Discover more from Generating Conversation
Rumors of startups’ deaths have been greatly exaggerated
Two quick announcements:
We have some exciting news about a new product release coming the week after Thanksgiving. If you’d like early access, reach out!
With the holidays & travel coming up, you’ll see fewer posts from us for the rest of the year. We’ll be back regularly in January!
After OpenAI’s Dev Day announcements last week, there was a slew of takes about how OpenAI had killed off a number of startups in a morning. These were fairly reactionary takes, and we disagree strongly. OpenAI certainly has Sherlocked some startups that are building basic wrappers around GPT, but we would argue that OpenAI is actually creating an ecosystem for applications to flourish. This who we think should be worried and excited.
Who should be worried
We’ve all seen examples of applications that are thin wrappers around GPT, especially those rely on prompting to get the model to behave in a certain fashion. The new GPTs feature, while it might be limited in scope, makes it easy for users to create their own customized personas, and given the ease of use, there will be a large number of GPT-wrapper equivalent tools in the GPTs ecosystem. This will perhaps lead to a long-term flourishing of an ecosystem of GPTs, but it’s not a promising sign for GPT wrapper products.
OpenAI’s support for retrieval has also led to skepticism about the future of vector databases and companies relying solely on retrieval augmentation. While we’re confident OpenAI will improve over time, the current state of the feature isn’t incredibly impressive. Storage is incredibly expensive, and the quality of retrieval is not particularly impressive. Startups doing basic retrieval or building vector databases will absolutely have to differentiate themselves beyond what OpenAI offers, but they’re not dead (yet).
Who should be excited
Hint: It’s mostly everyone!
For everyone who’s not in one of the above two categories, we would argue that the Dev Day announcements have pushed OpenAI towards being an ecosystem for applications to build on. Most tactically, the features that our team at RunLLM are the most excited about are JSON mode across GPT-3.5 and GPT-4 and price cuts on fine-tuned GPT-3.5 models. (OpenAI’s price cuts have been impressive across the board.)
These improvements unlock more use cases, more depth of model usage within existing applications, and better user experience. Realistically, OpenAI knows it’s not going to build every application on its own. The Apple/App Store comparisons have been flying around on X this past week, but we believe OpenAI it’s setting itself up to be the AWS of the 2020s. It will be the first account every makes, and it has the advantage of pricing based on usage. (We’ll sidestep the App Store revenue share controversies here.)
It’s become obvious to us that one LLM won’t rule them all. We fully believe that there will be an emergence of application-specific models, and at today’s prices, OpenAI is undoubtedly the place to start.
This moves OpenAI squarely into the category of core infrastructure. AI-powered applications will layer on top of OpenAI (especially as privacy concerns lessen), and this will create a value-accretive flywheel. OpenAI will gain more insight into how users are customizing models, allowing them to provide better capabilities to their enterprise customers. Meanwhile, companies building with AI won’t have to deal with tedious & costly GPU deployments and scaling challenges.
We believe OpenAI is poised to create the next public cloud provider category. Of course, they’ll compete with startups building equivalents to their core technology, but they’re clearly working to simultaneously create an ecosystem of products building on their infrastructure.
That means we’re poised for an exciting, if tumultuous, 2024. Stay tuned!