Rethinking AI Collaboration

If you've seen any of my previous posts, you'll know I'm a passionate advocate for AI, particularly large language models (LLMs). LLMs are a type of AI that can understand and generate human-like text based on vast amounts of data they've been trained on. I've integrated this technology extensively into my daily work and it's greatly increased my productivity. When I first started using LLMs like ChatGPT, I sought out YouTube videos and online articles to learn how to create effective prompts. I found tons of information, but one crucial aspect seemed to be missing from almost everything I encountered…LLMs are at their best when I treat them like a human collaborator, not a digital servant.

We've all been conditioned to view technology as a tool that responds to commands. We ask a search engine for information, and we expect a neat list of results. However, if we use an LLM in the same rigid way, we're bound to be disappointed. It's like asking a world-renowned chef to make you a grilled cheese sandwich – they're capable of much more. In my experience, many people who have experimented with tools like ChatGPT, Claude, and Gemini may not yet realize their full potential. They're still stuck in the mindset of using LLMs as a simple question-and-answer tool, rather than engaging in a dynamic, collaborative process. By failing to provide context, ask follow-up questions, and explore ideas together, they're missing out on the true value of these powerful AI models.

LLMs are not simply search engines or basic tools. They're capable of generating nuanced, creative, and even surprising responses that go far beyond the initial prompt. For example, in my work, if I'm engineering a process dependent upon chemical reaction rates, I might simply ask an LLM for a list of relevant citations. While this could provide some useful information, it barely scratches the surface of what an LLM can do.

To me, the LLM should be approached as if it were a human expert.? If I were conversing with a chemistry expert, I wouldn't simply ask for a list of citations and leave it at that. I'd ask probing questions, seek their insights on complex issues, and leveraging their expertise to explore new ideas. That’s how I approach my interactions with LLMs.? By engaging with them as if they were human experts, I've found that I can tap into their deep knowledge and problem-solving capabilities in a much more meaningful way.

To illustrate this point, let's return to the example of chemical reaction rates. Instead of merely asking for citations, I could frame my inquiry with context and open myself up to a broader range of ideas. For instance, if the reason I'm interested in reaction rates is to understand how to optimize an industrial process, I might ask the LLM:

"I'm working on optimizing an industrial process that involves a series of chemical reactions. I need to understand the reaction rates involved in order to identify potential bottlenecks and speed up the overall process. Can you help me explore the factors that influence reaction rates, and suggest strategies for accelerating them in a real-world setting?"

By approaching the LLM as a collaborator and providing context, I invite a deeper, more insightful conversation. Such a prompt results in the following LLM response (in this case, Claude Opus):

"There are several key factors that influence chemical reaction rates, including temperature, concentration of reactants, presence of catalysts, and surface area of reactants. In industrial settings, these factors are often manipulated to optimize reaction rates and improve process efficiency. For example, increasing temperature or concentration can speed up reactions, as can using catalysts or increasing the surface area of reactants through techniques like grinding or using porous materials.

To identify bottlenecks in your specific process, I'd suggest first gathering data on the current reaction rates at each step. This can help pinpoint which reactions are slowest and potentially limiting the overall process. From there, you can experiment with adjusting the factors mentioned above to see if you can accelerate the rate-limiting steps.

Some specific strategies to consider:

• Increasing temperature (within safe limits for your process and materials)
• Increasing concentration of reactants
• Trying different catalysts to lower the activation energy of key reactions
• Finding ways to increase surface area of reactants, like using smaller particle sizes

I'd also recommend running some simulations or small-scale experiments to test the impact of these changes before implementing them in your full-scale process. This can help you optimize your approach and avoid unintended consequences."

While this example is purposely generic, it highlights a dynamic exchange of ideas, where the LLM acts as a knowledgeable partner, helping me explore complex concepts and generate novel solutions. By treating LLMs as collaborative partners rather than mere tools, I've unlocked their full potential and found innovative solutions to the challenges I face in my work.

If you've been disappointed in your first attempts to engage LLMs, I suggest you reframe your interaction. Instead of treating it like a simple question-and-answer tool, approach it as if you were speaking to a brilliant colleague. Challenge it to offer deeper insights and push the boundaries of what you think is possible. If the answers aren't complete enough, don't hesitate to ask for more information or clarification. If you disagree with a point, explain your reasoning and for another perspective.

Many people in the AI community talk about the importance of prompt engineering, crafting the perfect prompt to get the desired response from an LLM. While there's certainly value in understanding how to structure your queries effectively, I've found that the real key to success is much simpler. I treat the LLM as a collaborative partner. I engage in a genuine dialogue, just as I would with a human expert. I ask follow-up questions, explore tangential ideas, and am not afraid to dig deeper.

By embracing this mindset shift and engaging your new digital partner fully, you'll unlock the true potential of LLMs and discover possibilities you never knew existed…just as I have. The power of these tools lies not just in their vast access to knowledge, but in their ability to help us think in new ways and solve problems more creatively. Don't settle for simple interactions.? Dive in, explore, and let curiosity guide you. The productivity you uncover may just surprise you.