospects of success remains to be seen. Trust, once lost, is hard to regain. It is a gift to open source and self-hosted AI.