1 comment

[ 2.6 ms ] story [ 22.2 ms ] thread
For last few months, I had been heads down on my own startup and totally ignored all the noise around generative AI. This weekend, I finally decided to spend time to tinker around, explore how GPT-3 works and try to get a product out.

Few days back - I had this craving to solve random math questions but never found anything. So thought could be a good idea to actually do it using GPT-3.

After 24 hours & a few stupid mistakes, learnt the following: * GPT-3 is not so good for Math (all the gurus saying "obviously, it's an LLM".. yeah, I get it, now!) * It's hard to actually parse text from GPT-3 into a JSON / structured format * DONT ask chatGPT for gpt-3 code snippets! It's not updated to the latest openai documentation :p

Yet to fully try: Fine-tuning model