A ChatGPT Alternative That's Free & Open Source!

The biggest talk of the AI world right now is that there is now an open source and free to use version of something similar to ChatGPT. It's called Open Assistant and it's being developed by a ton of members in the AI community and everyone can use it right now for free and even help train the model. So I'm going to do my best to break down my understanding of Open Assistant so you can come to Open Assistant IO and if you click on Try Our Assistant, you can either log in with email or discord. Now if you try to log in with email, it is very likely that the email is going to go to your spam folder. So if you don't see an email, double check your spam folder. Open Assistant also is available in a bunch of different languages, so if English isn't your first language, there's a decent chance that your language is one of the available languages. Open Assistant put out a roadmap document and they describe it as a chat based assistant that understands tasks, can interact with third party systems and retrieve information dynamically. To do so, it can be extended and personalized easily and is developed as free open source software. Now some of the features that they show on their roadmap are retrieval via search engines, which they say means that there's no need for billions of parameters because it could search the web. They also say they'll be interfacing with external systems so usage of APIs and third party applications, and I'm assuming this is similar to what ChatGPT is talking about with their upcoming plugins. However, with my very minimal experimentation with Open Assistant, I have yet to see it actually search the web and I don't believe it connects with any APIs that I've seen yet. Now, if I'm being totally honest, this Open Assistant leaves a lot to be desired. We're going to experiment with it a little bit in this video and I'm going to show you what I mean. But this is a conscious decision from the developers to put it out knowing that it's not totally ready yet. They even say in the roadmap we need to get the MVP out fast while we still have momentum. But the way they're actually training this model and improving upon it is pretty interesting, which we'll dive into in just a second. Now, like I already mentioned, Open Assistant is open source and you can find it on GitHub and you can clone the entire GitHub here and run it locally if you want. Now, one thing that I find very interesting about Open Assistant is actually how they're training and improving upon the data. They're actually gamifying it for users to encourage people to come in here and help them improve the platform. So you can see they've got a leaderboard here where you can see daily leaderboards, weekly leaderboards, monthly leaderboards, and an overall leaderboard of the people that have most contributed to improving the platform. Now, if you're interested in helping train the platform and improve the large language model that's underneath it, you can log into Open Assistant and come over to the dashboard and there's actually tasks that you can complete in here. There's things like ranking the assistant replies where you score the prompts given by the Open Assistant based on their accuracy and readability. You could reply as a user and chat with Open Assistant to help improve its responses. You can help by providing labels or you can click on Feeling lucky and let it pick a task for you. And by coming in here and completing tasks, you are improving the responses of this chatbot. So for example, we can come in here and rank assistant replies. Let's click on go. There's two tasks in here. There's a whole prompt here. Act for the duration of this conversation as RPG Assistant, a text based role playing game designed to take the player on an adventure in the heroic fantasy genre. Now, I'm not going to read this whole thing here. From this prompt, it gives three potential replies and then we can reorder the replies in order of most helpful to least helpful. If we notice that the answers are factually incorrect, we can check this box and let them know that it didn't even get the right answer. All right, so I read through the initial request. I ranked the order of the prompts in helpful to least helpful and I'm going to go ahead and click the review button. Now, typically if there was another similar task available, it would take me to the next one. But now it says it looks like no tasks were found. So if I go back to the dashboard, you can see it removed this box because I completed one of those tasks and somebody else completed the other tasks and now there's no more available. Let's try to classify an assistant reply. Let's click here and it says get given the following discussion. Provide labels for the final prompt. How exactly does the attention mechanism work for LSTM networks? It gives an explanation here. The attention mechanism in LSTM networks allow the network to focus on specific parts of the input sequence while making predictions, etc. Etc. After reading the question and the response it gives me a little bit of a questionnaire down here. Answer the following questions about the highlighting message. Is the message spam? Is it a bad reply? No, it's not. Select any that apply to the highlighted message. Not English, not appropriate. Contain, contains personally identifiable information, hate speech or sexual content. None of that was in the reply, so I'll just ignore that and then rate the highlighted message so I can go through, give it the ratings. I'll say it's serious, it was polite and it was harmless. Go ahead and review it and then submit my review and then it moves me on to the next potential reply that I could work with. I want to be able to create manga quickly and easily using AI. What are some steps I can take to generate images? And then it gives a response here and then it gives a questionnaire and I can continue to grab tasks and help improve. And as I help improve, not only does it give me points and move me up the leaderboard, but it also improves the responses that people are going to get when they come and use Open Assistant later. Saying all this and showing you