Loading summary
Transcript19 lines
- [00:01]
A
Okay, we've just got access to the smartest AI on the planet, Grok 4, but can it actually do marketing? Now we're going to give you all of the latest on Grok 4, how it's performed better than all of the other AI models, but then we're actually going to put it through its paces and do some live work with Grok4 to show if it can actually do world class marketing growth and knowledge worker tasks. All of that on this episode of Marketing against the Grain.
- [00:33]
B
Here's a quick word from our sponsor. HubSpot helped Tumblr solve a big problem. They needed to move fast to produce trending content. But their marketing team was stuck waiting on engineers to code every single email campaign. Now they use HubSpot's customer platform to email real time trending content to millions of users and in just seconds the impact three times more engagement and double the content creation. Want to move faster like Tumblr? Visit HubSpot.com.
- [01:17]
A
Kip how excited are you to have the smartest AI model just newly subscribed today at your fingertips? And we are going to show our entire audience just how smart it is.
- [01:29]
C
We have a brand new smartest AI model in the world. It comes from our friend elon Musk. It's Grok 4 from Xai and Kieran. The early feedback on Grok 4 is that it's pretty, pretty amazing and much smarter than most models. We've been playing with Grok for a little bit. We're going to break down for everybody today. So some use cases of Grok 4 for your business.
- [01:55]
A
Okay, so why are people excited by this model? This is one of the most respected benchmarks this arc AGI how far ahead it is of everything else. Now, I have listened to three different podcasts today on Grok 4. I think there's like nuance in how all of these models benchmark themselves against benchmarks that I am not going to get into because I'll just butcher. But apparently you can for sure peer pretty well in benchmarks while still not being great at certain tasks. But there is no doubt that Grok 4 is an incredible model. If you're watching this, you're likely have watched something else previously already on Grok 4. What we're going to do in this episode is actually give you real world applications of Grok 4. The only other one I just like, did you catch the super heavy Grok? The super heavy Grok does is like really fascinating, right? It has a whole team of agents go and try to solve a task and then it basically looks to see what agent has solved that task the best and kind of gives you that answer. So it has like whole swarms of agents go and try to solve something for you. And apparently it takes a long time from what I have been told. It's like 300amonth. Yeah, it's 300amonth and it's pretty time consuming. But I think that is like a pretty interesting next way of having AI solve problems for you. Yeah.
- [03:14]
C
So, yeah, Grok, you know, there's a free version, there's $30 a month and there's $300 a month. I think I'm a $30 a month subscriber, Kieran. And you get Grok 4 and I think a little bit of Grok 4 heavy in there. But to really use it, you're looking at the $300 a month. There are a few takeaways here, Kieran, before we get into some of the use cases, for me, one, I think that's really important for everybody to understand is that AI innovation curve is still really steep. It's not like flattening out. It's not like that. The models are just getting a little bit better. They're still getting a lot better. And so you're going to continue to see this wave of innovation be very fast and very steep. And we're nowhere near what AI is going to look like in a couple of years. I think Grok 4 is a good.
- [03:57]
A
Example of that, as Elon Musk said, because he's changed his tune a little bit. He used to say, like, we want to ensure we open source AI to make it safe. And now he said, look, I don't know if this is going to be safe, but I want to be alive to see what happens on the live demo of Rock 4.
- [04:14]
C
Well, I'd also like to thank Elon Musk for ruining my wedding because I'm going to be getting married soon. And one of my other sure takeaways here is that within weeks GPT5 will come out because the OpenAI team seems very competitive to me. And they're normally, they don't like anybody to have the lead quote, unquote, for too long. And so I know I'm going to be like, about to get married and going to have to be like, jumping on some emergency pod because GPT5 comes.
- [04:41]
A
Out and Gemini have a new model coming out as well.
- [04:45]
C
It seems like it's going to be a very busy summer for the AI models and there's going to be a lot happening. And that's all cool and that's all good, but how does it actually impact the average person who isn't a complete AI dork like you and I? Kieran, in terms of, like, what are these new models for the sake of the show, Grok 4 actually enabling people to do in a new and better way than the previous generation of models.
