Garbage-in, garbage-out. This term resonates, doesn’t it? It doesn’t matter whether it’s the high-school athletics coach encouraging you to eat salad, or the motivational coach warning you off TV, garbage-in, garbage-out makes instant sense.

It jumped out at me the other day as I glanced at the title of an email newsletter from one of the leading AI copywriting firms (which I won’t name, to not cyber-bully with my massive platform hehe). By the way, when I say “leading”, I mostly mean leading in startup funding.

It’s fair to say that since I last picked on AI copywriting (wrote about Google and MS AI here then Persado here), there’s been a quiet revolution in the space. Mostly, this comes from the Elon Musk-cofounded OpenAI project. While every FANG has its AI lab, and while a few academia/venture-backed AI labs are out there, OpenAI holds a pretty unique slot in the space globally, at least in its three areas of specialization: natural code generation (as in software code, yes), image generation and editing, and natural language generation, the latest version of which is GPT-3.

GPT-3 has in its short two-year history spawned over 300 apps. As its website says, “To date, over 300 apps are using GPT-3 across varying categories and industries, from productivity and education to creativity and games.” (Was that written by a human? Is this written by a human?).

Of course, I haven’t had time to review them all, but as a long time marketing and messaging consultant, I have a special interest in GPT-3 apps that generate copy.

And I don’t even have the bandwidth to analyze all of the major AI GPT-3 copy/content writing firms on the market. But I’ve looked at a few, including most of the best funded ones: Shortly, Jasper, Copy.ai, Writesonic, Copysmith, Rytr, Speak.ai, ContentBot, Anyword and a couple others. For the most part, I have looked at others’ reviews of these apps on YouTube but I have tried Jasper, Copy.ai, and Speak.ai myself. Jasper seems quite good.

To be clear, though, this isn’t a software review article. Instead, I want to make a point that I think applies to any kind of AI, not just copywriting AI: garbage-in, garbage-out is a faulty basis for a business model yet many GPT-3 firms tend to embrace it.

Back to the subject of the email newsletter from an AI copywriting firm. It claimed, “Enter two keywords, and get a blog post”. That’s so convenient, so easy. But I actually tried it (many times) and it doesn’t work. AI will save you manual digital labor, but it will not save you *that much* intellectual digital labor. You have to feed it not just more than two words, but actual thought.

The promise is: put in 10 words, get 100 back.

But’s counterintuitive. That’s not how I write. I put in 100 words and get only 10 back.

Look, I get that AI is an assistant and not a replacement. For now. I get that its job is to generate ideas. Let’s call that’s a given. As Persado customer Chase said 4 years ago, “Chase plans to use Persado for the ideation stage of creating marketing copy”.

I bet that works for Chase because they don’t put garbage into the AI, they put thought into it. This is the basis (or should be the basis) for what OpenAI calls “fine-tuning”. They don’t frame it like that because they think like Big Data people - insight through scale.

Well there’s another way to derive insight - it’s the way that consultants do it, known as peeling the onion, 5 questions, socratic dialogue, intellectual sparring, asking why, etc, etc. All of it goes deeper into the mind of hand-selected individuals, rather than broadly aggregating the minds of millions of individual’s through their publishings, as the OpenAI index does. It uncovers the reasons-why that otherwise remains covered up. That’s not garbage, that’s the black swan, the black diamond.

Open AI’s broad input model obviously works, but the way to fine-tune it is to go deep - and feed it black diamonds instead of garbage.