Transcription
Someone just figured out a way for you to get paid to use Claude Code, and it's insane. There's a new tool that will let you newsjack to get publicity for yourself. You'll see it. Elon Musk just launched an AI tool that will make cinematic videos. All that and so much more. I've got timestamps to everything below, and of course, links in the description. Let's get into it.
Presented by Zapier, the AI automation company.
All right, Corey. The first drop this week to talk about is this one. What is "Get Paid to Wait"?
>> So, I think this is absolutely genius. So this guy created a uh, it's a plugin for Claude Code, and it really can plug into any of your IDE, any of your coding agents, and it shows ads inside your Claude Code while uh, while it's working, right? So the kicker is that it'll actually give you 50% of the advert, like what the advertiser spent on the ad, as a kickback if you click on the ad.
>> And this went super viral. It's at 7.7 million views.
>> He posted a follow-up tweet. He was at $3.7 million run rate in under 24 hours after tweeting this. So, I think that just affirms that if you've got a good idea and your execution is good, yeah, you know, he he benefited from a viral tweet that really helped promote it, but like this is the best time to be building anything. This is such a smart idea.
>> What a great, like this little screen here where it says "discombobulating" and whatever it is to make it look clever while it's doing work is something we've all stared at for hours. And he's got an ad here. And what a perfect group of people to advertise to. People who are geeky enough to actually be using Claude Code in something like VS Code. Oh, what a great product. All right, I love this idea.
Let's look at the next one. Firecrawl. I don't know what Firecrawl is. Let me read the tweet and then explain why this one is so important to include today.
"Starting today, you can try Firecrawl for free without an API key. Search, scrape, and interact with any web page. Plus, parse any PDF into clean markdown with no setup at all. Start using your endpoints and only sign up when you scale."
What is it?
>> So this is, I'll I'll tell you why this is important because, you know, how a lot of times when you try to search, like if you're using like Claude or just any any AI tool and you have it go and, you know, use web fetch to search a website, and a lot of websites are blocked, like a lot of websites don't let AI tools scrape it. So Firecrawl kind of just gets around that, right? And previously, in order to use Firecrawl, you had to get an API key. Sometimes you had to pay for API credits. It wasn't a huge burden, but it it did require some setup. Now Firecrawl is saying like, look, you can use it for free, period. No strings attached. No, even no API key required. And the only reason you would ever pay for it is if you start to use it as more of a power user. So, this is really going to unlock scraping and agentic search on a lot of websites that previously, like, you know, your everyday Claude or your everyday ChatGPT couldn't access.
>> Oh, that's great because even when I do searches for this show, it's a pain because X will block me, YouTube will block me.
>> Yep.
Okay, let's go on to the next one for today. Uh, newsjack.sh. "Open source skills that turn any agent into a full PR team."
So, I think this one's really cool because again, the biggest problem for most people, especially founders, is getting distribution. And this is an open-source GitHub repo that essentially has a team of agents go and search live news, X, RSS feeds, kind of figure out what is trending, and then drafts briefs and sends it to those websites or to those reporters with an angle for you being a contributor to those news sources. Right? So, this is one that I personally want to try out because I think this is going to help you get a lot of earned traditional media that is otherwise closed off to the general public. Like otherwise, you need to know one of these reporters, you need to have a relationship. This is kind of a way to do like cold PR outreach in a very personalized manner.
>> Yeah, I have seen that if a reporter is covering a story, they they want to know more about it. They want to do another follow-up. And if you can tag on to that, you could be a part of that follow-up. And I could see how this automates it.
All right, we'll include a link to this, of course, in the um, in the description. This one's unusual for us. This thing went super viral. Just a few, I guess it was yesterday when they posted it, late last night. Um, this video didn't really say much, but everyone from Elon Musk to Nikita Bear to a lot of our friends talked about it. This video that you shared with me says a little bit more. I'll just um, I'll just scrub through it while you explain what you think this is, 'cause it's still a mystery.