- [05:13]
A
So I got access to Grok4 this morning. I went into my AI lab and I came up with three prompts that I think are good examples that we can use here to test it. So the first one I thought would be kind of cool is one of the advantages that all the GROK models have is they have access to xdata. And so one of the things I really wanted to try was build an ICP for your content, right? So we have this ideal customer profile for your product and service, who buys it. But I was really interested in trying to figure out if you could build an ICP for your content, who is the ideal reader of your content, who will engage with it. And so we have this prompt and basically what it will do is it will say like, you are a data strategist and you will take. We're going to do this for HubSpot. We're going to do the niche that we're in. We're going to say our primary goal is awareness. And then we're going to have all these competitor handles to actually go in and out and do analysis on. It's going to do a bunch of analysis. It's going to segment those engaged followers into clusters, and then it's going to figure out for each cluster, like, who are these folks? And then it's going to basically start to create this icp. And the reason we're going to create this ICP and the next prompt, but the one I'm really excited about doing is we're going to try to build a content creator, which is basically a model that allows you to give it content and then it will grade how well this content will theoretically do for this icp. So I think there are two good use cases for people who create content. Marketers, founders, everyone who listens to this show, everyone wants to figure out how do I perfect my content, tailor it for someone, and then actually how do I figure out if this piece of content is going to attract the attention of of those people or not? I've only done the prompts. We're running this live. So we have GRO4 picked down here. We do not have super heavy yet. We may actually get super heavy for fun. Why did I decide these two would be good ones? To illustrate here. So in the demo, which was apparently a demo free from marketers and put together by engineers, did you see the slides?
- [07:15]
C
No.
- [07:16]
A
It was like, everyone is like, this is the worst presentation of all time. Like, what's in it is really good. But the actual presentation, the kind of like energy, the slides, I think it was all a bit of a letdown. They should use genspark to create their slides. I'm not a genspark affiliate, but that's. It's a really good product.
- [07:33]
C
We do need to get Eric from genspark on the show. We got to put that on the list to do. Ginspark is great product, but one of.
- [07:39]
A
The things they used, and this is a really interesting use case, this is one that you're going to want to dig into, which is basically they figured out that they gave Grok some prompt and said, who would win the World Series and give it a bunch of tools. Oh, they saw that and they took it from polymarket and actually got the predictions right. So now, like, I'm pretty curious on how it can predict certain financial outcomes.
- [07:59]
C
We're here talking about these Grok models. They are like much more solo user than business user models. Like, they don't have the same business user controls and scalability and everything as something like Anthropic or chatgpt. But if you're a solo user, there is some compelling reasons to use Grok in that I think the quality for the price is very good. I think the $30 Grok is now pretty close to ChatGPT's $200 a month subscription.
- [08:30]
A
Yeah.
- [08:37]
B
Hey, everyone, we'll be right back to the show. But first let me tell you about a podcast that I love. It's a podcast called I Digress. It is hosted by my friend, Troy Sandich. It's brought to you by the amazing HubSpot Podcast Network, the audio destination for anyone in business. And I Digress is great. It's got 30 minute episodes and the podcast is all about helping you eliminate complexity, complications and confusions in your business. It's really heavy in frameworks and strategies to help you scale and sustain your success. And recently Troy had a really great episode. It was called Branding Secrets. Leveraging style to change the narrative, boost influence and profits. And that was with his guest toy, Sweeney. It's really good. You should check it out and you can listen to that or all the episodes of I Digress anywhere you get your podcast.
