Unlocking the Black Box of AI: What Content LLMs Crave

In this issue:

• LLMs Want Answers: Why Your FAQs Matter More Than Ever
• Write better newsletters with Google Gemini Gems
• Reformatting data with ChatGPT (tip)
• Epic Content Marketing for Higher Education released
• Executives need to use AI, too
• Design quote of the week

At this point, we understand that LLMs are basically black boxes and we don't know what happens inside. According to MIT, nobody knows. The question is, how do we make sure that these AI systems are generating the responses that include us?

Rob Davis , the Chief Innovation Officer at MSL, was recently on The Trending Communicator with Daniel Nestle , and they discussed this quandary ... as well as some potential solutions.

If you think about what generative is built to do – provide answers to questions – then think about what information is wants to ingest in order to be able to generate those answers.

LLMs are looking for conversational, question-and-answer-type content. FAQs are a great way to tell LLMs "this is the question" and "this is the answer."

Rob and Dan discussed an issue with a major produce company, and they noted that Perplexity's answers were based on this Q-and-A content rather than recent content. The system was putting more weight on conversational content than on the recency of the content.

This is different than what we've been taught with SEO.

The difference makes sense when you boil down the big difference between finding information in a genAI system and a search engine: A search engine will give you millions of potential answers and genAI will give you THE answer.

NEWS AT THE INTERSECTION OF AI AND DESIGN

?? Evaluating Google Gemini “Gems” for Newsletter Editing

Rolf Mistelbacher tested Google Gemini’s new “Gems” feature for AI-assisted newsletter editing, providing insight into its strengths and limitations compared to other AI tools like ChatGPT and Claude.

? Custom Power-Ups: “Gems” allows users to create custom instructions, which the user used to set up a process for refining newsletter drafts. Although helpful, the feature is mostly a time-saving mechanism for creating reusable prompts rather than a groundbreaking innovation.

? Effective, But Needs Fine-Tuning: The AI-generated content was mostly usable (80% retained), but required some editing to align with the user’s personal style. However, the feature still lacks the depth and flexibility found in ChatGPT’s custom GPTs and Claude’s project tools.

Keep reading

?? Reformatting data with ChatGPT

Speaking of trying new things, I use Bitly for the LinkedIn version of this email so I can track clicks (I want to see what info you find most valuable and give you more of that). Unfortunately, I use the free plan, so I can't download my data to dig into it.

I found that manually copying the information from my analytics page and dropping it into Excel is a mess. Nothing is formatted properly.

So I take that spreadsheet and upload it to ChatGPT with the following prompt ...

“Please reformat this spreadsheet so that each data group is organized into rows across multiple columns, keeping similar content (e.g., dates, engagements, links) in their own columns. In the ‘Engagements’ column, extract only the number and remove the word ‘engagements.’ Also, remove the ‘Tags’ and ‘Actions’ columns. Please export to a csv.”

...and it gives me a csv that I can then use to slice and dice the data. Let me know if you find this helpful.

?????? Epic Content Marketing for Higher Education

Brian Piper just released his latest book, Epic Content Marketing for Higher Education.

Last year, Brian co-wrote the updated version of Epic Content Marketing with Joe Pulizzi. This year, Brian is taking that one step further and focusing on his area of expertise, higher ed.

I haven't received the book yet, but I look forward to reading it and seeing what I can learn. I picked up some good tips from last year's version.

Order now

NEW RESOURCES FOR YOU

Are Your Executives Using AI?

After talking to many firms, I have come to believe a key factor in successful AI adoption is whether the executive team actually experiments with AI to try to get work done themselves. Those who do tend to feel urgency and push for transformation.

Most c-level folks (the vast majority of executives I talk to) still haven’t even tried LLMs. Changing that should be the highest priority of anyone who wants their company to succeed at AI.

As we discuss ad nauseam, AI is transforming the ground we stand on. There are people in every organization that are working to adapt their jobs and businesses. But if there isn't interest at the top, then businesses won't change.

Ignoring AI, or worse, fighting AI, is like fighting or ignoring gravity. As the saying goes, You may hate gravity, but gravity doesn't care.

AI is the same thing. It's coming to change your business. If the people at the top of your organization aren't exploring independently, it's not a good sign for the business's health.

Design Quote of the Week

“It’s not about knowing all the gimmicks and photo tricks. If you haven’t got the eye, no program will give it to you.”

– David Carson

My AI disclaimer: I write the main stories by hand (with occasional assistance from Grammarly). The quick news hits are articles I've found, which I run through a ChatGPT prompt to generate the synopsis and bullet points. However, I write the My Thoughts bullet. If AI generates the images, I include the prompt so you can see how I got to that image. Nothing in this newsletter, or anything I publish anywhere, reflects the views of my employer.

Next week, we'll dive into the news from this week's MAICON!

HOW CAN I HELP YOU?

If you want to learn more about what's available, here are some links:

• Coaching
• Courses
• Articles
• What’s After Design

Thanks for reading!

-Jim