Haha i also thought about that Y Combinator video. Yep, their prediction didn't age well and it's becoming clear that openAI is actually a direct competitor to most of the startups that are using their api. Most "chat your own data" startups will be killed by this move.
No different than Apple, then. A lot of value is provided to customers by providing these features through a stable organization not likely to shutter within 6 months, like these startup "ChatGPT Wrappers". I hope that they are able to make a respectable sum and pivot.
I think almost each startup is focusing on enterprise as it sounds lucrative but selling to an enterprise might qualitatively offset its benefits in some way (very painful).
Personally I love what Evenup Law is doing. Basically find a segment of the market that runs like small businesses and that has a lot of repetitive tasks they have to do themselves and go to them. Though I can't really think of other segments like this :)
If your entire startup was just providing a UI on top of the ChatGPT API, it probably wasn't that valuable to begin with and shutting it down won't be a meaningful loss to the industry overall.
There's a typical presumed business intuition that any large company will confer business to a host of "satellite companies" who offer some offshoot of the product's value proposition but catered to a niche sector. Most of these are however just "OpenAI API + a prefix prompt + user interface + marketing". The issue is (which has been brought up since the release of the GPT-3 API 3 years ago) that no startup can offer much more value than the API alone offers, and thus it's easier, comparatively, than in analogous cases of this type of startup model with other larger companies in the past, for OpenAI to capitalize on this business
This has been the weirdest part of the current wave of AI hype, the idea that you can build some kind of real business on top of somebody else's tech which is doing 99.9% of the work. There are hard limits on how much value you can add.
If you want to build something uniquely useful, you probably have to do your own training at least.
Any startup that is using ChatGPT under the hood is just doing market research for OpenAI for free. The same happened when people started experimented with GPT3 for code completion, right before being replaced by Copilot.
If you want to build an AI start-up and need a LLM, you must use Llama or another model than you can control and host yourself, anything else is basically suicide.
>Any startup that is using ChatGPT under the hood is just doing market research for OpenAI for free
It's not free if you have paying clients.
> If you want to build an AI start-up and need a LLM, you must use Llama or another model than you can control and host yourself, anything else is basically suicide.
You're still doing market research for OpenAI. Just because you aren't using their model doesn't mean they can't copy your UX. Prompts are not viable trade secrets after all.
No early stage start-up has revenues covering their expenses. But in fact you're right, it's not even “free”, it's “investor-subsidized” market research.
> You're still doing market research for OpenAI. Just because you aren't using their model doesn't mean they can't copy your UX. Prompts are not viable trade secrets after all.
Prompt aren't viable trade secret, but fine-tuning datasets and strategies, customer data[1], customer habits, user feedback, etc. are. And if you're using OpenAI, you're giving all that to them. Also, given their positioning, they cannot address any use-case that involve deploying your model inside your customer's infrastructure, so this kind of market research is completely irrelevant for them.
[1]: And don't get fooled by wordings saying that they don't train on customer data, they are still collecting much more info that what you'd like them to. For instance, even just knowing the context size that users like to work with in different scenario is a really interesting data for them, and you can be sure that they collect it and adapt to.
TLDR: This might have just killed a LOT of startups