- [09:33]
A
Okay. We had to do this show in two parts because when we tried to do Grok 4 Live, just as it had come out, everyone seemed to be having problems with it, us included. The results were not very good. I was playing around with it all weekend and it seems to be back online properly giving me output. So we're going to show you some use cases. Is Grok 4 any good for the general use cases that marketers or knowledge workers will really want to use it for? We see it's great for the benchmarks, we went through that, but is it any good? Is it any good for the general user? So let's give you some use cases that are actually tailored to show the strengths of grok, which is it's an incredible reasoning model with access to X data, which makes it pretty unique from the other models. Right? It has a very unique data set. So let's go to one of our first that we're going to try to see how good it is. We're going to run this live and then I'm going to show you ones I had prepared earlier to take you through them and give you my kind of summary of where I think we are with Grok4. So here we're going to do something I think is like really cool because it has access to X data, what's on X, tons and tons of great creators and writers. And so one of the things you can do is say, well, let's try to write like one of the creators. Now I've said this for forever when I've been showing people how to do things with AI, you should never copy and paste someone else's style. But there is a copywriting method to how you can learn pretty rapidly. That's called copyworks. And copyworks is you literally take a copywriter you admire, look at their post and actually handwrite at their posts so you can memorize the things that they do really well and then you can build your own style from that. So this is very similar to that. It's taking our creator style and using that to create a piece of content that you can then build upon and make your own. So in this we're going to give it some global write in rules. Here we're going to give it a task which is to write a newsletter on the theme of Copywriting is the only timeless marketing skill that matters in the age of AI. And I think copywriting becomes ever more important because AI creates content, copywriting creates attention. And so being able to craft attention, grabbing copy is always going to be an incredible skill to have we're going to use one of our friend of the show, Sam Parr, ex founder of the hostel. The time range, we're going to say 24 hours. We're going to keep it to about 200 tweets to analyze. And so what we're going to do is try to recreate Sam's style by analyzing his tweets. And so we're going to pull all of his tweets for that time range. We're going to sort by the longest because we want to take the ones that were longer in nature because it's easier to kind of train on those. We're going to look at total engagement retweets and replies. GROK has access to retweets and replies, does not have access to likes. Then it's going to create a rule set to actually mimic his style. So keep these rules handy and then you're going to actually mimic that style. So let's actually run this. Should run pretty fast this time. Okay, so we're back. So here we are. Why copyright and trumps every other marketing skill in the AI era? Let's see if it's got a bit of salmon here. I used to think tech tools would kill off old school marketing skills. AI changes everything. Take this guy I interviewed. He built $150 million moving company from scratch. So this is actually from Sam's podcast. So it's pulled in things from his podcast, his secret killer copy and ads and emails. No fancy AI. So he's got a bunch of things here. Why does copyright and win a bunch of things here? Want proof? Look at my own stuff. I host podcasts, share business stories. The episodes that blow up. Ones with hooks and descriptions. Simple words grab attention, no fluff. It's a little bit like Sam. Knowing Sam's right and it's not exactly like him. It's analyzed 61 of his posts. For marketers, here's how to level up your copywriting today. Read old ads, study classics like Olga V, write daily email posts, whatever. Edit ruthlessly. Test ab, join groups to join. All right, so like it's fine. I think it's fine. I think I could probably continue to work on it, but I've tested GROK a lot with creative output and I think it's just fine. I think this is a good example of it being just fine. I think this prompt is pretty good. I will tell you, I did run this prompt in O3 on ChatGPT and the output was much, much better than Grok4. Okay, let's go into a couple of others and then I'm going to run one other really cool prompt to see if it's any good. So this one here basically looks to see one of the things they did in the demo video was they predicted bets. They were using Poly market data, which is a known platform that you can kind of bet on certain trends, really any trend you want. It's like not normal betting on horses or things. It's like things that will happen in the world. And there's been some good examples of people using Grok 4 to predict things and then bet on Polymarket about those things and it being correct. So I wanted to know could it predict what campaigns would do really well because again, it has access to so much data in terms of what people are retweeting, reposting, replying to. And so here we came up with a fake campaign for a startup called Brief Buddy, AI note taking app. And basically this is a AI note taking app specifically for marketers to move them away from the admin of marketing and back to doing the craft. And it's a five day public sprint in which marketers use that tool to reclaim five creative hours by offloading note capture, task tag and follow ups and first draft generation. So Testament's post a before and after love lettle carousel. And so we have a primary target audience here where we're trying to target solo content marketers, fractional