David Shapiro
Patreon: https://www.patreon.com/daveshap (Discord via Patreon)
Substack: https://daveshap.substack.com/ (Free Mailing List)
LinkedIn: https://www.linkedin.com/in/dave-shap-automator/
GitHub: https://github.com/daveshap
Similar Posts
23 thoughts on “Train GPT-3 on Any Corpus of Data with ChatGPT and Knowledge Graphs – SCOTUS Opinions Part 2”
Comments are closed.
keep in mind there is an attention span break point for HAL (chatgpt)… another guy got it to play chess by letter-numbers but after so many moves HAL started making up moves lol
Don't you have to add import en_core_web_sm to your step02_dedupe_characters.py to use it..
That is my AI coding workflow exactly. Ask for the code, put it in, put the code with the error back in, voila!
Also, I have tried out many OCR solutions. G-Cloud document AI is the most expensive and the absolute best.
It can intake any PDF, and can even do photocopies or handwritten notes
What more, is that it can deal with double columns. Other OCRs get everything jumbled up
"English, your ability to describe what you want is going to be the primary programming language from now on. Period. End of story." Love it <3
Hi David, perhaps you’ll be bringing this up in subsequent videos related to this topic, but is the plan to ultimately add this set of law knowledge to a fine tuned model? I thought I read you can add knowledge in this way by having empty prompts and just provide the responses.
Is this true?
Are there best practices?
How can we add meaningful knowledge to GPT-3?
This part was so great "I'm going to address the elephant in the room and that was it like a week ago I made a video saying like meh chat GPT isn't that great um and uh so obviously like I kind of had my foot in my mouth because this is amazing … it also looks like it failed" 😀… glad you came around!
I just use the native speak-to-text on mac. Works like a charm right into tje input fields. Windows, although inferior 😅, must have something similar?
Also – great content 🎉
Dude you’re the dev whisperer. That’s great for devs what about us non-devs who want to learn this whole new world and now think we can but just need knowledge graph? Make us a knowledge graph to learn how to use Ai expertly. We got the English part down
Thank you for sharing! To streamline the deduplication process, you can manually input the problematic text into GPT-3 and ask it to correct any spelling errors that may cause duplicate entries. 🙂
My first reaction to the 'poor OCR' result was that the script was appending characters to the same file instead of replacing it because the script ran multiple times, errors or not. I'm not a python expert, is that even possible?
WHen you get errors , you dont need to do anything except copy and paste the entire error and hit enter. It will spit out fixes. You don't need to explain anything really.
It's the birth of the computer from Star Trek….stunning
To do the de-dup, you could try:
"write a python program that takes a list of words and maps each word to the closest English word using the smallest string-edit distance to the English word. Use the Spacy library to get the list of English words."
I really like how you think through problems. It's very inspiring.
Nice video. Thanks for sharing, David.
chatgpt is becoming annoying with its time limits. Every now and then it stops working saying "too many requests in 1 hour". If it didn't make mistakes in ffmpeg .bat files, it wouldn't take that many requests to fix it.
I believe human mind is still superior.
The chatgpt is for the lazy p-fools.
You are doing very nice work with these AI vids–have been going through a few of them. Have you tried VS Code and Jupyter plugin for your interactive coding sessions? It's not a coding panacea but Jupyter greatly accelerates your "code -> test -> refine loop" because your session remains live until you restart. In one page, you can test individual blocks of code (not just line at time like console), import from existing modules just like regular python code, and get graphical and formatted output. I start all of my exploratory work there then migrate stable code back to regular python modules. Open source and works on all platforms.
Do you remember when lexus nexus database was released? It had the same impact as chatgpt now.
I wish someone would teach GPT unreal engine blueprints proper
Great video! Usually, for me, I am much productively and fast using my own daily common wrapper (higher level) libraraies for the things like you are spending so much time to "teach" GPT3/4 to how to write to correct code. However, the part of extracting legal document(s) into JSON-LD, GPT is doing great job for that. But, for the coding part,, I usually can simply just few lines of calls to my dailly document processing libraries from PDF (or any format) to Text to whatever. Maybe, in the future, GPT can catch up my high level libraries function.