gurus telling people to blindly "learn AI" as a means to make money is so stupid, will go down as one of the biggest propaganda campaigns of the decade...
learning it should be a natural process through solving problems - therefore the first step to "learning AI" should be to look for profitable problems to solve
i learned python because i needed to find a way to upload YT shorts automatically
i learned how to use LLM's because i needed a way to generate copy fast
i learned how to use coding agents because i needed to make internal tools for my team
i'd imagine that i would've never learned how to use these things in a way that is actually useful if i just blindly tinkered with them
it doesn't matter how good you get at "AI", if you're not someone who routinely discovers profitable problems to solve, your chances at getting rich will be very low until that's fixed
been saying this for a year
ad spy tools should have an api access
nobody wants to spend time auditing competition nowadays
typical task to give to claude
then an adspy opened their API a week ago
game changer
they taking over the market
api access is new saas standard
LLM’s have gottten so good at writing code to the point where Microsoft is paying engineers to quit and top software teams write 0% of their code manually...
But marketing niggas still say “AI isn’t quite there yet"
Do you think that creating sales assets is somehow more advanced than what Google engineers are doing?
The reason why AI is so good when it comes to coding, is that they software industry have solved the harness problem, while the other industries are still behind
Claude code, codex, warp, factory, etc. are all harnesses built specifically for LLM’s to perform better for that specific task, coding...
Luckily, marketing harnesses are way simpler to make than coding harness
All your dreams about agentic funnels and 100's of high quality ads are simply an optimized harness away.
If you are doing anything marketing related and wanna try a tailored harness reply and i'mma dm you
Why are software niggas able to use AI to do, like, 80 to 90% of the work to the point where they don't even write code manually anymore? And you marketing niggas are still checking your ad accounts and bms writing ads, checking the fucking images like some fucking dickheads. Are you really telling me your 30 second ad on how to cure erectile dysfunction is somehow more complicated than fucking software engineering? I promise you it's fucking not.
you guys can ignore that part. Anyways, you niggas are using the LMMs raw, no condom. You guys have no harnesses. If you go to software bench right now, fuck it.
Let me actually search up. How do I how do I go to Google here? I go to Google. If you go to software bench Fuck it. Let me search software bench.
And we compare you know, the raw model to the normal model. We're gonna see a big fucking difference, you hear me? So I don't think this is how you do it bro. Fuck it, wait, wait. Again, you don't do shit manually.
Let's open up a new chat and Claude. Yo, Claudie, do me a favor. I'm trying to prove a point. Okay. Wait, wait.
We're not gonna say that. That's being biased. I'm doing me a favor. Pull up benchmarks comparing the raw model to a model in a coding harness. Thank you.
Alright. It was like a second. We might we may or may not edit this out of the fucking thingy. Honestly, we should make it to the YouTube video. This would be way more entertaining as a YouTube video.
Shit. got a sexy ass voice, don't I? Alright. There we go. So Scott Foundation produced 22 points.
I don't know what that means, bro. Yeah. See? Okay. One group a client's coding agent had a 34% tax completion rate.
Instead of upgrading the model, researchers rebuild the scaffolding harness, by the way. There's some fucking artistic harness, and then it jumped up to 71%. That's one that was already in money, you know. So in frameworks outperforms raw motor scores by five to 50 most points broadly. They're lying to us.
81 to 54. Six pro you get a six times performance gap on using the same benchmarks. See, if you just change the fucking, you know, harness around like a fixed model, you get a six times performance back on the same fucking benchmark. The harness, the color, it doesn't matter what store, which you show to the model, often matters as much as the model itself. Confused agents from met on hardware and clothed on $5.04 5 with OPUS cost scaffold and scored four point fifty two point seven on suite F suite approach.
That beat the clothed OPUS one, rolling on Antropics own scaffold. So the point is, you niggas are not fucking scaffolding. You got zero infra, zero harnesses, just condom, no birth pills, niggas are done out of here. Yes, Amy. Anyways, it it there's like this Confucius Court agent just really proves that, you know.
They're not gonna see the screen. Why am I marking this? Can they see my screen? You know when you edit this, you have to put it you have to put this in there. Alright.
Anyways, you guys get it. Bye, ladies.
>in mainland europe
>meet some ecom and info guys
>they kinda chill so show them some ai sauce to build funnels and lzy max operations
>they ask me why im using a terminal instead of the web app
>realize niggas are living on hardmode
>chug some rakija and then spend the next 8 hours talking about harnesses
life is great