|
|
|
|
|
by weitendorf
766 days ago
|
|
I want to push back on the idea that a “a skin over chatgpt” has no value a little, because it’s a common sentiment, but the tide is shifting amongst a lot of people I’m in contact with and what I’m building indeed makes API calls to OpenAI/Anthropic. In my opinion foundational models are essentially becoming commodity products with only temporary functional differences (eg voice) and ever-changing relative quality and aptitude for certain problems. Yes, OpenAI/Google/Anthropic/Microsoft can launch their own apps built on top of foundational models, but so can almost anybody else, and presumably there will be (or already is) a plateau in quality improvements so that open models may catch up and drastically reduce costs for those calling LLMs in their products. In my opinion reducing applications that call LLMs to skins over <insert model name> is like calling web applications skins over JavaScript, or a database, or a Java backend. Yes the technology has value in itself and they’ve commercialized it with conversational interfaces, but there are many more business applications or technical products that could use LLMs (on top of other things) and don’t exist yet. And surely many attempts will fail, but I think it’s going to become the primary way that LLMs become commercialized as time goes on. |
|
With that said, I think the main problem faced by these new businesses is not making people use the apps, but the cost of running the models themselves
Right now they're essentially being subsidised by Microsoft (in the case of ChatGPT at least). I think it's inevitable that there'll be a price hike and that it'll destroy the value proposition of these skins
ChatGPT is currently $20/month for a private user and from what I can gather from different news sources, OpenAI is still taking a loss on those sales. Most "skins" will probably have to charge at least the same for their service and I an app honestly has to be pretty amazing for me to fork over that amount of money every month
These are of course just back-of-the-envelope guestimates and I could be wrong about a lot of the assumptions I make here. Given the consequences it would have for the people running all these new businesses, I certainly hope I'm wrong