Fine-tune GPT 3.5 or build an assistant?

Question

I am currently developing a platform to help MBA candidates write unique MBA essays. I applied to HBS and INSEAD in the past, and when I used ChatGPT to assist me with the essays, the answers were really generic and the text used very complex words.I have access to a database of essays from successful candidates and different universities, and I want to use this data to increase the quality of the outputs from GPT. I am just trying to decide if it's better to (1) leverage OpenAI assistants, feed the documents as knowledge and optimize prompts for each essay or (2) fine-tune a model such as GPT 3.5.Any suggestions/recommendations?

hackerlight · Accepted Answer

Use GPT-4, it's better than 3.5.If you use a free tool, use Gemini Pro (which is the free model on bard.google.com), it's also better than GPT-3.5 (but not better than GPT-4).The output doesn't have to be verbose, you can instruct it to be concise.There's a GPT in the GPT store which translates verbose GPT output to human-like output. There's probably other custom GPTs that can help too.You can make your own custom GPT with the essays as context, upload them as .txt files.