Incredible New AI Use Cases Available Right Now

Welcome to this week's episode of AI news.   You can use. We have a new best-in-class AI   music generator A little AI app for all you  chess fans out there and this has been the   week for open source AI models So much released  this week So with that being said, let's have a   look at all the new AI releases that you could be  putting to work today Okay, first things first the

music generator. Yurio This is a new player that  entered the game and it's essentially a slightly   better version of Suno Some people claim it's  twice as good. There's a discussion to be had   there. But one thing is clear It's way better at  certain instruments and it's way better at clear   audio. There's a lot to talk about So I created a  dedicated video talk about this tool and all other   AI Music tools that you might want to consider  if you want to just play around with it You can   go to Yurio. com sign up for free and as of today  April 10th 2024 you can generate something like   a thousand something tracks a month, which is  ridiculous It's really impressive and absolutely   amazing. This tool absolutely just raised the bar  on what AI music is today So, there you go, did   you hear that it sounds really good many  examples on here You can just go into the home   page check out some of these. Here's a dune themed  one Listen Okay, let's get serious, all right next

up quickly I wanted to cover this just because it  became so popular on github But there's this brand   new AI software builder called planned X that  has been getting more and more attention. This   is super fresh I have not played with this yet But  essentially what it does is that it's one of these   many copies of Devon If you remember a month ago  Devon the AI assistant that builds apps for you   Well, this doesn't include vision yet all of it  just works inside of the terminal but I wanted   to bring this up because there's a whole open  source movement forming around Projects like this   where people are trying to build a eyes that build  software and they do what they can right? There's   a lot of limitation context windows reasoning  abilities. These models have not been created   to run autonomous agents perfectly Nevertheless  people build things like this and they open source   it So if you're looking for AI coding engine  for complex tasks that works in your terminal,

there you go Moving on to some fun updates here  So let's talk about a new feature that Poe added   and I want to keep this brief But they added a  thing where you can essentially monetize your   chat bots for every single chat a person takes  with them And I wanted to feature this because   I feel like this might be outlining the future of  many of these products now Obviously the GPT store   came out that got a lot of people excited and now  nobody really cares about it, right? Like what was   the last time you browse the GPT store to find  something interesting? I'd have to think for a   second All I use is my custom built GPT's that do  very specific things that are only relevant to me   and most people I talk to Find the same thing So  right now might not be the moment where stores   like this really shine and there's all these  useful chat bots But I find it really interesting   this approach to monetizing it where if people  find your bot useful They're gonna be paying per   request What an amazing model and I reckon as they  evolve over time with things like planned X being   available through a GPT soon-ish, right? It's just  a question of time Yeah I can see a lot of people   paying per request to build complex software with  AI Easily for something like chat GPT, by the way,   if you didn't know open AI started talking to some  GPT creators I saw this on X recently. They're   basically trying to pilot something similar to  what Poe is doing today here But this is largely   unexplored territory Like how do you monetize chat  bots that people could build themselves and they   probably can be prompt injected so you can extract  all the info That the maker put into it. I don't   know thought this was an interesting development  paying per message something We might see more of

in the future Okay Next of all I'll be showing you  is a no code chat bot builder called chat link and   they're actually the sponsor of today's video But  bear with me because they built a super simplistic   way of building chat bots yourself rather than  me Just telling you let me show you as you log   in you can easily create a chat bot like so you  can either start from Scratch or use one of the   presets. I'll just use the basic AI chat bot It's  that easy When you head on over to the builder,   you'll see a basic flow So as you can see here  It starts with a message and then it presents the   user with a form where it captures the email first  We need to head on over to the knowledge base to   add some data sources You can upload documents add  sitemaps and much more So what I'll simply do here   I'll add my own little PDF with a guide on how to  use a specific keyword in my journey and there go   now the bot Has custom data and now I get the  choice when the user talks to the chat bot Do I   want it to use the knowledge base or an AI model?   I can switch over to AI model here GPT-4 turbo   and as you might imagine we could add all of these  blocks here and make this more Complex and repeat   the same thing over and over again But let's  just preview the bot right here welcome message   will fill in our email and now it opens up the  conversation Hi Igor, how can I assist you today?    And it uses AI to answer here What can you do for  me? And as you can see it successfully answers   with GPT-4 Now there's so much more in here  between the building blocks and integrations with   Zapier You can add knowledge bases and once done  You can easily publish this and add this to your   website If you want to start building AI powered  bots yourself easily check out the link at the   top of the description and now let's look At some  more AI news you can use okay But now let's talk

