#30 - Greetings from Stresa 🇮🇹 & The Art of Asking

Is How We Communicate with LLMs Today the Right Way?

#beyondAI

I know I should relax and enjoy my vacation as promised. But I do feel relaxed in this moment. A weird sense of relaxation, not yet consciously perceived. But here I am, thinking and writing in a wonderful place in Stresa, and it feels good.

So, while I’m here, I was reading some papers about prompting… yeah, being in Stresa and thinking about prompting is everything but romantic 😂. Don’t judge me.

It’s funny how even in a setting like this—surrounded by the peaceful beauty of Lake Maggiore and the stunning view of Isola Bella—I find myself contemplating the intricacies of large language models and how we’re only beginning to understand their true potential. I mean, take a look at this view.

It’s serene, isn’t it?

The lush greenery, the tranquil waters, and the majestic Alps looming in the background.

You’d think such a scene would pull me entirely into a state of complete relaxation. And it does!

And yet, in the middle of this peaceful landscape, my mind still sometimes fully engages in exploring the complexities of AI.

Weird, but not unwelcome. It’s also inspiring.

The Art of Prompting: More Than Just Asking Questions

The setting here in Stresa is like the perfect metaphor for the relationship between humans and LLMs.

No, I’m kidding. It’s not. 😂 But here’s what I’m thinking:

The more I dive into prompting techniques, the more I realize that the limits of LLMs are largely our own. Sure, they’ve been trained on massive datasets, and they can produce incredibly insightful outputs.

But the prompts we give them? Often too simplistic.

When we get more sophisticated with how we prompt, we start to see just how much more these models are capable of.

Think about how detailed and layered the gardens on Isola Bella are. Much like those gardens, prompting can be complex and multi-faceted, guiding LLMs to explore deeper layers of reasoning.

What a ridiculous comparison 😅 Anyway.

One of the techniques that stood out to me during my vacation reading is chain-of-thought (CoT) prompting. This method essentially walks the model through complex reasoning tasks in a step-by-step manner, helping it arrive at much more accurate conclusions.

CoT prompting adds structure to the AI’s problem-solving process. It breaks down intricate problems into manageable steps and guides the LLM through each stage, allowing it to “think out loud” and come to the right conclusion.

Why Prompting Is Like Lake Maggiore 😉

The landscape of Lake Maggiore and the historic charm of Isola Bella remind me that there’s often more to something than meets the eye. LLMs are much the same.

It’s fun trying to be poetic 🤩🤣 Is it because I’m in 🇮🇹?

We tend to view them as powerful but perhaps limited tools, but that’s because we’re still learning how to engage with them. The beauty of the scene in front of me mirrors the untapped potential of these models.

There’s a lot happening beneath the surface that we’re not yet able to fully appreciate.

One prompting technique I found particularly fascinating is self-consistency prompting, which works by generating multiple reasoning paths and selecting the most consistent one.

It's kind of like hiking up a mountain path, where there are several routes to the same summit, but some are smoother and more reliable. By exploring multiple routes, the model increases its chances of getting the right answer.

This parallels the layered scenery I’m looking at. There’s the lake, the trees, the beautifully structured palace gardens, and beyond that, the mountain range. Each layer contributes to the overall beauty of the place, just as each step in self-consistency prompting adds depth to the model’s reasoning process.

Omg. This is an even more ridiculous comparison than the one before. I should stop. ✋ But it’s fun, I can’t help it 🙃

We Haven’t Reached the Limit Yet

The more I read about prompting, the more it becomes clear that we’re only scratching the surface of what LLMs can do. There’s still so much we don’t know. So many undiscovered ways to tap into their potential.

It’s like taking in the view of Isola Bella from multiple angles: there’s always something new to appreciate, another layer of beauty to uncover. The more we experiment with prompting techniques, the more we realize how much more there is to discover about LLM capabilities.

The Future of AI Interaction Is Here

LLMs have layers of potential waiting to be unlocked. How we interact with them—how we prompt them—will determine just how far we can go.

We’ve only just begun to tap into the true potential of LLMs, and there’s so much more to discover. But I also believe we may never fully understand whether we’ve explored all possible, yet undiscovered, prompting techniques.

Why? Because prompting, like language itself, is inherently open-ended. Just as there are countless ways to express a single thought in natural language, there are an infinite number of ways we might guide an AI to explore and process information. Every new prompt reveals another facet of the model’s capabilities, but there will always be more unexplored paths, more subtle nuances in how we can structure questions or provide context. The dynamic nature of human language, combined with the complexity of AI, means that the possibilities are endless—there’s always another layer of sophistication we haven’t yet reached.

If you’re as fascinated by this as I am, I encourage you to explore the world of prompting further. There’s a whole landscape of possibilities waiting to be unlocked—just like the view I’m enjoying here in Stresa.

That was my last attempt at being poetic and justifying my AI outbreak during vacation, I promise. 😉

Thanks for reading.

And don’t worry. I’ll get back to enjoying my vacation…

JBK 🕊️

---

P.S. Let me know if you’ve tried any advanced prompting techniques! I’d love to hear your thoughts on how we can unlock even more of AI’s potential.

---