CMOs and one person marketing teams. We have a little bit about their age demographics, we have a little bit about their pain points here, scattered notes, lost follow ups. We have a little bit about psychographics, right? Productivity app enthusiasts, share wins publicly motivated by time save metrics. We have an action we want them to take. We have a little bit of brand tone, we have a test budget, we have an evaluation horizon which is one week which we want to run this experiment. And then we say return eight sections in the following way. We have a Persona snapshot, we have a competitive landscape. What are the other options? Why are we different? What are the risks? How do we do one week validation, the messaging kit, the channel mix and then the self assessment. So this is what I mean here. It creates a campaign and then it assesses how it performed against that campaign. And the reason we're doing this actually is because there's a lot of things in here to test. GROQ4, there's reasoning, there's a lot of math and logic, there's a bunch of different steps which is like kind of multi chaining together tasks. So I specifically Chose this to try to test the capabilities of Grok 4 and also it's trying to lean into the fact it has access to Twitter data. So it gives me a Persona here. Talks about first person I'm juggling client calls, scribbling chaotic notes. Has the competitive landscape gets it pretty much right. I don't know why otter AI is in there. I think that's like a short form video tool. Talks about key differentiators and I will say in defense of Grok 4, I did not give it a lot of context here. If I was doing this for a real company, it would have had a lot of context about that company and it could have leaned into that context, make something really good because these key differentiators is just making it up because there's no public data on this fake company. So it comes up with some key differentiators. It gives the kind of risks of why this wouldn't work, which is pretty true. No one really joins the campaign. It's going to be pretty bad, right? Has a ton of things here goes for a one week validation plan. LinkedIn X Google search, Facebook. How can I prove this within one week? Gives me a messaging kit for solo content marketer ditch note chaos snag five creator hours back for fractional CMO tired of hurting tasks instead of crafting genius for one person. Teams wave by to lost follow ups allowed to draft magic and then gives me a channel mixed budget. It starts to assess itself, thinks it has done a pretty good job that the actual creativity behind this is not very good. And so this one here I thought was interesting. You could play around with it for sure over time you can actually add your real campaigns. Give it a bunch of context about your company. Again, I compared this to 03 and I think it's fine. I think it's just fine. So let's quickly add our last one here. So what we're going to do here is show what content people who bio because it has access to people's bios, right? X profiles are engaging with. So we have retweets, quote tweets and replies because again it does have access to these. It doesn't have access to your bookmarks, doesn't have access to likes obviously anymore and cluster those engagement into clear ranked topics. So what are we trying to do here? We're trying to figure out what our Persona on X is reacting to. So we have a time range here. We have max tweets to analyze. We're going to really push it. We're going to have five Target clusters. Identify all the accounts whose bios include VP of Sales. We could say within the us, within the time range, gather every tweet. This could actually go on forever. Then it's going to cluster all of those together. And I put that into a table because again, we want to start to get some content ideas. The core thing here is what are great content ideas and for this audience. Summarize any recurrent meme formats, hooks or phrasing patterns into those clusters. Instructions to GROK4 stream and analyze X data live with a specific time range. Restrict all analysis to cohort defined task one. Think step by step. Again, this is specifically tailored to test GROK as a reason and engine with access to X. And so this would be great if it worked right? Because anytime I want to figure out what is like good content to produce this week for my job titles that I'm trying to target, this would give me exactly the kind of content they're reacting to on X. So let us run this. Okay, so I've been here for a long time waiting for this to run. You can see it's trying to analyze for VP of Sales is definitely having some trouble actually getting VP of Sales from the bios. I suspect probably what the easiest version of this would be is to create a target list of people yourself and then give it to GROK to analyze. That would have probably cut it way down. So it's kind of stuck here, which is needing to collect all engaged tweets text for clustering into five top groups. But luckily I had run a version of this that is slightly the same. I had work to improve it. What it basically does is very similar. It says, okay, like what's a trend in topic that's related to your brand? And then how can you actually use that to hijack that topic and create timely content around it? So again, trying to lean into the strengths of grok. And so what we do is we give GROK some keywords and then again we give it a time window. Talk about the max tweets to analyze, talk about a cluster in that we only want certain accounts to be featured, which I probably should have put in here, to be honest with you, like a thousand to fifty thousand. I just put this in to try to narrow the scope to actually reduce the amount of time to do this, which I think is a good thing because. But you really wouldn't want to cap. I think the max, you want to cap the min. And so then we return these things here. So we return the volume snapshot. This kind of