Should I Be Here?

Beginners
1

I am a new user with a above average IQ, some linux experience, limited programming experience, and limited compiling experience, and no machine learning experience.

I want to learn about machine language models to help me with writing. I know ChatGPT is an easy option, but I don’t like that ChatGPT filters out objectionable or controversial content.

Where do I start? Is this a good website for me? It’s popped up in my searches several times.

I would also like to learn how to use machine learning to do various things I do at work. Some of it is boring, but also specialized, despite not being that high skill. I don’t even know how to train anything yet.

I also don’t know anything about machine learning. I can’t ask my work to let me spend some of my time on it with no understanding at all. I don’t have time to get a machine learning degree. Can I self teach myself?

I have written programs before.

3

I am coming into this with zero knowledge of anything.

What should I start with? Is there stuff here to learn?

Colab was a pain in the ass when I tried it before and it kept thinking I was a bot when I ran tasks for a long time, then kicked me off colab and blocked my google account as fraudulent, which was devastating because I used my real name, Frederiqua Psuedolicious, and I may be blocked from google forever as a result.

Is paying for GPU access expensive? Should I just buy my own GPUs?

Is this something I can do casually while working or will this take up all my time?

I also don’t even understand how training works. If I train a LLM, does the training become something I own apart from the open source files? Can I train a LLM for specific things? What is the best way to quickly learn this?

4

Also, when you say “programmatically curating data” I have no concept of what that means.

5

I don’t know much but I at least have gpt4all and have messed with other repos that use its python module, so I don’t know much, but…

I bought a 900 refurb PC that can work as a chatbot. Main thing I went for is Intel i7 or higher, Nvidia RTX works well with Intel (and I game some so it has good pixel quality) and at least 16 GB of RAM, but honestly I got 32 and I’m glad I did, like way glad. Also, SSD tossing 500 MB/s around during indexing / answering.

So there’s the hardware way, still slower than ChatGPT tho, by like a minute. Also, dumber, unless you smarten it with ingested data, but then that increases processing time if you make it really precise. Haven’t gotten GPU working yet for custom models in gpt4all yet, though. That likely would improve performance from 30-50% from stuff I’ve read, depending on settings.

Don’t know about cloud rental, that’s obviously a service and I just imagine like the learning curve for that as there are for the hardware way.

Also, ML is statistical based and I have the advantage (or inefficiency) of playing with some honestly hard statistics-based programs before now, so I don’t expect interpretating the lingo will be super hard. Not sure if I would curate data, though, maybe mess with transformers more, if anything.

What you should do depends on what you’re naturally good at, or willing and allowed to improve, hope this fills some gaps kinda spotty like.

6

(post deleted by author)

7

totally get how overwhelming it can feel to dive into ML, especially with hardware limitations and all the terminology floating around. Here are some tips that might help clear things up:

  1. Where to start: Since you’re coming in with zero knowledge, a beginner-friendly course would be ideal, especially one that explains concepts in simple terms. Right now, 365 Data Science is offering open access to all their courses for 21 days. They cover ML basics and have interactive exercises to make learning more hands-on and straightforward. This could give you a foundation without immediately needing complex setups.

  2. Cloud vs. Local GPUs: Cloud GPU access can get pricey if you’re running large tasks, but it’s often more affordable than buying and maintaining a GPU-heavy setup unless you’re doing this full-time. Since you’re just starting, experimenting with smaller models and datasets can be effective even with your current hardware. Once you’re comfortable, you might explore cloud options like AWS or Google Cloud, which offer free-tier options and discounts for beginners.

  3. Training and Ownership: Training a model means fine-tuning it with new data to adapt its responses. Once you’ve trained it, you own that trained version (the “weights”) and can use it independently from the original open-source model. You can specialize the model to certain topics, but keep in mind, this requires a solid understanding of model architecture and training processes.

  4. “Programmatically curating data”: This usually means selecting and organizing data in a way that’s useful for training models, often with code to automate it. It could involve filtering, labeling, or arranging data sets to ensure the model learns the right patterns. With your stats background, you might find data prep interesting and it can improve model performance significantly.

  5. Casual vs. Full-time: ML can be casual, especially at the beginner stage. Once you start experimenting with custom data or training, it might take a bit more time and focus, but it doesn’t have to consume your life. Building up gradually with smaller projects could be a great way to explore it without getting too bogged down

Wish you luck man