[video] Fine-tuning GPT-3 for generating questions

Here’s a quick video about the fine-tuning work I’ve already done. I just got back on the horse to continue my finetuning work on my Core Objective Functions. But for now, here is a question generator.

Why would you need a question generator?

Good question! The short answer is lots of reasons. The first reason is that I needed to generate questions was for NLCA. You see, my Natural Language Cognitive Architecture relies very heavily on asking questions. It uses questions to find answers from its internal database, which it then uses to generate better output. However, there are plenty of other things you could use this for - one example could be a question-asking chatbot. Just asking good follow-up questions would make a good chatbot.

The data is ready to go

Just grab the questions.jsonl file and you’re good to go!

Also, I am starting up a mailing list to keep people updated on my research progress. If you are interested in joining, please just PM me your email address. As soon as I finish the Core Objective Functions fine-tuning experiments, I will be sending out a newsletter!

7 Likes

That is really cool and helpful thanks for sharing it

1 Like

Absolute legend, Dave!

1 Like

Can you please tell how did you manage to crate ./questions.jsonl file ?

There’s detailed documentation provided by OpenAI here that is easy to follow in order to create the JSONL properly!

Let us know if you have any trouble along the way so we can help out :smiley:

2 Likes

I used a Python script, you can find it in the github repo!

1 Like