>> Yeah. So, they didn't really give us like, you know, this is a four and a half minute video. Obviously, we're not going to watch the whole thing. We're just kind of skipping through it. But even if you watch the whole thing, like, it's still kind of unclear what it is. From what I can tell, so Midjourney, for the folks in the audience, Midjourney is an image generation AI tool that lives in Discord that doesn't even have a web app. Like, to use it, you have to be in Discord. So frustrating.
>> They made one of the the biggest pivots that I've seen anybody in the AI world make. They went from like a Discord image generation bot to now their uh healthcare startup is kind of what it looks like. And this Midjourney scanner that they released, it it looks like it it is essentially going to do like full body scans of your body, your vital signs. Like here it's kind of giving us the things that it's scanning, like your vertebral processes, your spinal cord, your...
>> Your like, I don't even know what these terms are. They're like medical terms, but it's it's going to do some sort of like medical scan. I assume for, you know, prevention of serious diseases or just help biomarkers. Again, we don't know a lot, but that's kind of what it looks like.
>> You know what? And I was worried about them because they were the first or one of the first ones to do AI-generated images and they stuck with Discord, which nobody can use. And then all these other competitors came in. And now I can see where they're going. Makes sense if they can pull it off. Uh, I love these types of tweets.
For the next one, it's Headroom. Uh, where someone says, "Hey, so and so or someone on the internet built this and that." But this one actually is a really good source. A Netflix engineer built an open-source proxy that cuts AI token usage by 60 to 95%. Um, and this one's been especially popular uh during that small window of opportunity when we all had to use, when we all got to use Fable, and it just killed all our token usage, and what it was supposed to do was save us tokens. Right?
>> Correct. So what's interesting about this one is again, it was created by a Netflix engineer, so it has credibility behind it already. And it's, it's a, I guess essentially a plugin that sits between your app and the LLM, right? If you're using like Cursor, Claude Code, whatever.
>> And uh, and it kind of like compresses certain aspects of the, you know, of the output.
>> And they claim like zero, zero loss of quality, right? It said quality degradation was like 0.0000000. But the So what's interesting to to find out is that the co-founder of the Hermes agent...
>> Yeah.
>> Used it as a test with Hermes and found out that it actually cost more tokens to use this with Hermes, except for one specific use case.
>> So he's like, for the one use case where it did save tokens, we actually integrated it directly into Hermes. For everything else, it actually cost us more tokens. So, it's one of those examples of like, you can't just blindly download a tool and expect it to do what it says it does. And it's another example of knowing when to use the right tool for the job.
>> Right? Like it might make sense to use for Claude Code inside Cursor, but if you're going to use Hermes agent for anything other than the one use case that it mentions, it's actually not going to do what it's supposed to do.
>> Great tool, but yeah, worth checking out for yourselves. I saw Elon Musk tweet this. I said, "I got to try it." Grok now has better videos. And look at this one. And then I'll show you someone else's example, and I'll show you what happened when I used it.
>> Yeah, Grok 1.5. It's pretty good, I guess.
>> Oh, shush. I'm sure it's good for other things, too.
>> Yeah.
>> What you notice about it is it's not just better video, but notice her lips and the words and the sounds, how they match up well. "Grok 1.5. It's pretty good, I guess."
>> And like the facial expressions, like it, it is truly flawless. Like I was trying, like I was looking so intently trying to be like...
>> Trying to find any evidence that this was AI-generated, but this to me looks like a commercial that would be on TV, like a real one.
>> Yeah. Even like her hair, it looks off, but it makes sense that it would be like that. I don't know what this is over here. It looks like half an animal with a towel on it. That's the weird thing. But overall, damn good.
And then on the website, they show how you can use this as part of real movies. I then said, "Let me go and try it." I basically said, "Here's my prompt." I said, "Landscape mode video." That didn't, didn't even do that. Um, a woman sees a fort in the year 2050 and she decides, and there's nothing 2050-ish about this, and she decides she wants to take it over using nothing but Nerf guns. She goes, shoots, and takes it over. Let's see what we got.
"I see that fort. It's mine now. Time to take it over."
And >> that to me, like those Nerf guns look super AI-generated. Even her, like she looks AI-generated, you know?
>> Yeah. Like even the, the what comes out of the Nerf gun is weird.
