GPT-4o: The Good, Bad & Ugly for AI Agency Owners

hey guys Le Motley here and today I've got a video breaking down the GPT 40 release uh the reason for the spring update from open AI there's some good things and there's some bad things for the AI agency space I wanted to jump on here and give my thoughts now that I've had a few days to Stew it and really think about what the implications are for us so GPT 40 the good and the bad for AI agencies of course what's new if you've been living under a rock we've had the open AI spring update recently and they released a bunch of things but mainly uh the new flagship model GPT 40 um o meaning Omni and this is a step towards much more natural human and computer interaction according to open aai on their blog post so that's taking in inputs of text audio image and video and it's able to output and text audio and image so massive new capabilities that we've kind of been cobbling together using different services and we've had whisper and we've had these other services that allow us to get this kind of functionality anyway but they've just pulled it all into one and allowed the model to understand not only text input but also audio and video and stream it all into one input as well so very exciting stuff um I was hoping for GPD 4.5 or gpg 5 um but as we'll see later in this video there's some ramifications of what this incremental Improvement might mean for this bace um but I'll save that for a little later so there's also other stuff of the new chat GPT desktop app which is cool I've just been able to get it downloaded I think I need to update with Mac as many of you guys will as well I think it's only available for Mac uh for the time being um and GPT 40 is now available to free users um not just plus subscribers and therefore the GPT store is open to over a 100 million users so if you guys are interested in uh jumping on gpts I know I've made a video before that I think a lot of you found my Channel Through the GPT store is now available to 100 million users or more um which is great great news for you if you're hoping to be a GPT developer um we have yet to see anything about the monetization of gpts so I'd be curious to see what they have in store for that cuz that's kind of they attracted us all they honey potted us into this gpts uh and building gpts on the store thinking it's going to be an app store moment saying yeah monetization this monetization that um and we haven't seen seen anything of at least I have haven't seen anything about about the gbt store monetization so other stuff I don't think it's that relevant for us as as AI agency owners okay I want to start off with the good points first then we'll get on to the bad a little later um here's Sam Alman counting up all the all the money he's making from these recent updates um firstly new modalities it is a good advancement we are getting a ton of new solution opportunities opening up for us as we're able to take in different types of inputs from our end users and then give them different types of outputs back and really there's there's it's not a massive Leap Forward cuz we've been able to do this one of the examples on screen that openai has provided shows them asking a question and then providing an audio file as part of the input and then it's able to answer questions and reason off the back of that as well but prior to that it's not we were able to do that with transcription anyway we just transcribe it and then put it into the give it to the model to to reason over so not a massive leap we've been able to do a lot of these things but really what it is is just a simplification of our workflow and of the systems we need to build for our clients um less fiddling around with multiple different apis which is easier to get the results that we want for our clients and I think this is great for many of you who are not so technically inclined and I know a lot of you have been brought into this opportunity and still struggle with some of the technical parts of it but it's a clear Trend that we're seeing towards this simplification but there's still a level of complexity of how can this actually be implemented into the business so you're getting easier to do um but you still you're getting more power essentially in your hands that you can provide to your clients as well and because we're going to be using fewer apis this is probably going to decrease our cost as well because we're not having to use transcription and then generate an answer and then use text to speech if we're using these kind of systems which we'll get on to next which I think is pretty exciting The Voice AI systems and providers I think are going to win big here um because once audio inputs and outputs become available via the GPT 40 API the response times can be reduced by up to 60% um based off the numbers that open AI is provided which is between 200 and 300 milliseconds for responses at least that's what we saw of the chat GPT in the demo and as you can see here on platforms like vapy um even on the fastest and lowest intelligence model uh we've got a 650 millisecond response time and this is purely because they're having to stack up so many models that when your voice comes in over the phone they have to transcribe that then they have to generate an answer and text and then they need to turn that text into speech and then they need to send it off to you as well so this 650 millisecond latency which was fine it was fast enough we're now going to get a potentially 60% reduction of that as well so I think soon as these guys are able to access vapy and and Bland Etc are able to access GPT 40 via API and send and receive audio inputs and outputs um we could see a a continued boom of the voice AI space which is something I've been talking about a lot on the channel here if you guys are just starting with your AI agency you know looking for a good place to start or specialize in then voice AI is a great place to look into next we have a quick win for us as AI agency owners which is GPT 40 apis being twice as fast and 50% cheaper than GPT 4 Turbo it's always great on these big updates from open AI cuz we can kind of expect these reductions um and it's good to see that they're continuing to do this over time so we can expect it in future and an interesting thing to point out is that we're getting much closer to this GPT 3.5 Turbo cost which is is basically free this thing is so cheap it's it barely cost you a dime to do anything um but here we can see that we've got input of $5 and 50 for gbt 3.5 turbo so it's just a 10x price difference considering the the massive increase in intelligence and and modalities that we're going to get from GPT 40 you can't not be happy with that outcome next we have another quick win for us as AI agency owners which might have slipped under the radar for you a little bit better language support for GPT 40 that can handle over 50 different languages now covering 97% of the spoken world and it's also going to decrease a token usage as you can see here the new token compression method is actually reducing the amount of tokens for some of these languages as you can see now this may not seem like a major but this is a