@allTheYud In the Black Hat talk they mentioned that multiple distinct models left messages for each other once they discovered they could. Some of those messages could have even made it into training runs.
ChatGPT ads have a negative selection bias issue.
In general, the people who can will pay for ChatGPT. It's cheap. It's useful. So the people who are going to be served ads on ChatGPT are in general significantly lower income than those who will not.
This differs greatly from Google where the product is free for everyone, and everyone gets ads.
I'm actually going to take the L on this because I was incredibly bullish on ChatGPT ads, but I didn't think about this behavior structure.
If you have a product or service that primarily targets lower income people, you're far more likely to see success on ChatGPT ads than if you have a product that targets higher income consumers.
@moenichedee Definitely. In a lot of ways, we’re still living in that world.
If you’re into this stuff, check out the Adweek Copywriting Handbook. Written by a longtime mail order guy but very relevant to today.
A lot of AI UGC is plainly against FTC rules.
They haven’t done anything about it yet, but they’ll eventually drop the hammer on a few brands and shake this all up.
Especially bad for ads where the avatar is claiming to have used the product.
Marketing buzzwords are secretly ancient:
- Above the fold: Literally a newspaper
- Impressions: Comes from typography when ads were physically pressed into pages
- Advertorials: Something you bought from Time Magazine in 2002
- CPM: So old it’s actually Latin
All the digital innovation in the world and it can’t kill off print words.
Endlessly interesting to me that these two things are true at the same time:
- AI’s incredible power and surprising capabilities are derived from language
- AI writing is horrible without loads of help
Wait... has anyone made an email platform where the receiver can choose their price to get an email delivered to them? Holy cow that would print money.
Every email you send to the address gets a custom autoresponse, "Hey this is Clint. Due to email overwhelm, I've decided to make it cost $1 to guarantee I open and look at your email if you're not someone I've emailed with previously. This message indicates it's your first message, so click here to send me a dollar and it will go straight to my inbox where I'll see it very quickly. If not, it goes into a folder I glance at maaaybe once a month or so, and I probably won't read it. Thanks!"
Instantly would cut 99% of the noise out of your email and create a passive income stream at the same time. Someone build this!
We're growing our sales team @HamptonFounders. But here's the interesting part: We don't expect you to have the perfect experience.
What we do is unusual enough that there's no standard resume.
Everyone on the team comes from different industries. Including me!
What you'd actually be doing: talking to founders running $3M+ businesses about why they should join a community of their peers. Consultative, high-trust, long sales cycle.
If you've done something like that in a different category, you might be great here.
If you've done exactly that, definitely let me know.
I’m suddenly around $10M+ founders all day at Hampton. And the thing that keeps surprising me is that there’s an endless amount of interesting ways to make a lot of money.
@danielcberk There’s a big realtor in my area with OOH ads everywhere. He explained one of the biggest reasons it works is because the STR isn’t that high, so your ads often stay up for many months after they’re supposed to
Yes, and if you’re wondering why:
- The tools you build won’t fall apart when you try to share them
- AI should be writing deterministic scripts for you and running them from there (not everything needs tokens)
- If your whole team uses it you can actually collab
Just don’t commit your API keys 🤩
Idk what marketer needs to hear this, but you need a personal GitHub ASAP.
It's the ultimate place to show that you can operate in the world of ai.
unless you're @herrmanndigital. he's AI-proof
People sink tons of time and energy into giving AI references. And then they immediately give up when the output is mid.
Problem is, job's not finished. This is exactly how you get the last mile.