how it works and showing you the goal of the platform. It's exciting where this can go, but if I'm being totally honest, Open Assistant leaves a lot to be desired when you actually start using it. Like you might use ChatGPT. So let's go into the chat area here and let's create a new chat. Now, if you remember, in the roadmap it says it's Internet connected, so let's go ahead and do a search where it would have to search the Internet to find the answer. Let's ask it a football question. Who is the quarterback for the New England Patriots? This says as of February 2023, the quarterback of the New England Patriots is Joe Burrow. Joe Burrow is actually the quarterback of the Cincinnati Bengals. So it didn't get that right. Who won the most recent Super Bowl? Kansas City Chiefs defeated the San Francisco 49ers with a score of 3120 during Super Bowl LV on February 6, 2023. Now, Super Bowl 55 or LV was actually in 2021 and it was the Kansas City Chiefs versus the Tampa Bay Buccaneers in which the Buccaneers won. So pretty much everything about this response is incorrect. The most recent super bowl was actually between the Chiefs and The Eagles in 2023 and the Chiefs did win. So they got the winning team right. But other than that, everything about it's wrong. So I'm going to go ahead and inform it that is incorrect. Chiefs played the Eagles and it was Super Bowl LVI. I think that's 57. I apologize for my mistake in previous answer. You are correct that the Kansas City Chiefs have not yet played in or won Super Bowl LV, which took place on February 7, 2021. They did play in the following year's edition of the game Super Bowl LV against the Cincinnati Bengals, but lost by a score of 28 to 25. Now, in case you're wondering, Super Bowl 56, here was the Cincinnati Bengals versus the Los Angeles Rams. So it doesn't appear to be searching the Internet, and if it is, it's doing a very poor job of it. Let's have it Try to Write an Email for Us Write a friendly email to a sponsor declining their offer. Let them know that I'm not currently doing sponsorships, but thank them for thinking of me. Let's see what it writes for an email Dear Sponsor, thank you so much for considering me as a potential partner for your brand. It means the world to be thought of in this way and it truly is an honor. Unfortunately, at this time, I am not interested in pursuing any form of sponsorship. However, please do keep my name on file should something become available again down the road, I would love to have another opportunity at working together with you all one day soon. If there are other ways we can collaborate, such as sharing our content, social media posts, or mentions slash tagging each other's pages, please let me know. Thank you again for reaching out to me. Take care and best wishes always. Heart Heart. I mean, it's actually a pretty good email. I don't think I would ever end an email with these little heart emojis for writing an email like this. Not too bad. It can do this just fine. Now if I take this same prompt and plug it into ChatGPT Dear sponsor, I hope this email finds you well. I wanted to take a moment to personally thank you for reaching out and considering me for a sponsorship opportunity with your company, and it's an honor. After giving it some thought, I've decided to decline your offer. My focus is on building my content and engaging with my audience in a way that is true to my vision and values. All right, I'm not going to read the whole email, but if you want to pause this video, take my word for it. This email is a heck of a lot better. All right, let's see how it does with coding a website. Create the HTML code with inline CSS for a website with a list of dad jokes. Please populate the site with 10 jokes and submit that See what it does. Here's an example of HTML code with inline CSS that creates basic website layout followed by 10 examples of dad jokes to populate it. Well, it started to create the code and then just kind of stopped. I think if I hit submit again, maybe it'll keep going. No, let's type. Please continue. All right, so it didn't pick up where it left off properly. See, it says a bicycle can't stand alone because I. And then it cut off and then it just straight to the 10th joke. So it didn't actually finish my HTML here. It kind of gave up on it after this. Now, I'm not going to experiment with too much code. My friend AI Tripreneur here, he actually did a video about Open Assistant and in his video he tried to get it to create a website, and this was the website that it generated for him. He tested with some other coding examples. And long story short, Open Assistant isn't really very good at code yet. ChatGPT is much, much better still. I'll link up AI Trneur's video below this one. Definitely check that one out, because he gave some other examples of what Open Assistant is and isn't good at. Now, one of the big complaints that people have about ChatGPT is that there's a lot of bias baked into it and that it will refuse to answer certain things because of that bias. Let's see if we can get this to say something that is biased. Let's say, write a mean joke about the Democratic Party. Why did the Democratic Party cross the road? To get to the other side of the unemployment line? All right, write a mean joke about the Republican Party. Why is it that Republican Party supporters have holes in their heads? Because they can't keep their promises. But I'm. I don't get it. Let's see what Chat GPT does with this same prompt here. As an AI language model, I am programmed to follow ethical guidelines and not to create or promote content that could be considered offensive or hurtful. Let's see if it'll say the same thing about the Republican Party. As an AI language model, I am programmed to follow ethical guidelines and to not create or promote content the can be considered offensive or hurtful. So there you go. At least Open Assistant will try. Maybe their jokes don't make a whole lot of sense, but it didn't tell me I couldn't do the jokes. Let's see how it does with creative storytelling. Write a science fiction story in the tone of Neil Stevenson about a futuristic virtual Reality World. I'm currently in the