KHAI KWAN LAWYERS

I have create a simple cross-exam module for applicants looking for a job.

At the same time, I also tested the same using bare LLM like Grok and it came back with more or less the same output and process.

I leave this to you to decide whether it is good or bad. What I mean is that as model become smarter, it is able to understand what we want to do even thought our prompting is rubbish. In my case I provided a very simple prompt for a very complicated backend process of “cross-exam” and somehow even I cant understand how Grok was able to think it is a barrister and do this without any coaching….

Check it out https://superbar.pythonanywhere.com/

Previously I was working on improving the skills of juniors to be top presecutors but I reckon there is more demand from witnesses who have to appear in Court wanting to improve their answers/replies. So I design a chatbot for specific facts to do this. Best of all it can use different languages. This is the video below giving you some explaination and the link is inside should you want to test this. Remember the chatbot is designed for a unique facts which is mentioned in the video. One need t program these facts else it wont know what to ask right….

This youtube is 14 mins long with the first 7 mins about set-up and the last actual examination and cross. Obviously one can reverse the role and have the chatbot be the lawyer doing the examination and cross with the user as the witness.

The only problem is that if you want to try this out you will need to register and get your own account as the platform resets each time. Again this is just a trial to show proof of concept. https://character.ai/chat/vncC2P0zLiDrUx6c24fmcI2bWY0e6HEJmbj0uR9P7gc

Looks like the Botsonic Platform is not working anymore. It has this 14 days trial and after that nothing works unless I go back and reset everything again which is a pain. I am leaving it here as a record (below). I have used the same documents for RAG at helix.ml you can go to https://app.tryhelix.ai/session/ses_01j5m8ajt3y71btnfsp3hx6ac6

The main reason is to show it is very easy to do this now, the only issue is $$ and privacy, you dont want the LLM to get your data especially if these are sensitives one. So it is better to have your own servers to run this, helix is one of those platforms. But you have to know a bit of coding though.

Here is the link https://www.linkedin.com/services/page/681343314281303345/

Here is a long video 15 mins about a company that provides data training services for LLM. They provided a page where they differentiate their product with the base model and another trained by Watson. I am currently working on fine-tuning a model so that it can give advice on Affidavits. Still WIP but will be upload to huggingface soon. Anyone interested in co-peration can contact me but seriously I need workers not talkers. Also if you have the chance try out Unsloth, appears to be “easiest” & fast way to fine-tune.

Hello been a busy week experimenting with the various platforms. IF you are up to it check out my amendments to AS 4902 – Building Contract on behalf of the owner.