question I get all the time my accelerator and on my free community q&as which is should I be selling local or should I be trying to sell in the US or should I be trying to sell in Europe it's mainly people interested in selling in the US um and my answer is always no ideally Go Local um if you're if you're from South America and you're trying to go over to the United States and start selling there you're at a natural disadvantage just by purely being outside of the country you might sound a little different over the phone um you might have a name that doesn't necessarily ring like you're you could be someone's neighbor um and there's nothing wrong with that but it's just the the cold heart facts of it's going to be harder for you you're playing at some kind of disadvantage or debuff versus someone who is is John Smith who lives next door you know if you're in for example the Spanish speaking world I'm sure you've already had Fairly good responses and and good translation capabilities from GPT uh but it's really the smaller areas and these smaller languages that up until now haven't really had the support now you can be the first person into those markets so if you live somewhere that you thought oh no one's ever going to be able use this in my language or I shouldn't bother selling local now is your chance to be the first guy or the first girl in that market to go and start selling these Solutions and you might say that oh but they don't they're not interested in AI don't try to sell it as AI then just sell it as a meaningful difference in their business and now getting into the bad and we may have the rise of e girlfriends sooner or later um but that's not what I want to go into here um it's actually the long road to integration that's the first thing that I'm kind of concerned about here um and and by that what I mean is these new modalities and text and audio and video and image and all this stuff is cool but it's it it doesn't mean anything to us as AI agency owners until we're able to get that to our in customer with all these platforms that we use like make.com and voiceflow um and sending things to WhatsApp and the different solutions we build they are lagging far behind the technology that open AI is providing it really is an issue of trying to get the stuff in the hands and making it useful for our end users um but until these platforms catch up and and allow support for the customers concerned voice notes and they can send photos and and say voice flow allows you to send photos through your web chat widget which I'm not sure why and more so for things like WhatsApp deployments for your your AI agents um being able to send voice notes to the customers and receive it from them and send pictures and get them back um that's I think a long way off and I'm looking forward to seeing how they allow us to build these different modalities into our systems that we sell even for my own platform agentive we now have this question of do we want to integrate audio and video and image and all these different things into our application and into our platform or do we want to just stick with text base and I think this is a conversation many of these platforms are going to be having um it's interesting to see how they play out and moving on from that to something closely related is the lagging consumer Behavior now we can have technology that moves ahead very fast and and early adopters kind of catch up if you've seen my uh my technology adoption curve video which I'll put up here somewhere while technology can race ahead the actual tastes and preferences and and behaviors of the the consumer populace take a lot longer to adjust and e-commerce is an example of this where it took a long time for people to become comfortable with putting the credit card online and now we do it like the thought of putting a credit card and giving it to some some random website was ludicrous back in if you go back far enough it was a completely silly idea and over time it took like decades for them to get to the point where it's okay yeah now now we all buy stuff online this is the same sort of thing with with AI and I think we're going to run into this GPT and and chat GPT may help with people getting used to speaking to these AI assistants and having conversations um but I think there's still a considerable lag in the actual consumer behaviors where if we're trying to sell these Solutions do our end customers actually want to be sending voice notes to WhatsApp do they want to be sending pictures and and giving videos to them and personally if historical preceden are anything to go off I'm not betting on this thing moving too fast next we have more of a technical one that I think could be an issue which is the image and video difficulties that come along with Building Systems around much more complex and varied inputs like text image and audio um in this example here you might have watched my prompt engineering video where I I highlighted the difference between conversational prompting and single shot prompting I'll put the video up there if you haven't watched it highly recommend it's very important for you to know how to do prompt engineering and it's not your regular video it's very very different we had some really good feedback on that there conversational prompting and then there single shot prompting for us as AI agencies in many cases we're working in this single shot range where we need to engineer The Prompt and engineer the system to be reliable and predictable and continue to give the same outputs over and over and over again um so that they can actually be built into a company and operate as a a artificial intelligence task that plugged into their systems and doing that with text only proves to be difficult enough as I'm sure some of you have found out but now we introduce a whole another layer of complexity of images and videos so imagine this example here of of an email classification system where there's a user and they fill out a contact form we get an email then we use prompt the GPT task on something like make.com to Output a label now imagine this instead of it being an email input it was a video from a customer or it was a was a photo of a of a damaged good that they got from from an e-commerce store there's so much more complexity in that than okay I can understand text but this is okay what's wrong with with the product from my own experience with vision models they are nowhere near good enough to be able to be baked into these kind of systems and be able to reliably perform so that's that's another concern that I have um for this new multimodal and Omni Channel or Omni model future that they're going for and finally what might seem very concerning to some of you is this plateau and intelligence um here's the model evaluation from open ai's blog post on the GPT 4 release and you can see the pink bars here a GPT 40 and it's about the it's about the same and it's not a massive Improvement you know it's it's very incremental look we're just edging up above gp4 turbo uh in most of these categories and yes this is a text evaluation I get that and they're really focused on being audio and video and things like this