Issue #42: AI research, fused models, and Prompt Caching magic

Good morning.

Remember when I said AI would make the great even better and the mediocre fade into obscurity?

Well, today we're diving into the nuts and bolts of that reality.

In this issue, we're not here to rehash chatbot hype or gawk at the latest image generator.

Instead, we're peeling back the layers to explore the AI advancements that are quietly but fundamentally altering our work as marketers.

I'm talking about AI that acts as your personal research assistant.

I'm talking about fused AI models that becomes a unified whole—and could do the work of about 50 really good marketers.

And I'm talking about prompt caching that simplifies most, if not all, of your tasks as a marketer.

I'm also sharing some tools you can deploy right now to tackle those mind-numbing marketing tasks eating up your time. Because while the future is exciting, we've still got campaigns to run today.

Let’s get into it.

—Sam

IN TODAY’S ISSUE 👨‍🚀 

• The AI Scientist: Your new research sidekick.

• Fuse Chat: The Swiss Army knife of AI models.

• Anthropic's prompt caching: Long prompts, short bills.

• Hermes 3: The customizable chatbot you didn't know you needed.

• AI workhorses: Tools to deploy now for immediate impact.

Let’s dive in.

Automated Insights: How AI is Reshaping Market Research

Research is the necessary fuel of marketing—the rabbit hole that swallows your time and spits out your sanity. 

It’s the foundation of killer campaigns and content that actually converts, but it's also the bane of every marketer's existence. 

But what if AI could be your personal research sherpa?

The AI Scientist promises to automate the entire scientific research process, from ideation to publication. 

While it's primarily geared towards academic use cases, savvy marketers shouldn't overlook its potential. 

At $15 per paper, it's tempting to think we've stumbled upon a gold mine for market research. But let's take a closer look before we get carried away.

This system could be adapted to tackle marketing challenges, from consumer behavior studies to trend analysis. 

Imagine leveraging the same rigor used in academic research to dissect your target market, or provide quality inputs for technical whitepapers. 

However, remember that this tool isn't a plug-and-play solution for marketing insights. It'll require some creative thinking to apply it effectively to our field.

Let's boil this down to the essentials.

The Potential:

• Rapid-fire market research: Picture churning out in-depth studies on consumer behavior or market trends in record time.

• Content arsenal on steroids: Supercharge your data-heavy whitepapers and case studies with AI-powered precision. Churn out high-caliber content for high-tech B2B and enterprise marketing that'll make even the most nitpicky subject matter experts nod in approval. Your competitors will be left wondering how you're cranking out Fortune 500-level insights while they're still struggling with their executive summaries.

• Faster, Cheaper Research: Bypass those eye-watering analyst paywalls and arm yourself with C-suite level intel while your rivals are still piecing together breadcrumbs from free whitepapers.

Sounds great right? 

The Reality Check:

• Quality assurance: AI isn't a replacement for human insight. We'll need to be vigilant about the relevance and accuracy of AI-generated research.

• Ethical considerations: There's potential here to flood the market with subpar or biased research. Using this tech responsibly is on us.

• Context matters: it’s not the first (or only) AI research tool on the market, I can see this being highly applicable to technical content, but there are probably better solutions out there for voice of customer research if that’s your main focus.

My take? This tech is a powerful assistant, not a replacement for human expertise. 

Use it to create empirical sources for technical content, or handle data crunching and initial analysis. But apply your own skills to interpret results and temper those with other transitional academic or analyst sources.

And please, verify, cross-reference, and use your judgment. We're marketers, not automatons.

Matt Berman has an interesting video breakdown on the “intelligence explosion” of AI-assisted research, click the link to see the full video here.

Whatever your stance is on the intelligence explosion or the continual dawn of AGI, this tool has the potential to level the playing field for smaller marketing teams in terms of research capabilities.

Fuse Chat: The Frankenstein's Monster of AI (In a Good Way, Eventually)

When you tell the normies at your family BBQ that you leverage Gen AI for marketing do you usually get that blank stare?

Followed by “I can’t even keep up with you, I’m still figuring out my smartphone.”

Little do they know your job is really just piecing together a hodgepodge of prompts and AI tools, each with its own quirks and capabilities. 

But hey, let them think you’re a tech-genius.

For most of us, leveraging AI isn't nearly as sexy as your luddite relatives make it out to be. 

You're cobbling together cloud solutions to get the job done, and juggling multiple chatbots just to crank out a single piece of content.

This is where a solution like Fuse Chat should make you sit up and take notice.

It represents an approach in AI development that could make the "middle-ground" of AI adoption a lot less tedious—by merging knowledge from multiple large language models without starting from scratch. 

This article from Synced back in January is pretty decent:

Think of it as the AI equivalent of a supergroup—like the Crosby, Stills, Nash, & Young of machine learning.

Here's why you should keep it on your radar:

Cost-effective AI power: Eventually, this could mean getting the best of multiple models without breaking the bank or selling a kidney. It's like buying a greatest hits album instead of every single record. 

