Hey! Don’t say that. I’m thinking a few months to maybe 1-2 years don’t say 10-20. What model you used for your ai
10-20. 30, since you don’t have much resources.
TinyLlama/TinyLlama-1.1B-Chat-v1.0
Noo you know I’m not trying to initially make large chatgpt like ai right
2 Likes
Depends on what exactly you mean by ‘ChatGPT-like’. Because ChatGPT is very complex. If you mean 'an AI that can predict the next token in a series, then yeah, probably doable in a few years (depending on how complex your dataset is, how well you can train it and how strong your PC is). But if you want something remotely close to even GPT-2 (which is not good, but still), 10 years is a good estimate.
Though, 10 in binary or 10 in decimal, that depends.
1 Like
I meant chatgpt like as in large scale and complex kind of and what I thought was more focused on one thing so probably doable in a few months to 1-2 years but I probably won’t work on it all the time so there’s that
1 Like
Ah, then yeah, brings it down to 10 years.
Unless you scrape the data from elsewhere. There are plenty of data objects on Hugging Face you can use.
1 Like
No what I’m try for I doubt is much on the website you mentioned but I did think of the data for training too for that one
what about your ai did it take 10 years no right
1 Like
I downloaded an existing model (gemma4) from the internet. I didn’t design the LLM itself, I just designed the framework around it. I gave it the system prompt, handed it the tools, and gave it a small database to work with.
Oh wait, is that what you were referring to?
1 Like
fine fine no I didn’t mean that I meant to make a model but in a different way like I don’t exactly know the traditional way so you’ll make fun of me but yeah leave it I’m perhaps rambling now
1 Like
No, I’m not making fun of you. I’m impossibly blunt and that’s important because you need someone to reality-check you when everyone’s telling you “you’ll do great”. You need someone to tell you “you didn’t do well and will likely fail”.
In this situation, I think the task you’re thinking of undertaking is just massive, though still doable because people in fact do do it. In fact, nowadays I think, with rising prices of API keys, people are looking to move towards SLMs (small language models), so you actually have a lot of scope.
The only issue is even “small” language models can consist of hundreds of thousands of parametres, which might not be feasible. Might, because you can still embark on it. People do build their own PCs from scratch, so you can train your own model.
In fact, if you do, I’ll do the same.
1 Like
Aare na, I didn’t say you’re making fun of me I said you’ll make fun of me, probably, and nobody said in that way that I’ll do great on it (maybe because I didn’t tell much people) and what I’m thinking of is different like it will take a lot of time probably but minimum time not 10 years but yeah anyways people generally don’t understand at first what I’m trying to say I don’t know if I talk different because sometimes chatgpt misunderstands even
1 Like
Ah.
Don’t worry, I won’t make fun of you. I’ll make fun of your current lack of skills. When you gain the skills, I’ll stop making fun of them.
Yeah, probably not ten years, but then again it depends on how much you want to go into it. I want to pursue a masters in ML, so maybe I’ll get into something similar too.