(untitled)

trying to build my own chatbot a lot like ChatGPT3 anyone know any good resources for teaching myself to do something like this? just trying to make my own chatbot thanks!

>>34255 Good luck making anything close to the level of "intelligence" that ChatGPT3 provides. If you're interested in neural networks, the book [url=https://nnfs.io/]"Neural Networks from Scratch in Python"[/url] is great, and it has a free accompanying course [url=https://www.youtube.com/watch?v=Wo5dMEP_BbI&list=PLQVvvaa0QuDcjD5BAw2DxE6OF2tius3V3]here[/url]. If you just want to make a chatbot, learn python and have it extract key words and phrases from your sentence, and combine this with rudimentary sentiment analysis and pre-programmed phrases to get something semi-realistic (at least for the first 5 minutes).

>>34257 what do you mean by this exactly? Thanks for the info, you think maybe I could just custom tailor something using the ChatGPT3 API?

>>34258 If you're using the ChatGPT3 AI's API, that's not really you who's made the chatbot, you'd only have made a frontend. The first line (including nnfs.io) was me linking resources for if you're interested in learning about neural networks in general. The second line was suggesting ideas for how you could create something semi-convincing in python. What I mean overall is that you probably can't achieve something on the level of ChatGPT3, and depending on if you're actually interested in neural networks like ChatGPT3 uses, or if you only want to make a "chatbot" and don't care about complicated neural networks, I've given ideas either way.

>>34259 well thanks for the ideas. What if it's something worth looking into using a neural net to make something for some very use specific task rather than generalized? Like, if I wanted to make something but only specialized around a certain area or aspect like, gardening or something or being able to answer questions about a certain type of plant (like, take a picture of a plant and then it can tell you some relevant information about the plant and then answer questions about the plant as well)

>>34260 A strong suit of it is categorization, so it wouldn't be too hard to get it to tell you which plant it is out of a set few options based on a low-resolution image of it. In fact, naming clothing items is an example project in nnfs.io In terms of answering questions about that plant, where the AI would come into play is not creating knowledge about the plant but detecting what information your question is requesting, so this might be limited to a set few questions it can answer about a plant and it uses the AI to detect which question you want answering, even if you phrase it differently. Another place where AI excels is in the "creativity" department. If you have a dataset of plant pictures and their names (several thousand of them), it wouldn't be unreasonably hard (once you know how to make neural networks, that is) to make an AI that names plants based on an image or generates an image based on a name, which you can make up. Good luck!

>>34260 You probably need to learn to program first. It seems like if you're coming here asking such a general question your skill level to get started is probably not there. Python is basically "the" language used for AI stuff these days i.e. PyTorch / TensorFlow etc.. Not to mention it will take lots of data and compute time to actually build a decent model. GPT-NeoX is basically the closest thing to an "Open Source" GPT3 ... out side of the Megatron-LM. Megatron isn't realistic for most people to use in general, so I discourage you using that even if you have a budget of several million dollars.

>>34255 You heard about llama? Zucc was working on a language model to compete with closedai's model, and an employee leaked the entire thing. You can run it locally, and it doesn't require a GPU or even good specs. Been playing with it a bit. It's nowhere near as good as GPT-3 but it's decent enough, and like I said, it runs locally. If you're interested, a couple gh repos to search for are llama-dl and llama-cpp.

Saw an article not too long ago about using chat GPT to train facebook's model into a near chat GPT level model that you can run locally. https://newatlas.com/technology/stanford-alpaca-cheap-gpt/