Recently I got access to GPT-3 and have been playing with the tool. Things I have tried:
- Generating tl;dr for paragraphs
- Extracting information from news headlines (e.g. "Edtech startup Microverse raises $12.5M to bring income share agreements to the developing world" => Startup = Microverse, Funding amount = $12.5M, Category = Edtech)
I have seen a lot of emails and copywriting tools made with GPT-3. What have you made or are working on?
I was building an app using GPT-3, but eventually I abandoned this idea.
It was a copywriting tool that generated product descriptions based on given parameters (in Polish language.). The app worked really well. Together with my teammate, we showed PoC to some people, and they were truly impressed.
In the meantime however (when app was about 80% ready) I dived more into safety guidelines section and found out that our app doesn’t fit OpenAI requirements (yea, I know - I should have checked it earlier…).
Anyway - output length in our app was about 4x longer than required by OpenAI. When we adjusted the app to safety guidelines, output descriptions became very short and useless.
We tried to pivot this idea, but the feeling of such significant dependency from OpenAI was frustrating. So we basically stopped to develop this project.
I’m pretty sure I will use GPT-3 in one of my future projects. I know however it will be responsible only for part of app functionality to avoid full dependency.
GPT-3 is an amazing tool, but in many cases safety guidelines simply don’t allow you to use its incredible power. It’s totally understandable - not everyone will use GPT-3 only in good faith. It leaves this strange feeling of limitation though.
Thank you for your insights. I was also reading through their guidelines and its worth going through it once if you want to release a serious product.
What a shame! But many companies are using GPT-3 to make copy writing tools and they seem to be operating. What's different in your product?
Also I see companies/products like this one - https://shortlyai.com/, who seem to also need to have longer outputs than copy writing. I wonder how they are handling this issue? Any insight?
Hi, sorry for late feedback.
As far as I know, after some time of using GPT-3 in an app, it’s possible to ask OpenAI to have expanded access & give them more detailed usage feedback/data in return.
In such way some tools are treated individually and can enable their users to generate much longer text forms.
How did you get a key for the API? I am trying to get one for over a year now and I have not succeeded. Any pointers would be welcome :)
From my experience: you need to show in your application form that you are engaged in a real project. If you fill the form with purpose to use GPT “for fun” you can be 100% sure that your application will not by accepted. Few points worth attention:
It worked well for for me - hope it’s useful.
I'm working on getting access 😏
any tips how to get one? I would love to try this out
I've been building Sidekick AI with in-house models when I got accepted, so recently I've incorporated it into my system to work alongside my other models (www.sidekickai.co).
This sounds like a great use case of GPT-3, will keep an eye on this product.