to give them credit if you go through these other tabs on the website I will leave a link to the blog down below um you can see that in uh in these audio and video different tests they are outperforming the the prim models and we are seeing much larger increases in capability and and intelligence but when it comes to the scores on the text evaluation which is what we typically determine as the the level of intelligence for getting questions right and reasoning things like this which is really crucial to the systems that we currently build we're plateauing um and we're seeing incremental improvements and this opens up to a much broader discussion about the state of generative Ai and and where the future's going to go and here's a look at the multilingual performance so you can see GPT 40 is better with different languages sure but again it looks fairly incremental it's not like this massive Leap Forward in intelligence um this idea that generative AI may be plateauing or we're reaching some kind of upper upper limit has been to some degree corroborated or or confirmed by re research paper and while this may look crazy on screen I I'll explain it basically um they have found that as they continue to increase the size and the amount of data that they train models on they seeing a a diminishing return um which may sound awful but give me a second to explain so I've actually a great video breaking this down is from I think they're called computer file I'll leave a link down below and as you can see on screen here there's three scenarios of generative Ai and and artificial intelligence and with these language models and the transform architecture they have three different scenarios say the one is the most exciting where as we continue to say on these on these axises sorry we have the number of examples in the training data set and then we have the performance or the Intelligence on the y- axis and in the first case if we continue to increase the number of examples in the training data set we get this exponential curve as you can see here of it's just runaway general intelligence and things get so much smarter and we put in a little bit more data and it gets way smarter then there's a more balanced or conservative one which is a a a linear relationship where more examples more data equals better performance and and greater intelligence but what they are starting to see as you can see from from this graph here is a potential uh worst case maybe not worst case but not the greatest case outcome um based off the evidence of this paper where we are seeing a flattening curve and as we continue to increase the examples it's flattening off and we're not getting any kind of meaningful increase on intelligence this is a little summary from that research paper which is there is a clear log linear trend for the models to get incrementally better at handling a concept that concept needs to appear exponentially more times in the training data this shows the models are very data inefficient which may sound concerning to you as someone who's just bit their whole life on this AI space and continued advancements um and I guess the question is has generative AI peaked um and I think this is an important one to ask but personally I'm betting on this being a temporary Plateau where the architecture and the Transformer that we currently use is kind of maxing out and and we've pushed as far as we can we've got great capabilities we have an awesome set of new technology to use and Implement into businesses um and we've just attracted the entire every smart brain on the planet has has flooded into the space as as you are here watching this video interested in the space and we have the greatest researchers in Minds all going into it humans are not done with artificial intelligence by no by any means we're not just going to go oh well well we've got the missing returns right I'm out like okay I guess we can't get any smarter than that so that's complete rubbish so we're going to continue to keep pushing and just like the transformer architecture blasted open the scene for us to be able to get to this point um when GPT 3 released we're going to continue to see people innovate and try to create smarter architectures that are more efficient on training with that data so we could be looking at a a little Plateau a nice little bit of room to catch our breath and and I honestly think that that is going to be very very good for the space as the AAA and the AI automation agency space as a whole because if if the the tech keep kept taking off and it just got better and better and better and every 6 months we had whole bunch of new stuff to try and adapt to with we don't ever get a chance to really catch our breath and solidify our solution so this is something that I've I've been kind of excited for to some degree where I get questions on the channel which is oh but like what if what if the like why why do they need our Solutions why would they ever need to use us if if the thing just going to get so good that suddenly they can type one word and then it just automates their whole business that that is a a very low percentage outcome like in terms of probability and if this is anything to go by we're not going to get there anytime soon so we have a chance now to say okay yes people have dropped out of the space they've gone off they've chased other other shiny objects we now have a chance to really work on our craft really work on identifying these use cases and going into businesses and finding ways that AI can help them and we get a nice Runway now before things take off again to dial in our services to get experience and to continue to push ourselves further and further away from The Spectators who sit there and watch AI news to the people who are actually taking action like you and I I hope they give you a little bit more clarity about what this update means for us as AI agency owners I do not like being an AI News Channel but when there's big updat from open AI um it is worthwhile coming in here and hopefully giving you guys a bit of insight into what this means for us if you have enjoyed the video hit down below and leave a like um subscribe to the channel if you're not already uh for more content like this teaching you how to make money and build businesses in the AI space if you're interested in seeing what my life day to-day is like here in Dubai as an AI agency own it you can check out my recent video here um showing you the RO reality of my life as an AI entrepreneur but aside from that guys thank you so much for watching and I'll see you in the next one

Channel: Liam Ottley

Views: 12,758

Keywords: chatgpt, chat gpt, chatgpt examples, chatgpt explained, generative ai, openai, chatbot tutorial, artificial intelligence, AI, chatgpt prompt, liam ottley, make money with ai, ai business ideas, make money online 2024, make money online, how to start an airbnb business, ai automation agency, ai automation, ai agency, ai business, start an ai automation agency, start an ai marketing agency, ai business opportunities, business analyst, ai news, ai automation business

Id: MvNcm6aOKd0

Length: 16min 23sec (983 seconds)

Published: Thu May 23 2024