The Code Interpreter for Marketers (plus latest news)
The chat battle is heating up.
Depending on the prompts I use, right now, GPT-4, Bard, and Claude 2 outputs are the same in “quality”.
They’re incredible close in generating the output I’m looking for.
Some prompts I modify for a model, others are exactly the same.
You’ll see more Large Language Models “compete” against each other in the next few months.
Bard and Claude have a small advantage in that they both allow you to upload files (and interact with them) without having to use a Plugin (like you do with ChatGPT).
But the Code Interpreter function of ChatGPT is incredible and allows for file upload, as well.
I’ve had alpha access for several months and now Plus users have access.
And it blows everything else out of the water. I expect that to change but for now, it’s the “best” in the game.
In today’s issue, as I mentioned last week, we’re focusing on
Latest news on AI for marketing.
The Code Interpreter tutorial.
I’ve got 6 most important things for you from the past couple of weeks:
1. As mentioned above, ChatGPT is getting stiffer competition.
Anthropic launches Claude 2 and it’s now available for public testing in the US and UK, and it's available with a 100k context window (enough to write a book or sales page or, really, anything in one shot).
The model also has significant improvements on coding, math, and reasoning compared to previous models.
I tried out several of my copywriting and marketing prompts—and it nailed it.
Upload an image and it can describe it to you, extract the text, and more. You can now also more easily adjust the tone and style of responses, choosing between simple, long, short, professional, or casual tones—and you can listen to the responses in 40+ languages.
If you had previously written off Bard because the responses were terrible (they were), try it again. I’m now using it daily.
2. Speaking of Google, have you seen their NotebookML?
Their AI-backed note-taking tool, NotebookLM is launching to a select group of users in the US.
You can select multiple documents and ask questions or create new content, automatically summarize long documents, convert video outlines into scripts, generate new ideas, citations are integrated, and lost more going on.
Basically, Google is trying to give you your own personal AI, trained on your data and notes.
You can request access via the waitlist on Google Labs (click the link above for the announcement and waitlist link at the bottom).
If this is true, then I might just try it out:
We’ve built NotebookLM such that the model only has access to the source material that you’ve chosen to upload, and your files and dialogue with the AI are not visible to other users. We do not use any of the data collected to train new AI models.
3. Next, turn your sketches into images (perfect for marketers).
Stability AI releases their sketch-to-image tool, Stable Doodle and it’s a lot of fun.
Imagine talking to designers or anyone on your team and trying to explain something. Just sketch it out and Stable Doodle will turn it into an image.
It accepts both a sketch and a prompt to generate images and it’s made for both professional artists and novices, offering more precise control over image generation.
The quality of the output depends on the detail of the initial drawing and the prompt but it’s getting it pretty close to what I imagined every time I try it.
4. You’ve probably seen all the ways you can clone your voice (or, really, anyone’s voice).
ElevenLabs is one of the best tools out there.
Soon, you’ll see tools that provide completely natural-sounding voices, entirely generated by AI.
They’ll be used for Customer Support, Sales, and plenty of other things.
Air.ai is another excellent tool, and it can perform full 5-40 minute long sales and customer service calls over the phone that sound like a human.
And then perform actions autonomously across 5,000 unique applications.
As it says on their website:
“It’s kind of like having 100,000 sales & customer service reps at the tap of a button.”
There’s a waitlist to join but it is a tool that’s working right now on live calls, talking to real people.
And it’s not just for Support or Sales calls. You can use it for anything (therapy, talking to The Terminator, or anyone else you can imagine).
5. AI continues to roll out across every creative field and job you can think of. This time, it’s coming for web designers and developers.
Type a description into a box and answer a few questions, and you’re done.
Yes, you can create a complete website with text, images, and more in mere seconds. You can have text, images, layouts, product descriptions, and a lot more done with just simple prompts. They’ll add more features in the next few weeks and months.
6. Finally, another force to contend with in the raging LLM battle are open source models.
With the release of Llama v2, a model from Meta, open source LLMs just got a huge boost.
Open source LLMs are making headway fast and a few are as good as GPT-3.5 already.
Soon, I think we could see GPT-4 level quality for most tasks.
If you’re reluctant to use GPT-4, Bard, or Claude for your business because of proprietary information or data, there’s a case to be made for you fine-tuning an open source LLM, custom and based on your data, that will outperform them for your specific use cases (because it’s narrow).
I’ve seen it work incredibly well and fine-tuning is only getting easier (and cheaper).
I’m suspicious that the Code Interpreter is GPT-4.5.
Even though it’s, essentially, a Jupyter notebook for running Python, it does have a larger context window, has many more functions added, and when I stick with prompting for text, it does a better job, faster, than GPT-4 does.
If none or little of that made sense, just think of it this way:
The Code Interpreter is basically a juiced up chatbot that can run code, handle files, analyze data, create charts, process images, video, and more.