Customized AI assistants: Imagine an AI that's great at both creative copywriting and data analysis. It's the marketing equivalent of a Swiss Army knife—versatile, reliable, and impressive at parties. We're not there yet, but soon.

Enhanced customer interaction: Chatbots that actually chat? Revolutionary. We're talking about bots that could handle complex queries, understand context, and maybe even throw in a dad joke or two.

Multilingual marketing: The dream is to go global without learning 20 languages or hiring an army of translators. Your AI assistant smoothly switches between languages faster than you can say "où est la bibliothèque?" 

Instead of companies pouring billions into training new models we might see a shift towards combining existing models in clever ways. 

What does this mean for marketers in the future? 

Adaptability: As new AI models come along, Fuse Chat-like technologies could quickly incorporate their strengths. Your AI toolkit could stay cutting-edge without the constant upheaval and learning curve of yet another new tool. 

Specialization: We might see the rise of hyper-specialized AI assistants. Need an AI that's an expert in both TikTok trends and B2B lead generation? That could become a reality.

The bottom line? 

The concept behind Fuse-Chat highlights the need for more powerful, flexible, and accessible AI tools for the average marketer. 

For now we'll have to stick with our hodgepodge of AI tools – but hey, at least we know a sleeker future could be on the horizon. 

And who knows? 

Maybe one day we'll look back on our current AI workflows the same way we now view flip phones and dial-up internet—with a mix of nostalgia and chuckles of "how did we ever manage?"

Anthropic's Prompt Caching: Making Long Prompts Cheap Again (and Project Mode Magic)

Anthropic just released prompt caching for its Claude models.

Here's the deal: You can now cache your elaborate prompts and refer back to them, slashing costs by up to 90%. It's like having a photographic memory, but for your AI.

Why should you care? Imagine working for a tech company with a product more complex than a Rubik's Cube.

You need targeted messaging for each customer segment, and you need Claude to understand your product's intricacies for every piece of content you ask it to create.

With prompt caching, you can load up all your brand guidelines, product specs, and customer personas once, and Claude starts each chat armed with that knowledge. No more uploading of documents or provide the context again, every time you need a new email subject line.

Right now, you can give this capability a spin in Sonnet 3.5's Project Mode.

Check out what Steve Ranger reported for ITPro:

"Users can now organize their chats with Claude AI in Projects, which allows them to bring chat activities and outputs together and share them across teams... Projects are available on Claude.ai for Pro and Team customers and can be used with its latest large language model, Claude 3.5 Sonnet. Each project includes a 200,000 token context window to allow users to add relevant information, which Anthropic said is the equivalent of a 500-page book."

A 500-page book of context? Now we're talking. 

This isn't just about remembering a single cached prompt. It's about retaining entire conversation histories, documents, and context. 

You can work on complex, multi-step projects without constantly refreshing Claude's memory.

Here's why project mode is great for marketers:

Long-term context: Claude understands and builds upon previous interactions, making each subsequent task more efficient and contextually relevant.

Document retention: Upload key documents once, and Claude can reference them throughout your project. No more repetitive copy-pasting.

Iterative work: Need to revise or expand on something from earlier in the project? Claude remembers the details, saving you time and ensuring consistency.

Task continuity: Start a task, take a break, come back days later, and Claude picks up right where you left off. It's like having a tireless assistant with perfect recall.

Imagine crafting an entire marketing campaign—from initial research to final copy—in a single, coherent project, while Claude keeps everything organized and contextually linked.

Hermes 3: More Flexible AI for Marketers

Let's talk about a “new” LLM on the block: Hermes 3. 

(Previous Hermes versions have been around, but this is a step up).

While it's not as easily accessible as ChatGPT or Claude (yet), it's definitely worth keeping on your radar.

Here's the deal: Hermes 3 is built on Meta's Llama 3.1 405B model, but with some key tweaks that make it particularly interesting for us marketers. 

The big selling point? It's more flexible and customizable than many of the corporate AI offerings we've been working with.

AI trend-watchers like Aakash Kumar Nain are calling it “highly steerable”. 

You can check out his full report here. 

Beyond the tech-pundit analysis, I think Hermes 3 offers some good possibilities for marketers who want to stay with cloud-based LLMs: 

Less Restrictive Responses: Hermes 3 is designed to be more "open" in its responses. This means you can push it further in terms of creativity and role-playing scenarios without hitting as many guardrails.

Improved Reasoning: It's built to handle complex, multi-step tasks better than many alternatives. This could be great for content planning or campaign strategizing.

Customization Potential: With its "unlocked and uncensored open weights," Hermes 3 is more steerable. In plain English? You can potentially tailor it more closely to your brand voice or specific marketing needs.

Agentic Capabilities: This is where things get interesting. Hermes 3 can use XML tags for structured outputs, generate internal monologues for transparent decision-making, and even create visual communications using Mermaid diagrams. It's like having a more autonomous assistant.

