Sam Altman's talk was undoubtedly the most packed session at last week's Y Combinator Alumni Reunion, and for good reason. The vast majority of YC companies, including ours, are building on top of the OpenAI platform. It would be beneficial for us to understand precisely what the mastermind behind the operation is thinking in terms of developing the platform further.
An important question that Dalton asked Sam was about the inherent platform risks behind OpenAI, which was certainly a question that we were all contemplating. Apple is notorious for its high level of platform risk. Here is an article about apps that Apple discontinued with just one keynote
Thankfully, Sam was quite honest when it came to what we probably should not be building. He mentioned a couple of important considerations to take into account.
Strategies to Minimize Redundancy with OpenAI's Development Roadmap
Building something that OpenAI will build anyway is clearly something to avoid, but he provided us with important insights on how to identify such areas. Things like making marginal improvements to the ChatGPT UI or addressing issues in the GPT models are likely aspects that OpenAI is highly interested in addressing in the future.
If you're working on a ChatGPT wrapper solely aimed at enhancing ChatGPT's user experience, you might face some platform risk. OpenAI is heavily invested in the long-term improvement of ChatGPT's user experience. However, from my perspective, building such a wrapper can offer valuable insights into the AI chat's UI/UX process. While it might be on OpenAI's radar, it may not necessarily be their top priority.
Similarly, if your main focus is addressing OpenAI's inaccuracies, your core business proposition could become obsolete as OpenAI continually advances its model's accuracy and cost-effectiveness. OpenAI is undoubtedly dedicated to long-term improvements in model accuracy and cost-efficiency. However, it's essential to note that there are various model options available in the market today, such as Antropic, each with its unique challenges. This concern primarily arises if you rely solely on covering OpenAI's specific idiosyncrasies.
Prioritizing AI Solutions for Targeted Challenges
He did mention some promising use cases for building on ChatGPT as a platform, particularly in the realm of specializing ChatGPT for handling specific types of problems. For instance, the creation of an AI instructor, akin to CGPGrey's Digital Aristotle, was suggested. Additionally, he raised the idea of developing an improved version of WebMD for diagnosing health issues, an area in which IBM Watson has made significant progress over the past decade.
He also discussed the potential of assisting companies in adopting AI, including the possibility of creating a private GPT. This is a field that OpenAI currently appears to be steering clear of, as they prioritize expanding the capabilities of their high-end models. Thus, the use case of training a smaller, less resource-intensive model that doesn't rely on extensive GPU computing power seems to be a reasonable direction to explore.
We've been hard at work crafting a solution that aligns perfectly with what you need – a completely private GPT tailored to your internal documents. If you're interested in bringing the full power of AI to your company within a secure platform, let's schedule a chat.
Make sure you sign up for our newsletter for future updates!
If any of this looks interesting for you, book a demo with us!