It also seems to have a much larger context window, meaning you can have longer prompts in one shot and get more output.
I’ve tested and managed to get it to accept files up to 100MB upload or download (including .csv, .xls, .png, .jpeg, .mov, .mp3, .epub, .pdf, .zip)
Lots of smart people have figured out that it comes “preinstalled” with over 330+ libraries like pandas (data analysis), matplotlib, seaborn, folium (charting and maps), pytesseract (OCR), Pillow (Image processing), Pymovie (ffmpeg), Scikit-Learn, PyTorch, Tensorflow (ML)9, and so on.
Again, this is technical stuff but it means you can do a lot of really useful things.
Because it’s flexible and you can infinitely combine all the capabilities, it’s impossible to say for sure just how much you can do.
I’ve yet to run into anything in ordinary marketing and copywriting tasks that it can’t do.
You can have market simulations created, dashboards, preprocess data, clean data, draw memes, analyze complex data from anywhere if it’s exported into a spreadsheet, help you make sense of data, create diagrams and charts, and more.
Imagine you had a data analyst on your marketing team, who can do the most complicated analysis.
And help you discover hidden opportunities and roadblocks in your marketing, then come up with a plan of action—and then make most of the marketing assets you need.
The data analysis part is hard to overstate. Most marketers glance at Google Analytics, some other data platform or their email marketing platform data, poke around, look at basic numbers, and call it a day.
But you can do so much more.
Don’t take it from me. Here’s Ross O'Lochlainn, a Conversion Engineer and marketing expert, doing amazing things:
Imagine having to help you calculate LTV, AOV, crunch the numbers on your latest advertising campaign or email campaign, analyze your product line to find the most profitable product, evaluate your content marketing metrics—the list of capabilities just continues.
It can perform most, if not all, marketing and sales analytics you could ever want—and you don’t even need to know what to ask for.
If you aren’t already paying $20 per month for ChatGPT Plus, the Code Interpereter alone is easily worth a lot more than $20.
And it coudn’t be simpler to use.
You log in to your account, go to settings, and enable it:
Once you’ve done that, you’re good to go.
Let me illustrate a few examples on how you can use it:
That’s how I started one recent chat with the Code Interpreter.
It told me this:
And it continues (I’m collapsing the “Show work” part, as it would take up a lot of space—it’s really just showing you what it’s doing with code).
And on it goes:
Great basic analysis.
But what else?
Now we’re getting somewhere.
Sure, let’s do that.
How can we make this more digestible?
Keep in mind:
All this is happening over the course of about ~5 minutes.
Could you do this type of work in 5 minutes?
Let’s look at what other types of analysis could be useful.
I asked again:
The analysis continued with all those six things. I’ll stop sharing screenshots as it would just take up too much space.
But the point remains the same:
You can have in-depth analysis done to your marketing and sales data, even if you don’t know what to ask for.
You just ask the Code Interpreter what you can do—and it’ll do it for you.
It can be as simple as you uploading a spreadsheet and prompt with:
Analyze this spreadsheet on 12 month ad performance. Generate hypotheses. Test them in a sophisticated way. Write up results.
The Code Interpreter will oblige and do this for you.
A few more suggestions on what you can do:
Analyze your landing page for optimization:
Upload a sheet (or file) with landing page data (traffic metrics, conversions, etc.) and prompt:
Analyze the traffic data and landing page data I’m sharing, and identify patterns in user behavior, conversion rates, and bounce rates. Show me the patterns you detect, make graphs when appropriate, and give me a list of recommendations for optimizing the landing page as well as the user journey that happens before and during the page interaction.
Minimize customer churn:
Upload a sheet (or file) with customer data (with sales metrics, etc.) and prompt:
Analyze the customer churn data I’m sharing and identify the main reasons for customer churn, trends, timeline of churn, any behavioral indicators, and propose strategies to improve customer retention.
Find effective social media strategies:
Upload a sheet (or file) with social media campaign performance metrics and prompt:
Compare the performance of these social media campaigns across different platforms and provide insights on which platform generates the highest engagement, the type of engagement, trends in engagement, conversion types and trends, any return on investment if applicable, and provide recommendations for future campaigns that are unique and different but can still help achieve more engagement and conversions.
But it’s not just for analyzing data. You can analyze language, word, and copy with it to.
Upload surveys, documents, text—whatever—and have it perform different sentiment and language analysis.
Can you see how this tool, alone, is a marketer and copywriter’s best friend?
Imagine the kind of value you can unlock for your company or clients.
You could get in-depth data and language analysis done, in minutes.
And you don’t even need to know what to ask for—just ask the Code Interpreter what it thinks you should do with your data.
You can make yourself more valuable and expand on your own capabilities.
That’s all I’ve got for now.
More explorations and tutorials coming.
See you next time,