Hermes 3 isn't something you can just log into and start using today. It requires some technical know-how to access through Lambda chat with an API. 

But don't let that scare you off. As with any cloud solution, it's likely to become no or low-code in time. 

For those of you who've been frustrated by the limitations of current AI models—maybe you've bumped up against content restrictions or struggled to get the tone just right—Hermes 3 offers a glimpse of a more flexible future.

While Hermes 3 is exciting, don't feel like you need to jump ship from your current AI solutions if they're working for you. 

Instead, keep an eye on how this technology develops and maybe give their free beta a spin once it goes mainstream. 

The point is, as less restrictive LLMs become more accessible, it opens up strong capabilities for personalization, content creation, and campaign planning. 

In the meantime, focus on mastering the AI tools at your disposal. The skills you develop now will serve you well as more advanced models like Hermes 3 become mainstream.

As I’ve been encouraging you to do for years, focus on developing your: 

• Curation (Selection)

• Perspective (Point of View)

• Discernment (Situational/Contextual Awareness)

• Judgment (Decision-Making)

• Tastemaker (Sense of Taste)

• Articulation (Expression)

• Wisdom (Right Thing, Right Time)

I know, that sounds esoteric but I’m telling you, when every marketer on the earth has access to the same AI tools, and can do the same things as you—you need to aim for deeper work than clicking buttons and chatting up ChatGPT.

The past year I’m finally seeing others repeat the talking points above. They’re missing a few things but that’s okay. I’m just glad people are repeating them.

And I have a different take on those skills above than most, which I’ll soon share.

Tools for Your Stack Today

Here are interesting AI tools that you can deploy today to tackle those mind-numbing marketing tasks that eat up your precious time. No PhD required, just a willingness to work smarter, not harder.

Let's dive into the nitty-gritty of AI tools that'll give you an edge while your competitors are still fumbling with spreadsheets and generic chatbots:

MECO, secret weapon for trend-spotting

This newsletter aggregator is like Marie Kondo for content consumption, but instead of sparking joy, it sparks ideas. 

Smart filters, personalized recommendations, and built-in note-taking mean you'll never miss a trend or forget that brilliant campaign idea again. 

Gamma: Making slide decks EASY

Most of us would rather watch paint dry than create another PowerPoint deck.

Enter Gamma, the AI-powered presentation tool that's here to save us from death by bullet point.

This bad boy uses AI to whip up presentations, documents, and websites faster than you can say "synergy." 

Whether you're an entrepreneur pitching to investors or a creative trying to keep the client awake, Gamma could be your secret weapon in the war against snooze-worthy presentations.

Stop Eating Your Keywords

Search Engine Journal recently released a practical workflow using OpenAI's text embeddings to identify keyword cannibalization. 

This approach helps SEOs efficiently detect when their content is competing against itself in search results.

It's part of a new series focused on integrating Large Language Models (LLMs) into SEO processes to scale tasks and improve efficiency. 

These articles cover different embedding models and their applications, providing a solid foundation for SEOs looking to leverage AI in their optimization strategies.

If you're serious about staying ahead in SEO, this workflow is worth investigating for its potential to streamline your content analys

Good prompting is like practicing Applied Intelligence: 

• Critical Thinking

• Abstract Thinking

• Concrete Thinking

• Divergent Thinking

• Convergent Thinking

• Holistic Thinking

• Sequential Thinking

• Scenario Thinking

• Intuitive Thinking

• Reflective Thinking

• Metacognitive Thinking

• Systems Thinking

• Pattern Thinking

• Analytical Thinking

• Creative Thinking

There’s a lot more but not enough time and space to cover this here.

But things like this, cognitive architecture, and so on—these are shifts in perception that have helped me use AI for my own business, projects, and clients—long before ChatGPT became a thing. 

You have to understand: 

I’m the dumbest person in AI. I don’t have a degree in anything. I’m not an ML researcher or scientist. 

I’m somewhere between AI academia and drawing with sticks in the sand. People either look at me and think it’s adorable—or they’re looking for nice doctors to take me away, ha ha, hee hee. 

My only advantage? 

Somehow I can make (some!) sense of AI—as it is, and where it’s going. 

Somehow I can apply what I’m seeing into real-world projects, problems, and so on—and it works!

If you want to learn some of what I do (I can’t share it all, not enough time), then you’ll probably be interested in my upcoming workshop on prompts.

I don’t have a name for it yet. But it’ll be in-depth, not focused on making AI into a cash machine (I’ll cover that later), and expensive. 

Until then: 

The key is to approach these tools with equal parts enthusiasm and thoughtfulness. Don't just jump on the bandwagon because it's new and shiny. 

Ask yourself: How can this genuinely make my work better? How can it free up my time for the tasks that really need that human touch?

Until next time, keep your marketing bionic and your thinking human.

Talk soon,
Sam Woods
The Editor