Issue #23: Office hour replay, Midjourney for products, better than ChatGPT?
A few weeks ago, I held a public open office hours.
I went deeper into AI for marketing, Code Interpreter, prompts, and so on.
Lots of people signed up and turned up—thank you!
You can now watch the replay here.
Here’s what I covered:
0:00 - Intro
3:41 - Prompt Flow: From Comments to Tweets
7:55 - Questions and Answers
22:44 - AutoGPTs
34:36 - Prompt Flow: Plutchik's Emotion Analysis to Copy
46:00 - Code Interpreter
1:02:35 - More Questions and Answer
1:08:28 - Prompt Flow: From Conflict to Landing Page, Emails
1:22:03 - The Next 3-12 months of Marketing and Copywriting
Great discussions, answers to questions, and I do prompt walkthroughs as well.
I’ll do more of these (and announce them here on Bionic Marketing).
Speaking of AI and video, you can now also watch and listen to Issue #20:
Let me know if you can tell what part of my voice is AI generated vs. me-generated. Listen closely.
Anyway, more good things are coming soon, like tutorials.
In today’s issue:
LLMs/GPTs in your marketing & business (how to decide).
Product “photos” with Midjourney (easy tutorial, crazy results).
Generative AI saves marketers 5+ hours per week.
How AI is "killing the old web" and making it new.
Using ChatGPT for data enrichment (great for leads).
Let’s dive in.
Last week we talked about predicting your customer’s behavior and how to do so more accurately by using AI.
This week, let’s take a deeper peek into how it’s made possible (and how you can do it dirt cheap).
In a recent post, I shared some of what I’ve done over the past few years (and still do) with clients (integrating Large Language Models, GPTs, etc. to scale).
What, specifically, you do all depends on your desired outcome.
Most marketers and businesses will do fine by just using existing tools (ChatGPT, Midjourney, and so on).
Others should use a search vector solution (like a chatbot), especially for simple tasks like a support bot.
Beyond that, you can unlock more value for your work with a fine-tuned LLM, either open source or with GPT-4.
It’s not incredibly easy but it’s simpler than it used to be.
The final step beyond that, as far as LLMs go, would be to actually train your own. But it’s still somewhat expensive, takes time, requires development work, resources, etc.
The higher up the levels you go (with a custom-trained LLM at the peak), the more value you create internally—and the more use you’ll get out of AI and ML.
But, regardless of all that—every marketer and copywriter should go beyond typing prompts into ChatGPT as soon as possible.
You’ll have more control and better output if you access GPT-4 via API.
And you don’t need to be a technical expert.
So, let’s grasp an understanding of what’s going on behind the scenes to help you better organize your marketing efforts to scale.
But first, what does API even mean?
To keep it simple, these Application Programming Interfaces are what allow systems to communicate with other systems.
And as per the post, you can use different ones to personalize and train your own GPT to “see around corners” and strengthen strategy, marketing, sales and even snoop around with competitor intelligence analysis.
API calls collect all that data, crunch it, and organize it for better dashboards, streamline workflows, and create highly customer-focused (even segmented) campaigns that give any competitor a bumpy seat in the back of the bus that thinks AI is a joke.
And with the new AI Cloud that Salesforce dropped, they are going to put a mark on those virtual assistant mills—with MarketingGPT, SalesGPT, CompetitorGPT, etc.
But you don’t need Salesforce. You can create your own LLM/GPT ecosystem by building out models via API—for as much as it costs to have GPT+.
Because if Salesforce has all this, you know Open AI probably wants you to have it, too.
You can now, basically, make API calls to plugins—and have execute tasks.
LabLab AI also compiled a list of podcasts that put all that marketing AI into context, covering everything from why you should be listening, practical picks and data skeptics to how to actually use the information you get from the podcasts you listen to.
If you’re not into the nerd-side of AI with API calls, and just want the easy peasy stuff…
This simple Midjourney tutorial shows you how to stage an Instagram-worthy product photoshoot, all without a camera.
Of course, it’s not as simple as asking Midjourney to “create some cool photos of bottles for my skincare brand”
The tutorial involves the entire creative process, from setting out the vision to nailing the perfect composition.
