Whenever I need to lift some weights at the gym I use gloves. Gloves don't make me stronger, but the extra grip helps and the layer of fabric between the bars and my skin makes me endure the pain of exercising for longer.
This is how I prefer to see LLMs and AI at the end of the day: tools that help me improve, so that I can become better in the tasks I'm applying AI to.
Often, when I use AI, I end up not learning as much as I could. While I get things done, I don't accrue the bits of knowledge that make me improve.
This feeling can be validated by a recent study by Nataliya Kosmyna et al., where people tasked to write an essay using ChatGPT displayed weaker brain connectivity in comparison with people only using their brains or brains plus Google.
I'm far from a specialist in the topic of learning, but I can tell you, doing something again and again, breaking through obstacles and having those "eureka" moments are essential parts of the learning process, and when we prompt ChatGPT, get the results back, tweak here and there, and that's it, we skip parts of this process.
Back to the weightlifting example, we end up using the AI as an air wedge. There's nothing wrong with air wedges, but those things are for the super heavy stuff we can't handle by ourselves.
The other day, for instance, I was asked to write a recommendation letter for a friend of mine. Had I just explained to ChatGPT which type of letter, described my friend, set the tone, I would, for sure, have the recommendation letter as output. But what about me getting better at writing?
One can say: I got better at prompting. Sure, that may be true. But is this the skill I want to get better at?
I write letters myself.
Then, I ask the LLM to act as coach, to not rewrite my text, but point out my flaws. I ask it to point out my grammar mistakes, my logic flaws, effectiveness and more.
Here is an example of the prompt I use:
I was asked to write a recommendation letter to my friend Jane Smith.
This recommendation letter is going to be used by for a job offer my friend received from ACME Inc., a SP-500 company where she is going to work as a Senior Software Engineer.
You are a coach with previous experiences as a Hiring Manager. Evaluate the following letter checking grammar, consistency, conciseness and effectiveness.
Do NOT rewrite the letter. Suggest improvements, point me how to improve.
The letter:
```
(...)
```
From that point on, I get the feedback and apply into the text myself. This process reminded me a lot of writing my final dissertation at college, where I went through an iterative process between myself and my supervisor. Each revision, each iteration, made my text better, made me better at writing. It was a tiresome process, but the learning was embedded in it.
Note that LLMs can't help themselves and always sneak a bit of a rewrite here and there, I would then hold the urge to copy and paste whatever it suggested into my letter.
Another benefit is that people on the receiving end of those letters, CVs, and even code, say they are starting to identify LLM generated content. By following those steps, we make sure we also stand out.
There's much potential for growth that we are leaving on the table by not leveraging the new tools that are available for us.
AI is changing the way many people solve problems. It's helping people do things they never knew they could, however, choosing not to learn about topics of our interest is a mistake.
By using it to elevate our knowledge, we grow, and we ensure the output contains the layers of learning we accrued from interacting with the world.