topic trend cluster, which is what I was trying to do here. Then kind of this same thing. Any recurring media formats, right? Because we want to know like what are the content types I should create around this micro influencer radar. So one of the great things about this is you can actually use it to identify creators and influencers in your space. Then we kind of get into how you can use this to create content. I can get into the first part it does really well, which is the analysis. And the second part where it has to use its own creativity, it does not. So it analyzed nearly 12,300 tweets from the last 72 hours. I put in Grok4 or GPT5 because I want to create some content. Comparing the two, I know GPT5 is not out yet. The benchmarks have just been leaked. If you believe that is true as you're hearing this, you'll know if it's true because this is Sunday and this will be out in a couple of days. So here you can see it starts to cluster things together, provides a sentiment. So GPT5 release rumors positive talks about the kind of rumors about it being released. This one here is the one I really wanted to start to create some content on. Obviously you can't compare it yet, but just like what potentially these could look like. So Grok 4 versus GPT 5 comparisons. I don't know how people are retweeting that it crushes GPT5 because GPT5 is not out yet. AI ethics debate. Grok 4 feature hype so talks about the XAI upgrades the voice mode. Grok 4 real time voice mode is game changing for apps. This is what I have heard is the voice mode and I have not played around with it yet is like the best on the market. Oh, GPT5 whispers last and like rumored memes on AI supremacy when Grok4 and GPT5 argue. All right, so these are memes. So basically says what are some of the content formats, Image macro features, AI characters, quotes, retweets, roast and benchmarks. That's one of the ones makes sense for x one word memes like overhyped or revolutionary paired with gifs in 18% of cluster of three is AI ethics debate. Interesting. Then it has this micro influencer radar. So it actually shows you who are the tweeters. This is actually the most useful part of this prompt and again shows you the value of the X data. So it pulls in the people who are getting the most engagement within this time window and actually gives you the influencers and Gives you their actual content. Now I actually added in an icebreaker because if I want to kind of reach out to these folks and get them to work with me and then it gets into this, which is provide some potential hooks for my content. So let's take one of these GPT5 risks. Got you thinking. Join the AI. I don't know why it's using hashtags. Elon himself says he hates hashtags. All right, safe AI solutions lead the way. Okay, this is just a whole bunch of mess. The reason this is a mess I will give GROK for an out here is I didn't actually provide a good context of the type of content I want to create or who for. And then it just says why it matters. Hi. Check. And that GPT5 release rumor cluster could amplify your brand reach. So what's it done a good job of here? I think this is pretty good actually where you can start to break it down. The influencers, micro influencers in your space. This is kind of cool if we perfect it that we can cluster content together from all of the tweets in a three hour time range about a specific topic and then you can figure out like what is actually trending within my audience. Okay, so what have we done here for the recap? We did our mimic style using the data on X to show how good it is at trying to craft something from one of the creators on X. We used a lot of reason and logic to be able to do that. We created a campaign for a fake brand again with a bunch of different steps chained together using X, using some of the things that Grok 4 should be pretty good at. And then we use it to analyze content that is trending within our audience and cluster those together and then pull out these kind of micro influencers and then have GROK suggest content that would play really well within this space. And now what are some of the things we've learned now? You didn't see that I actually had compared a lot of these to O3. Right. I'm not going to go through the comparison to O3. And the other thing I would say what you should learn here because I'm doing this to test Grok 4, but because I didn't provide a lot of context, I used a fake brand in the marketing campaign. And then for the one where we analyzed clusters of content and then had GROK suggest what kind of content we could create around those clusters, I really should have a whole profile from my ideal customer profile. I should have examples of content that we've created that have done really well so it had something to guide its output. But aside from all of that, I think Grok 4, when I compare it to O3 for a lot of these kind of marketing knowledge worker growth use cases, it's fine. It's not incredible. I have not found it to be incredible. Maybe I will as I continue to like work with it prompt that improve my own prompts. But at the moment it's fine. It's just fine. I don't know why it performs so well on the benchmarks. I was really excited to try it out. I thought it was going to be a total game changer. I was going to just like be totally blown away and thinking wow, like I can just get such incredible results. Things that I could never do at 03. I'm going to be like the happiest person ever do doing amazing things here in terms of how I can use Grog 4 and I don't feel that way. I feel it's pretty fine. I'm going to use it for a bunch of things where I need access to the X data, the Twitter data. I'm probably not going to gravitate that much from O3 for it, but I'm going to use it for like certain use cases and that's it. I think that's maybe the end of the pod that it is fine. So go try Grok 4, try some of these use cases and tell us what you think in the comments.