A combination of MidJourney, Qreates Studio and a sprinkle of Photoshop are capable of creating some truly stunning results, straight from a web browser:
What would you do with an extra five hours a week?
According to ZDnet’s breakdown of Salesforce’s “latest research,” marketers can save a whole extra day in productivity gains with the help of AI.
Or just do work, faster and easier?
According to Adobe’s Future of Digital Experiences Report, 95% of marketing professionals believe generative AI is a creativity enhancer.
In other AI marketing news:
The future of marketing jobs: OpenAI Plans ChatGPT ‘Personal Assistant for Work,’ Setting Up Microsoft Rivalry. You’ll see things like marketing assistants become more a thing in the next 3-6 months. You can already have one, it just requires some API calls and webhooks.
What’s better than GPT? This: Google DeepMind’s CEO Says Its Next Algorithm Will Eclipse ChatGPT. Demis Hassabis says the company is working on a system called Gemini that will tap techniques that helped AlphaGo defeat a Go champion in 2016.
Online marketing is changing because the internet is: AI is killing the old web, and the new web struggles to be born. Generative AI models are changing the economy of the web, making it cheaper to generate lower-quality content. We’re just beginning to see the effects of these changes.
How to use ChatGPT for enriching your leads: Using ChatGPT For Data Enrichment. Submit an incomplete street address and have ChatGPT enrich your database with the state, zip code and country code for the address.
Good times all around.
And the cash continues to flow into companies.
$90MM: Synthesia, an avatar creation app, hit a $1B valuation in their latest round, led by none other than Nvidia. Superpowered GPUs, combined with Synthesia’s potential. We’re about to see some spooky good deep fakes.
$33MM: Striveworks is catching momentum outside of itself with its ML Ops that you can “deploy in days, not months”. It’s seen its ARR grow 300% two years in a row.
🤖 Definite lets you take that labyrinth of spreadsheet information and translate it into useful, applicable insights using plain-text English.
Shoot it simple questions and receive impressively intelligent answers.
Stop flooding the data-requests channel and instead simply ask, “Can you create a chart that displays our most successful marketing copy?"
The app operates exactly like a personal data analyst, where it shows up inside your Slack, Salesforce or Hubspot interface, whom you can message on-demand and available 24/7.
🤖 STUDIO AI is reinventing web design. Its interface includes an easy-to-use Photoshop-style editor that can create sleek, Squarespace-style sites with all the style yet none of the code.
Since the app uses both WebDesignAI and a ChatBot, you can input direct requests to finetune your designs, like: “Make all the buttons yellow”, or “Add space for a landing page”, or “Change every Helvetica font to Roboto”.
It can understand what you are designing, learn from your feedback to take them further, and turn those designs into live websites, instantly.
THE LAST BYTE
If you’re not running LLMs locally on your computer, why not?
With GPT4All, it’s the easiest thing in the world.
Download the app, install, and download an open source model or two.
The “best” ones are on par with GPT-3.5 in many ways (like Falcon or Phi-1).
Prompt away with no restrictions, filters, or censorship.
You can even interact with your own documents.
I do this (and have several instances in the cloud, too) and find it immensely helpful for exploring prompts and getting output that ChatGPT will decline to give.
There’s nothing like running the same prompt across multiple models, and getting the variety in output.
Which brings me to the most important point I can make this week:
There are few better skills to hone that curiosity.
Over the years, I’ve spent hours having GPT-3 (and then GPT-4) write novels, plays, argue with me, role playing, and so on—mostly for fun and sometimes for profit.
I’ve interacted with LLMs for years now. But I’m still mesmerized when I get usable output from a really bizarre prompt that I’ve been tinkering with for days or weeks.
You can literally type anything into a chat box—and get an intelligible response. Often, a response that solves a problem or helps with work.
On the mundane side of things, it can produce any kind of marketing or sales collateral.
On the more interesting side of things, it can solve complex problems, help you “see around corners”, or just write really good pulp fiction that people will pay for on Amazon.
What a time to be alive.
See you next week,