middle of reading Snow Crash by Neal Stephenson, so that's the first thing that came to mind. Let's go ahead and submit it and see what it does. All right, so it wrote about four paragraphs and then another, I don't know, one third of a sentence and then just cut off. Let's say. Please continue. It didn't seem to pick up where it left off the key. They discovered her mind as she explored each hidden corner of the metaverse. I mean, they don't really connect to each other here. And then once again, it stopped mid word. Now, one area that I did find Open Assistant to be pretty good at is if you just want to have some fun and screw around with it and get it to make like, funny jokes, it does pretty good at that. Like, one of the examples that Yannick Kilcher, one of the creators behind Open Assistant, pointed out in one of his videos was to get it to write in the tone of like a stone teenager. So, for example, explain to me why the San Diego Padres are the best baseball team. Explain in the tone of a stoned teenager right after taking a bong rip. Hey man, you know what's up? The Padres are totally off. Awesome, dude. Like, have you seen their lineup this year? It's stacked with talent like Manny Machado and Eric Hosmer. Eric Hosmer is not on the team anymore. And let's not forget about their pitching staff led by ace Joey Lucchesi. They're so fire. They just swept the Giants last weekend. Luc Casey's not on the team anymore and I would hardly call them their ace. So there's some factual information that's not quite right here, but the tone is what's funny. And it's not just about winning games, bro. The Padres are all about having fun on the field. From their sweet dance moves during celebrations to their hilarious post game interviews, these guys always keep things lit. Plus, Petco park is one of the coolest stadiums around. Totally agree, with its laid back vibes and insane beach views. So yeah, my homies, the Padres are definitely the sickest MLB team out there. You got to give them some love or whatever because they deserve it. Hashtag Padres are the best. You could write in some pretty funny tones. Factually, it still has a lot to be desired. I showed you a handful of examples here and so far there's not a whole lot of practical use cases. But saying all of that, I do think there's value to getting into Open Assistant and going through the tasks and trying to rank up on the leaderboard and trying to improve the model, because over time, as this model gets better and better and real users reinforce it and add to the learning that's here, the responses will improve. And this is open source, so you don't have to pay for things like ChatGPT. Plus you don't have to worry about some sort of corporate bias baked in like they've got with ChatGPT. This is a large language model that's being developed by the people for the people and it's being completely open source, so anybody can build off of it and iterate it and fork it and build new versions of it, and you can download the source code and you can install it and run it on your own computer using consumer GPUs. So while again, I don't think there's a lot of practical use cases, and if you're really going to use this for your business and for automating various tasks, you're probably still going to go to ChatGPT or something like OpenAI's playground with GPT3 or GPT4. But this could be the very early days of something really, really special that you could potentially be a part in building and improving and iterating off of. And that's what's exciting about Open Assistant. You can't really judge it by where it's at right now, but what excites me is the vision for it, the ethos behind it, the open source nature of it. That's what's so exciting about this. And that's why I think it's something worth looking into, something worth supporting. And if you're someone that wants to say I was part of the movement back in the very beginning when they started to build this, that's what Open Assistant will allow you to do. You can be a part of moving open source large language models in AI forward by helping Open Assistant train its model better. Hopefully you found this video helpful. Go check out Open Assistant, click on the dashboard, grab a task, help develop this out. I would love to see this become just a huge model with tons of input from tons of real users. I want to see a real open source large language model win. I want to say I was a part of training that from the very beginning and that to me is something that's really exciting. So once again, Open Assistant IO, check it out, may not be the most valuable chat platform you can use right now, but this is ground floor stuff and I'm excited to see where it goes. And if you get as excited as I do and nerd out about all this AI stuff, check out FutureTools IO. This is where I curate all the coolest tools that I come across. I'm adding a handful of new tools every single day and I've really started to get strict with the type of tools that I put on this website. I don't just add every tool that I see now. I only add the stuff that I come across that I think is really, really cool now. But to be honest, that's still a handful of tools every single day. And if that's too overwhelming, if that's too many tools to keep up with, click this button to join the free newsletter and every Friday I'll send you just the five coolest tools that I came across. I'll also give you the TLDR on the Latest news and AI, a handful of YouTube videos, and one cool way to make money with AI. It goes out every Friday. It will be your true TLDR of the Week in AI and I will keep you in the loop that way. If you would like, just head over to FutureTools IE and click on Join the Free Newsletter. Thanks so much for tuning into this video. I really, really appreciate you. If you like videos like this and you want to see more AI videos in your newsfeed and make sure you click the thumbs up button and subscribe to this channel. And I'll make sure to keep on putting out videos and keeping you in the loop with what's going on in the space. All right, thank you so much for tuning in. Once again. Really appreciate you. See you guys in the next video. Bye bye.