Discussions about ChatGPT with Sam Altman
Discussions about ChatGPT with Sam Altman
We were able to get some insights straight from the CEO of OpenAI on the future of ChatGPT as a product and the future of AI
Stable Discussion was able to attend a special event in Toronto with OpenAI’s CEO, Sam Altman. Sam is on his world listening tour to gain insight into how users are engaging with the AI models his company has developed. Toronto was the first stop on the tour.
In a fireside chat, I was able to ask a single question to Sam:
Ben: “Are you having challenges re-engaging customers who have tried and failed to accomplish a given task with a GPT model? Others besides early adopters fall into this category. What does re-engagement look like for an AI model, and have you put any thought to that problem?”
Sam: "We really haven't seen it as a problem for us. As we release new versions, there's enough buzz and sharing between our users that we haven't seen a falloff in user engagement."
While this seems like a fairly uninformative answer, it shows us a glimpse of how OpenAI is operating. We can assume how they’re thinking about the millions of users using their ChatGPT service. The answer also gives us insight into the continued focus of OpenAI. Given the context of the rest of the event and the focus of the listening tour, we get a pretty good idea of what OpenAI is planning for the future.
Re-engagement on AI Models
ChatGPT has been hailed as the fastest growing product on the market. But Sam has said multiple times that it wasn’t really intended to be a product. They thought it was an amazing tool, but the intent was research and not the creation of a hot new product. Now many think of OpenAI as a product company due to the wild success of ChatGPT.
If ChatGPT is thought of as a product, there are at least two important factors for its success: acquisition of new users and retention of existing users. Acquisition has been no problem, to the surprise of Sam, former Y Combination President. Marketing for the product has been easy with so much word-of-mouth and organic content generated from users. Retention is easy, for now.
Retention measures how frequently a user comes back to use a service or platform. ChatGPT certainly doesn’t have a retention issue right now because so many users are trying out new things. But how often do users try something that failed previously? That’s where re-engaging users comes into play. Re-engagement is how easily a product can get a user to attempt a desired task again.
If you have a home assistant (Google Home, Siri, Alexa) you probably only use a handful of commands despite that device being capable of so many things. There aren’t any clues about what more you can do with the device, aside from the ads you sometimes get in the form of “Have you tried X?” When you try something new and the device fails to give you a result that you’re looking for, you’re unlikely to try giving it that task again.
These negative feedback loops drive customers away from trying new things and discovering new features of the product. It isn’t until a competitor appears, or a product focused on a specifically challenging task appears, that customers will attempt that task again. Until that point, the only way you can try to get the customer re-engaging with the task is through advertisement, marketing, or training.
While I agree with Sam’s answer that re-engagment isn’t a problem yet, it’s going to be a growing problem for a product like ChatGPT. Over time people will be less inclined to try out the latest model, and over time they’ll learn not to try certain tasks with the model. This will hamper the product’s ability to grow and maybe even interfere with the research that is core to OpenAI’s efforts.
OpenAI the Research Company
Research is the primary focus of OpenAI, closely followed by policy setting around AI. Sam Altman has been meeting with world leaders to discuss the changes that advanced AI will bring. Because of their significant partnership with Microsoft, OpenAI doesn’t need to worry about being commercial or focusing on ChatGPT as a product. In fact, their hit product wasn’t talked about very much in the fireside discussion.
However, their research is greatly improved by the success of their product. Reinforcement Learning from Human Feedback (RLHF) is one of the most effective ways to train an AI model. You’ve probably seen this a number of times when websites ask you to rate how well a search worked or if the data you’re seeing is what you expected. Fuelled by the world’s attention, OpenAI can use that feedback to better train future models.
Services focused on coding, like GitHub Copilot or Repl.it, may draw users away from asking ChatGPT coding questions, leading to it having a harder time improving its ability to create a model that best incorporates code. Companies focused on research will eventually compete not just for raw attention but specialized attention on specific tasks, and that’s where re-engagement will begin to become an issue.
Where ChatGPT will Grow
ChatGPT isn’t going anywhere, and it’s unlikely to be unseated with the entire world marketing its incredible capabilities which continue to grow with each new release. Plugins, released recently, enables some incredible new capabilities as well as some scary security concerns. OpenAI is owning the discussion on the future of AI and are taking advantage of their success to be the first to lobby government entities on what the future should hold for AI.
However, without a product focus, ChatGPT will begin to lose its advantage of being first to customers and see other competitors enter the market and begin converting its customers over to their services. Because OpenAI is focused on research, they might be ok with ChatGPT being unseated as long as they remain involved somehow. Partnering with Microsoft and other companies to integrate their AI models into their services may be enough for them to further their research in meaningful ways.
While there’s a lot of excitement and speculation for the future of the ChatGPT product, the reality is murky and full of questions that have yet to be answered. I’m looking forward to seeing more AI events, as it was great to connect with the community here in Toronto and learn what others are thinking about when it comes to AI capabilities. If you’re interested in discussing this post or potential product AI, comment below or reach out to us via email.
You can watch the fireside discussion on YouTube but unfortunately the Q&A section was cut from the recording.
 | A guest post by
|