Wow, reading your interview session feels like a knowledge base in itself. Thanks for sharing!
Perhaps these mock interview can be a resource in itself later on?
We're looking to further optimize the prompt and definitely to cross check the facts that ChatGPT spit out (since its notorious to state false information as fact).
Yeah, these interviews can be a knowledge base. You could have links to past interviews by category for people to read through. But do give people an option if they want their interviews publicly listed or not. Great for practice and clarity of thought.
Edit: Just read my own chat again. I did the "interview" quite hurriedly. I can't believe how far we have come with these tools. I think a lay person who is not aware of this technology would not be able to discern that the interviewer is not a human and that's saying something.
Thats a great idea. I'm thinking there's a lot we can do with the idea. Especially through this layoffs season, might be a valuable tool for many.
Re your edit:
I know right! I've built a few thing with GPT-3 before but the ChatGPT API somehow gives a totally different result. Especially in regards to tone.
Slap Whisper API to this, you can practice with your voice and not only via typing.
you could link the knowledge base to ways to answer questions with context or create a fun gamified way to study for interviews, like leetcode but for general interviews
we are on the mission of AI-generated lessons for most content at revision.ai if you are interested in hearing more - you could upload that interview session as a PDF easily
Yes I do admit the interview is not perfect yet. More prompt engineering to optimize. I think with a few more optimization we can get ChatGPT to conduct a more quality interview session.
Sorry you didn't find value in the mvp yet, my teammate and I was taking ChatGPT API for a spin and share it as it. In hindsight we could've done more optimization before posting here.
If it's ok, I'll let you know after we optimize the prompt for you to try it again.
On the results, I agree. Definitely need more curation in regards to the voters for a more reliable result. So far I've validated the model kind of work and brings in some interesting insights.
Ideas I thought people would be interested in ends up having more not interested votes. Conversely ideas that I thought would be mediocre gather a lot of interest. I think there's a lot of work to polish this further.
In regards to registration, thanks for the insight. Initially I thought email is high friction, but perhaps I'm wrong. I'll try out login link and see how it compares!
If login link works, it's better for the platform too in fact, since I'll have a way to notify users e.g. send vote results, and create a way to bring people back to the site.
