Fine tuning LLMs (GPT4) without training with your own data is now possible

OpenAI released fine-tunning but the model GPT-3.5-turbo has less reasoning capabilities when compared to GPT-4

I will explain how you can load related part of your own data to ChatGPT before it answers you.

I LOVE UGLY DIAGRAMS here is the process:

Step 1 and 2

Head over to https://ln.keymate.ai/web and login with your email, then upload your PDF (Guide: https://help.keymate.ai/en/articles/8273516-upload-pdf-content-to-your-personal-knowledge-base#h_cc57199f3b )

Step 3 - 6

After successful pdf load enable plugins on your ChatGPT plus account at https://chat.openai.com , install KeyMate.AI Search on plugin store, log in with same email you used on step 1 and 2.

Prompt this

"Query personal knowledge base about [PDF TOPIC] and answer [Complicated question to LLM about the same topic]"

ChatGPT reads up to 5000 words from your personal knowledge base related to your query.

Enjoy you just post-trained your GPT-4 LLM. It's not as powerful as fine tunning but you will have references to the facts uploaded by you and GPT-4 is way more clever than GPT-3.5-turbo.

Some use cases

After Sales Support

Upload your product user manual and use that to answer about product support questions (i.e. We are going to use this at?KeyMate.AI to delegate our support process so that we can reduce our support ticket response times)

Book Related Tasks

Upload your favourite algorithms book and ask LLM about specific algorithm/data structure for a specific problem.

DOTA 2 / LOL Coaching Heroes Guide

Create an updated/meta knowledge base about heroes and interactions then ask specific questions to get pick and item advices.

Bigger Knowledge Base For Companies is possible

If you are a company that have a scenario in your mind. Please reach out to [email protected] we can provide bulk PDF uploads, secure stored PDF knowledge base and linking, referencing to specific page when an answer is given features.