Home Upgrade Search Memberlist Extras Hacker Tools Award Goals Help Wiki Contact

HF Rulez the UniverseHF Rulez the Universe
nodes
code vending machine
ai artificial intelligence LLM gpt openai bard llama

Knowing When To Use or Not Use AI

Posted Feb 16, 2024 12:30 PM
Knowing when to use AI, or when you should not is important and today you will learn why. You may think that an LLM is some sort of magician, or perhaps a God. "Wrong!" ~ it's just a bunch of lines of code put in a fancy order. It's predictive text at the moment, with some insanely deep contextual understanding & relationships. It can feel like it's the solution for everything, and yet there are times when you're better off without it.

One such example of when to not use AI is actually right now, as I am writing this. You see, I wanted to write about how I started a company with an LLC in Delaware years ago. I thought about writing a blog about negotiating equity. But there was just 1 problem - it's 12:28am right now, I'm limited on time and I can't jump into those topics without spending much longer than I can to think it through. Is that a good time to use AI? Well, I found out that the absence of my personal time doesn't automatically mean AI is a good fit even though it cranked those articles out in seconds.

When I generated those articles just now I realized one major problem. It sounds nothing like me... I even gave it my previous blog article here on Hack Forums & some other content to train it on my writing style. Nope, it still managed to use words I would never use, add more fluff than a stuffed animal, and sound ridiculous. I shifted gears and decided that while AI may be a superhero at client emails, anonymous blog articles, SEO rewrites, etc, it's not great at sounding like me for a more personal community like HF.

On the flipside, I released a tool called MD5HVC here not too long ago. It's a simple program that allows users to recover the values for their own MD5 hashes. While I did write the logic for the value checking logic, it would've been a huge waste of time for me to try to hand-code the entire MD5 algorithm. Yet there was one problem: C++ doesn't have MD5 natively and all the libraries that use it are way too bloated. Was this a good time to use AI? Yes! It pushed out a fully functional MD5 encryption function in less than a minute. It probably saved me hours, if not days or weeks of followup bug hunting.

These two examples showcase two unique uses of AI: Writing a blog for me, and writing a function for me. In the former, the main issue was that it didn't sound like me, and in the latter - no problem at all! This is a good example of how AI can thrive with unopinionated, "boolean" (it works or it doesn't) requests. It's fantastic at code, math, and similar topics for this reason. And yes, it's fantastic at writing too - but not if you want to maintain any sort of personality, pride, or sense of self.

Now, there are plenty of other examples of what AI can be great at or terrible at, but it's mostly subjective and personalized to you. For me, I couldn't justify using the AI written content here on HF for a blog article because it's boring, not me, and anyone can do it. I've always taken pride in writing, written blogs for too many years, and I'll probably continue to.

Yet there is also a middle ground. I can't just say that I love to use AI for "all" code because that's not true either. I don't like using AI for large sets of code as it's much better with single functions or smaller requests. I don't like using it too much as a code generator when I'm learning because then I won't learn as fast - but like with the MD5 hashing there are times where my goal may not be to learn, but to produce (a product/program). This is a grey area, and it can apply to the writing example too. For example if my goal was not to produce 5 original blogs here on HF but instead to produce 500 articles for SEO marketing, then of course AI becomes the perfect fit for that.

As LLM technology / AI models improve we will be tempted to use it for more, more, and more - and eventually everything else too. I think it'll take a personal commitment to not using AI when it does not make sense to do it. Do we want every blog to be generated by ChatGPT or Bard or Llama? What happens after decades of that and nobody (comparatively) has written an original article online in years, or at least not to the degree pre-AI writers are used to? Will that affect society as echochambers echo louder? Will writers suffer from lack of practice? And for coders - will we stop caring who knows how to code and hire AI-coordinators instead? Perhaps! Only time will tell, but one thing I think we should all do is take into account all possible outcomes. A 30-second article I could've copied & pasted into this textarea may have saved me some time tonight (it's 12:39, so it's been exactly 11 minutes so far for me to write this and I'm actually happy to have written it - mission accomplished!) - but I would eventually lose my ability to type, think, and write if I generated it always. The MD5 hashing function from GPT saved me time, but if I did that for all of my code I'd end up with more bugs than a desert, and certainly a lack of personal learning.

If you are interested in this topic feel free to shoot me a message, catch me in /convo.php, or otherwise say hi around the forums. I love to think about the future implications of AI, even though right now it may still feel magical.
Feb 16, 2024 08:48 PM
"This is a good example of how AI can thrive with unopinionated, "boolean" (it works or it doesn't) requests."
Great quote. As soon as AI runs into any "controversial opinion", problems arise.