HuggingChat: This is HUGE for Open Source ChatGPT!

This is probably one of the biggest developments when it comes to a free and open source alternative to ChatGPT. The CEO of Hugging Face just announced the release of Hugging Chat, an open source early prototype interface powered by Open Assistant. Hugging Face, if you don't know, is the company that has really democratized access to AI for masses. And they are hosting a large number of open source models, as well as data sets for applications related to computer vision, natural language processing, audio processing, and a whole bunch of multimodal models. Just to be clear, it's not a new open source model, but an interface where you can interact with different open source models. Before playing around with this new chat, let's look at some of the distinguishing features. So in terms of privacy, according to them, in this V0 of hugging chat, we only store messages to display them to the user, not for any other usage, including for research or model training purposes. So they are not collecting any data, unlike ChatGPT or OpenAI. And in order to use this, currently you don't even need to have an account with Hugging Face. So in this case, you just need to be very careful with your cookies if you want to keep your browser or chat history, because according to them, if you clear your cookies or switch the browser, all your conversations are going to be lost. As I said before, they're not collecting any data, but in future they might introduce a feature in which you can let the authors of the model collect your data and use that for training purposes, but you're going to have the control on the data that you generate. They're interfacing this with Open Assistant, but in future they plan to actually integrate other open source models as well. Now here is the best part. It's running in a space, so that means that the code for the UI is open source, which means that you can actually copy this into another space and customize it. So you can even right now replace the model with some other model that you want, or you can update the user interface. You can add memory to the model if you want to. Here's the chat interface. If you want to start a new chat, you can go and click on New Chat. It looks to be inspired by ChatGPT. It has a very similar feel to it, right? And then there is this button here where you can change a theme. So let's say it's a dark theme right now and you can change it the way you want, which is pretty neat. A few things before we play around with the model. So first thing you'll notice is the model is Open assistant. It's a llama 30 bin parameter model. Right? And then there is a small disclaimer. Generated content may be inaccurate or false. Right here it says that this is the V0. So keep in mind this is a very early version and the model may not be great, but it's just the start. Now the model information is all here. So let's say you can click on this, you can go and look at the model page. Similarly, there is a link regarding the data set that is being used for training the model. Now this is great to have all the transparency regarding your model as well as the data set. And as I said, you actually have the ability to replace this model with another model. You just need to copy this UI into your own space and then you can make changes. Another thing is since Hugging Face is storing all these different open source models, so there is a possibility that you can have different apps in here. So if you recall, there was a paper called Hugging GPT for which it gets very close to the everything app vision that people describe. Right. So essentially in that paper, Microsoft researcher used model hosted on Hugging Face to perform different operations. Now if somebody wants to do it, it's actually possible in here. Okay, let's look at some examples and here they are definitely taking inspiration from ChatGPT. So let's see, we're gonna click on this one. Write an email. Okay, so as a restaurant, I'll write a professional email to the supplier to get these products every week. So wine, eggs and bread. Right. And here is the email that it wrote. I hope this email finds you well. I'm writing on behalf of our restaurant to rep to inquire about your availability for regular deliveries of certain products that we use in our kitchen. Right. And so this is the quantity. Let's see if it got it correct or not. 1024 12. That's pretty nice. The rest of the email is also pretty coherent. Right now if you look at here on the left hand side, it actually auto generated a title for your chat. That is pretty neat. It again seems to be inspired from ChatGPT. Okay, let's click on new chat. Okay, let's ask it another question. Since it's an open source model, so give me five reasons why open source large language models are better than closed source models. So let's see what it comes up with. If you look at it, actually the generation speed is pretty good. It's pretty fast. Okay, so here are the five potential advantages of open source or closed source. So community involvement and contribution, customizability Cost saving, security, freedom of choice. This seems to be very valid reasons. So open source project often large community of contributors who can provide feedback, report bugs, suggest improvements and contribute code to the project. This means that the open source project open source software is likely to be more extensively tested and may receive more frequent updates and bug fixes. So these are pretty good responses. I think I'll give it at least a B plus or a not bad at all. Now if you want to actually play around with the underlying model, this Open Assistant model, I have a video on this, so I'm going to link it here. Now again, if you look here, it created another chat for you, so that's pretty awesome. And then here is a small link, share this conversation. So let's see what this does. So it creates a link that you can share on your social media if you want. Let me copy this and test it out in another window. Okay, I opened it an incognito window. It seems to be storing that chat so you can actually share your chat with other folks. Keep in mind right now, as I said, you don't really need to make an account, so anybody can play around with it. Now let's ask it a programming question. So here is a normal prompt that I used. Write a Python function to write a file to NS3 bucket using the boto library. Let's see if it can do that. It's a pretty trivial task, so if it does search online, it should be able to put something together and it's able to generate the code. So that's pretty nice now. Well, it seems to be working pretty nice nicely. One thing I wanted to highlight is that the focus is not really on the model in this case, but rather the hugging chat interface where you will be able to integrate different models that are either on Hugging Face or potentially your own models and play around with it. It's probably one of the best open source interface that I have seen so far, which is pretty awesome. It's pretty neat, pretty simple. It mimics ChatGPT interface, so it will look very familiar. And in my opinion, Hugging Face is in a really good position to actually build some really cool apps on top of their platform because they already have access to all the open source models you can think of and it could be a really good alternative to ChatGPT or OpenAI models. So we just hit 75,000 subscribers in almost four months. If you're new to the channel, I make videos and tutorials related to large language models, image generation models and anything AI related. Check out all the other videos on the channel, and if there are any specific topics you guys want me to create videos or tutorials, let me know in the comment section below. I'll be happy to make them. Make sure you subscribe to this channel for more updates and hit that bell notification button so you don't miss any new video here. Just wanted to give you a quick update on Hugging Chat. A more detailed video containing comparison with ChatGPT is coming soon. Thanks for watching. See you in the next one.