>> The fact that one popping out of the fact that she's randomly walking. It seemed like it was barefoot, but I guess not. It's nothing. Nothing. Even the basics didn't come out right. So, I thought, maybe it's just me. And honestly, maybe it is just me.
So, then I found this video from AI for Real, from a creator who is really good with videos. I like this guy because what he did was he tried it against another model and then he wanted to compare it. He tried to use the exact same prompt, and here's what happened.
"Exact same prompt that I used for the other ones, but I could not do that because this one maxed out at 2500 characters. So, with that maxing out at 2500 characters, the other one was longer than that for seed dance. That's something to keep in mind. So, I couldn't do the exact same prompt for this one. So, I went ahead and modified it a bit, cut it down a little bit, but try to keep the same elements and same bot. So this is the output that we got very first time from Grok, imagine 1.5. 720p is what they max out at. So if you look at this, it's actually really solid. And three to six months ago, I probably would have been flipping out over this because it's really solid. But now that we have gone to a different level, you have to pick it apart a little bit more. So, look at this. And there's a few things that I'm going to point out. And they're not necessarily that big because you can rerun them or comp it out, but that branch right there, or clip it out. But that branch right there, it goes through, gets them through the bus, ends up outside somehow. And it did not only do it that one time. I tried it again just to see what happened. Same thing. That branch is there."
>> Okay. There's two things that I'm noticing. Number one, he's getting better results because his prompt is better than mine. Mine was pretty short. Uh, he's also doing this through Higsfield. And then the other thing that I'm noticing is that even for him, there are clearly problems. So, it's great, but it's not...
>> It's not fully usable yet.
>> I, I feel like, yeah, with the AI video, it's like 90% of the way there. Like it looks so good in many cases, but then, like he said, it's like there's specific nuances that make it a deal breaker. Like you couldn't put that in an actual movie or like TV show. Like people are going to pick that out right away. But I mean, give it another three to six months, and I feel like it's going to be there.
Okay, final big release. New number one open-source model, and the gap is insane. So my favorite lab, ZAI, just released GLM 5.2.2. And get this, it's by far...
>> Look at the numbers. It's the only thing that beat it was Fable 5 on this benchmark model out there. Plus, it even beats the best GPT and the best Gemini across multiple benchmarks. That's how ridiculously capable this model is. In this video, we're going to go over all the incredible things that...
>> I'll have a link below so that people can see the actual examples. What do you think of this? You've used it.
>> Yeah. Well, so I haven't used this particular model yet, but because it only came out within the last day or two. But I mean, this is going to be a...
>> What did they say? Like 1/16th the price of of GPT 5.5, but it's even more capable. So this is just, I think, one of those examples where it's going to force the cost of inference down. Like, there's no way that Anthropic and OpenAI can afford to charge the prices that they're charging when there's better models out there for, I mean, practically for free, right? Like obviously not completely free, but very, very cheap. So I think this is just one of those things that's going to drive the cost of tokens down, and it's good for the industry in general. Like the better these models get, the cheaper they're going to be, and the and the higher the demand is going to be because of that.
Yeah. I think now that it's getting cheaper and cheaper but more effective, I've got to start adding this into my workload. The way that I've been thinking about this, Corey, is I might as well pay for the best because it's not that expensive. 100, 200 bucks a month, and you get the best model that there is.
>> But I still have issues like I did last night, where I was told that I had to wait two hours for my token thing to reset, and I go, "Dude, I I can't deal with this." And I was about to pay for more, but it was going to be a pain. And so I just moved on and went to pick up my kid from daycare, but I should just have a backup so that I could at least try it and start putting it in uh, into my workload. And another good use for this is Hermes Agent, which does suck up a lot of tokens.
>> 100%. Yeah. So, with my Hermes Agent, I just have the Codec subscription. So, I've got the $100 a month OpenAI plan running it that way. And um, this is a great backup if you hit your limits or or as an alternative if you don't have an OpenAI subscription.
All right, great. We'd love to see more of what you all are creating. So, if you have anything, share it in the comments.
>> Absolutely. We'd love to feature more people building in this space. Yeah.