First I want to preface this by saying that I see value in large language models and transformer technology, I've personally gotten value out of ChatGPT using it as an assistant in product and software development.<p>However I can't help shake the feeling that we're being sold a bit of snake oil when it comes to the APIs and business use cases. In the demos I've watched in events from OpenAI, Cohere, Scale etc I haven't seen any major compelling business use cases. Even today's GPT-4 demo, where they showed it understanding tax code and answering tax questions, what is the use case? If I'm developing tax software do I bake this into my product to answer questions? Why can't someone just use ChatGPT or any other bot built on top of it to accomplish the same thing? Why are they using my bot that uses the same API as everyone else?<p>Has anyone found and solved really compelling business use cases with LLMs?
> Why are they using my bot that uses the same API as everyone else?<p>Maybe different developers will use different APIs from different providers running different models?<p>Even better, from what I've seen and experienced, LLMs can perform very differently based on how they're prompted and/or fine-tuned, so Alice's TaxGPT may be better than Bob's SmartTurboTax, etc.<p>To the larger point, these could help make software a whole lot smarter and more tuned to personal preference. That, I think, is a selling point.<p>As a developer, consider a basic OCR library used to power an app that helps your user track their spending by reading receipts and automatically inputting amounts in some database. An LLM like GPT-4 can do much more than read what's on it trivially.<p>I can imagine an on-device model that has access to your health data can trivially read nutrition information on a label and suggest that maybe you shouldn't have that donut today...<p>I don't know, but easily building smarter software sounds like a strong business case...
You have to move paste the copy and paste and think about applying the API against groups of files and databases.<p>simple script to add comments to all your code files.<p>Text-SQL interface.<p>building a chatbot.
1. generate 50 ways to ask to take a flight.
2. now build a set of rules or reg expressions that match on those questions<p>scrape HN and categorize each of the titles and for each ML link go out and create a summary of each of the articles.<p>how long would it take you to do any of these with the api.<p>how about it the api didn't exist?
Not an answer but I do think this is a game changer?<p><a href="https://techcrunch.com/2023/03/01/addressing-criticism-openai-will-no-longer-use-customer-data-to-train-its-models-by-default/" rel="nofollow">https://techcrunch.com/2023/03/01/addressing-criticism-opena...</a>