← Marketing Against The Grain
Loading summary
Transcript3 lines
- [00:01]
A
I'm going to go through one of the most surprising yet powerful AI launches in the last 6 to 12 months. And why it's so surprising is it's not come from any of the big players, it's not come from a Google, it's not come from OpenAI, it's not come from Antropic. I'm going to give you some cool ways you can play with this new AI app. I'm going to give you a prompt that any marketer can use to analyze a campaign and get some real world tactics to make that campaign better. All of that and more. I'm Kieran Flanagan, co host of Marketing Against It Green.
- [00:34]
B
Hey guys, real quick. You know we love building custom GPTs on the show and we love sharing it with all of you. Well, we wanted to kick that up a notch. We just developed this free guide that teaches you how to build your own custom GPT on chatgpt. We've taken the guesswork out of it. We've got templates, we've got a step by step guide to design and implement custom models. So you can focus on the part that's actually fun, the part we love actually building it. And if you want it, you can grab a link in the description description below and go check it out now. Now back to today's show.
- [01:17]
A
First of all, the launch. The launch is from a Chinese company and the model is called Deepseek R1. Now what's interesting about this model is it's completely open source. They have open sourced the model, it's free to use. They've given all of their papers away around how they built the model, all of the information in terms of the weights they're using in the model, they give it all away for free for other people to be able to build on. And what's super interesting about that is it's comparable in terms of power to ChatGPT and OpenAI's O1 recent O1 launches in terms of intelligence and reasoning. And it's completely free. Now OpenAI's original mission was to give everyone AI tools for free. They wanted to open source it. They believed everyone should have access to this intelligence and reason and power. Along the way. They decided to change the mission because they said, wow, this is way too powerful. We don't want to open source it. That's a real threat to the planet. Now the skeptic could say, well they did that because they realized, wow, there's a lot of money we can make here. And so they brought it all in house internally and they do not open source that anymore. And so this Chinese company are kind of continuing on the original mission of OpenAI which is interesting because China is not known for its love of open source and they are giving all of this stuff away for free. And it's a really important launch because now if you're kind of on the fence about AI and you don't want to spend the money on any of the O models, you don't want to spend $200 a month for the O1 Pro. You have an open source model that you can start to play around with and it gives you results and gives you intelligence and give you reasoning on a parameter with one's model. So let's kind of jump into a couple of the tweets and then I'm going to show you a prompt that I ran with deepseek that I think is really impressive. It's going to analyze a marketing campaign and give me real world tactics to actually make that campaigns better and basically help me strategize how I do these campaigns. And actually you can take the prompt, start using it today, because I think it's a pretty great one. But I want to get into a little bit of the coverage around Deep Seat to give you what is going on in the world and what are people saying about it. You know, one of the most important things here is just how cheap it is, right? Deepseek R1 is 100% open source. It's 96.4% cheaper than OpenAI's O1 model. So it really does put pressure on these companies that are trying to differentiate through their models. I think open source, if we take deep seek at face value, it's like six months away from what the kind of large closed source companies are doing. So open source is trailing around like six months behind these companies, which is not very far if you just want to wait it out and get these things for free. OpenAI's 01 is $60 per 1 million tokens. Deepseek R1 $2.19 per 1 million tokens. And what's incredible to just think about for a moment is this is intelligence, right? This is the cost of intelligence and reasoning going down. Over time we're going to get to a point in civilization where intelligence and reasoning that is above what a human can do is nearly given away for pennies. And I think that's an incredibly mind bending thing to think about in terms of where will the planet be when everyone has access to these tools and everyone kind of does, because this is Totally free. The other thing I wanted to mention is it also has access to the web and so we are going to get into a prompt and I'm going to show you what I did. But you can run something like this, which is who is the strongest hero in Marvel movies. I love comic book movies. So I'm actually interested in the answer here. The first thing you'll see is it's doing, reasoning, it's thinking. And when you go through this, this is fascinating and I'm going to show you what it did for the marketing campaign. It's doing like an internal monologue, thinking to itself, reasoning to itself, thinking through the problem. It actually has one of the most natural language internal monologues of all of the models that I've seen. Like it really is like a person trying to think through a problem and it's a fascinating thing to read through. If you go watch this video and start playing around with Deep seq, go try it out, you can start to read through its internal monologue to see how it is actually thinking about solving a problem. That in and of itself is quite valuable to see how an AI model is actually approaching any problem you may have. But you can see it has access to the web very similar to OpenAI's models. It excites its resources. For free. For free. For free. For free. And then the other thing that I have been thinking a lot when I read through the Deep SEQ paper. So the cool thing is if you're a AI nerd, you can actually read through how they built this model and it's worth doing it because it is quite different from what the other companies have done. And so I'm not going to get into the technicalities of that. I would read that on your own because I will probably mess a bunch of it up. But it is fascinating. There's some papers out there, they've open sourced it all. But what I really thought about was, wow, like you can really start to build your own models because you have the foundational model like a deep seq, which is comparable to 01 for free. And now you can actually fine tune that model for your needs. For example, these models, 01 deep seq, they are really good at logical and reasoning