about all the open source language models that  came out over the course of the last week There's   so many and all of them are really good. This has  been a massive week for open source So what's new   first of all this command R model by Cohere then  this Mistral model 8x22b and then stability AI   came out with lm212b. What makes this unique?   Why should you care first things first Cohere   came out with R plus and already ranks at the top  of chatbot arena in terms Of open source models   as you can see This is the number one model here  for open source as rated by the people Benchmarks   align super positive sentiment all across the  internet towards this people are loving this   brand new model Why might you ask? Well, first of  all, it has a large context window of 128 K tokens   Secondly, it's optimized for retrieval augmented  generation aka rack Which means it's optimized for   you to attach your own context your own files to  it I think it really good for business use cases   Also, it is trained to be multilingual so you  can work with different markets all in the same   model and that I mentioned This thing is fully  open source So this is a really exciting one for   businesses and enterprises because you can run  this locally It has a large context window You   can add a lot of your very own context via rag and  it's multilingual and it outperforms all the other   models Whoo, kind of a big deal every week. We  keep receiving better and better models and this   is no exception But that's not where this week  ends beyond that Mistral released another model.    This one is called mixed rolled 8x22b It's also  fully open source And if we shift the conversation   to strictly benchmarks the new Mistral model  is actually slightly better than the Cohere   R plus model here Now mind you these just came out  and as with all of these language models It takes   some time for users to figure out if they're  any good in practice But from my very limited   testing and what I found in my research across the  internet Mistral which is built on the mixture of   experts is the one you want to go to for general  purpose usage and Cohere R plus as it is optimized   for rag is the one that you want to be using for  business use cases where you're gonna be attaching   a Lot of extra files knowledge  and there's one more model that just came out   which is stable LM 212 be the main selling point  here is this one is quite small and multilingual   So you got English Spanish German Italian French  Portuguese and Dutch? Yeah, let's do it Let's test   this with a German prompt. Perfect. Schleiben and  also to the penguin it Yes, we need this. There   you go If you didn't know these models that are  natively trained on multilingual data perform way   better in other languages than something like GPT  That can't translate it to other languages because   it has seen them but let me tell you using GPT for  in something like Slovak where there's 5 million   inhabitants in the entire country and Not a lot of  literature across the internet compared to English   It's just not the same model as it's translating  a lot of its knowledge and then a lot of the info   gets lost in translation So for international  audiences and use cases, this is preferable and   we're getting so many options these days. It's  absolutely amazing Just keep in mind that this   arena doesn't update every single day. So here and  there they updated right now We don't even have   the mixture model on here yet One more thing about  large language models this week GPT 4 actually got

an upgrade namely They updated the GPT for turbo  API their most capable API to a new version and   that version is also inside of chat GPT So to all  of you chat GPT users, it just got a little better   now How did it get better? And what are these  major improvements? Well, we don't really know   if you look into the comment section here on X all  these people share frustrations Hold up. Hold up.    Hold up editing Igor here. I'm about to release  this video and a few hours ago They actually did   release what is better about these they published  this graph with the different benchmarks and   improvements So on the left, you can see before  after essentially better on almost everything,   but here's the most important thing to me  They changed how chat GPT response now,   it's more conversational. So you get less of  these Bloated formulations like hey friends   name just checking in to see if you had a chance.   That's not how people speak Hey friends name just   a friendly reminder to RSVP for my birthday  dinner on dates Like this is how humans talk   and now chat GPT response more like this So that's  absolutely amazing One more thing that also just   happened today is that in chat put arena this  new GPT for turbo model that is also inside of   chat GPT so both chat GPT users and API users get  the benefits of this is the new king it overtook   Opus and for my first Tests I can confirm they  felt the competition and they improved several   things about it And especially this conversational  writing style is really enjoyable Go ahead and try   it out and chat GPT. It does make a difference  You can ask it to write messages and if you have   experience with the response before it you'll feel  the difference It's a bit more human now. So yeah,   the king is back. I suppose. Okay back to the  video But I did find Lewis here that ran various   needle-in-a-haystack Benchmarks on GPT for turbo  the new one versus the old one and that got way   better if you're not familiar It's a very simple  benchmark where they give it a very long piece of   text and at different points They hide a specific  piece of information and then they ask the large   language models about that information If it  quotes it correctly it found the needle in the   haystack and it scores Well, they do that over and  over again to test how good it is at retrieving   little pieces of info that are hidden somewhere  in The middle GPT for used to be okay at this   But especially Claude Opus was way better at this  now to seem to have gotten better as you can see   here comparing all the different APIs they had  the green line is the performance of the newest   one and a higher value is better So it's the best  GPT for we ever had in terms of retrieving things   from a long context Okay, next up I promise you  something for the chess fanatics in here and

I myself I'm not even a big chess player I used  to play a lot as a kid with my dad, but over the   years I became rather rusty And this is where you  can use this tool called Nocti dot AI I realized   that chess. com does something similar to this  but they have some unique features in here like   flashcards that it generates to teach You certain  moves and live feedback as you're playing and you   can also just play this rating game where you  play against AI and it Tells you what Elo you   have. Let me just quickly do that You All right,  821 on light chess, that's 1. 1 as mentioned not   great so terrible But yeah, there you go. AI  has been a thing in chess since quite a while   now But there's a slightly different approach  can practice openings just like with other apps   creates flashcards and apparently uses AI for all  of it Okay, then I have one more app and this one

is really simple, but very cool. It's completely  free as of now It's called fix blur and it does   one thing it fixes blurry images Let's just take  this blurry image right here. Let's see what this   can do fix a photo now upload an image Look, this  couldn't be any simpler, right? We're doing this   in real time It's uploading and there you go  five photos left today. Come on do a good job.    I want this tool to be really good There you go.   Wow, that's way better. What a cool little tool   and it's completely free Alright, so if you want  to learn more about AI music check out the other   video that we created other than that Don't forget  to subscribe because I do this every single Friday   leave a like and make sure to leave a comment  with your favorite Release this week. I'm always   super curious to see what parts of these videos  you like. That's all I got. See you next week