kind of problems like math, coding, science. The only model that I think is really good for creative tasks like Rhinon or I think is the best, I won't say the only model, they're all kind of good, but Claude is like much better for creative tasks. And I think that is one of the ways they differentiate. Even the way you kind of interact with Claude, it just feels like a different type of interaction. It more sucks you in to like be your kind of pseudo friend, right? Like it feels like an AI friend and their creative tasks are really good. But there is an opportunity to start to build some of these custom models yourself because I can take Deep SEQ and now I can fine tune it for specific creative tasks that might be applicable to the audience that I'm trying to build AI tooling for. So I think there's going to be a large market for actually taking these markets and building these different types of models that are very specific for a certain set of tasks and a certain audience. And I think that is going to be a large thing we're going to see come out of open source. And I've always been a believer in open source AI, that this is going to be one of the core ways that companies actually use AI within their businesses because they don't want to be tied to any one model because different models are good for different things. You can see in my previous videos, I'm using Gemini, I'm using Claude, I'm using OpenAI, now I'll be using Deep Seq. I don't think companies want to be tied into a singular model and so I think the way they'll navigate that is they'll be able to have open source models fine tune for certain tasks and have an orchestration layer that will call the different models when they need it. And I believe in that so much that I've invested in some companies doing that. So here's a deep seat, go sign up, it's totally free and I'll show you what I have been doing with it. So I give it this prompt. So it basically says I'm a market analyst for a SaaS company. We've given it an audience, which are doctors. We've given it, we want to increase revenue for the practices and we've compiled all of this data about our recent B2B campaigns. So you can see here I have basically just put it into an image. You can give it a file, you can give it a Google sheet, but I basically give it the campaign details. I give it the start date, I give it the end date, I give it the budget, I give it the ROI type, target audience, channel conversion rate, revenue. And so what basically I asked it to do was I give it some information about the quote, key features of our product. I told it about the primary audience and then I asked it to analyze the performance of each campaign considering the product details and target audience. Identify patterns or trends provide actionable recommendations to optimize future B2B campaigns. The other thing I asked it, which I think is really clever on my part because these things connect to the web and you should think about this is also look for external trends I should be mindful of around targeting doctors. You don't have to speak perfect English. You can see I messed this up. But it understood what I meant and could have impacted these results. Like large events, busy times of the year and so on. Right? When I'm running campaigns I want to know is there anything going on externally that actually could impact the quality of my results? And so it thinks for 30 seconds this cool internal monologue where it starts to break down the problem and says all of the things it wants to do. So it's really worth reading through this because it's fun to see how it kind of problem solves. And then it comes through the campaign performance gets it all right. Does a really good job of analyzing that graphic and basically talks about what's doing well, what's not doing well. Talks about patterns. The patterns did a really good job of. Right? It says group practices respond well to webinars. Hospital admins prefer in person events they trust Buildin for system integration Solo practitioners low engagement with blogs need concise actionable content. External trends it has some really good things here. Q1 lead generation aligns with clinics and annual planning cycles. So I should bear that in mind. June content market and potentially impacted by summer vacation reducing engagement so my campaigns around content campaigns were in June May industry conference coincides with medical conference seats and I would have never known that. Right? I now know that the medical conference in season is like May. It gives me some healthcare trends. Growing demand for telemedicine Economic pressures may delay budget regulatory changes. There's a whole slew of words. People laugh at me. There's another one that I can't say anyway words. Sometimes I'm not great with them which is why I love AI because it doesn't judge me and then provides actionable recommendations. Right? Expand interactive content formats because it's figured out that webinars work really well. Live demos work really well. Revamp the content because they have not done so well. Tells me how I can improve my LinkedIn campaigns. Talks about how I should allocate my budget time in and then the external trends. Any risk mitigation and then I basically asked it to build me the webinar schedule and then it's like built me an entire webinar schedule that I can just hand to my boss and say, hey, I've done that for you. And now I'm going to go and chill because I've done everything you've asked in like two prompts. So it's given me like the exact, you know, webinar title, target audience, key value proposition, description, suggested timing, because it's looked at the external trends and the internal trends and figured out like, when is the best time to do all these things? Because it's using reason and intelligence to do that. This is done with two prompts, right? I analyzed my campaigns. I got an incredible webinar calendar for the coming year that's mapped to the internal learnings and external trends in two prompts from a free model that is equivalent to the most powerful model that you can pay for today, which is O1. And actually it has performed better than O1 across certain coding benchmarks, MAT's benchmarks, and better than Cloud Sonnet 3.5. This is a wild time to be alive. These models being open sourced is just going to mean that people are going to start to fine tune these things and build upon them really rapidly. This year we are going to see an incredible speed pace evolution of models. So I would strap in for an incredible year. If you like these videos, I do them in my newsletter. Do some exclusive for my newsletter that you will not get on LinkedIn and you want to sign up to the podcast marketing against the Green because we go deeper in some of these insights. I hope you're finding them useful.