What do we do with ChatGPT?

Here are 5 game changing things we can do with ChatGPT (thread emoji) (hand pointing down emoji)

Just kidding

There’s countless buzzword-filled threads on Twitter detailing how ChatGPT or AutoGPT will change your life, but they’re mostly just trying to get their engagement numbers up. Plus, with all this AI doom and gloom, what do we do now that the world is supposedly ending?

AutoGPT, in really simple terms, wraps a loop around ChatGPT and allows it to iterate on its output. AutoGPT provides the outputs system commands, thoughts, reasoning, and criticism to determine what to do. That’s right, AutoGPT allows ChatGPT to execute shell commands automatically. Thankfully I haven’t heard of any cases where ChatGPT has taken over some datacenter due to an employee putting their production code into ChatGPT (another to-be-written article?).

AutoGPT seems more of a novelty than anything. While some examples seemed to prove that the model was able to function by itself, people reported that the AI would eventually get stuck in a loop. It was a really cool demonstration of how ChatGPT can iterate on itself, but due to the predictive nature of ChatGPT, I believe you can argue that little thinking is actually happening. At some point I’m going to write something on how I think we’ll define thinking in the scope of an AI (this is my newsletter plug, sign up at the bottom of the page!). It could also be that the model itself is not complex enough to handle this type of work.

So ChatGPT can’t think, then what can it do? Hallucinate. The trending AI models (Midjourney, GPT) are really good at hallucinating. They take some input parameters, and then try to settle on a single “answer” (image, text).

I think the above (Plinko) board is a good representation of how AI comes to answers. While we technically don’t know the intricacies of how ChatGPT forms its answers, it’s likely not actually thinking, rather its memorizing and just spitting out what it thinks should come next.

Prompt: I’m writing on article on ChatGPT and its usefulness. I’m diving into the technical limitations of LLM models. Could you tell me if you have a working memory, how much thinking you do, or if you just predict the next set of words? Even if its the last one, is there any “thinking” going on?

… However, it is essential to remember that I don’t have genuine comprehension or awareness. My usefulness lies in generating coherent and contextually relevant text based on the data I’ve been exposed to during training.

ChatGPT

It’s funny asking ChatGPT to describe itself in this way and taking its answer as fact, but I believe what it said is true. Let’s breakdown one possible way its coming up with this text at the end. It must believe that ChatGPT is a language model, and that language models are trained on data, and that training allows it to create similar text, and language models take what they’re trained on and produce similar output, so ChatGPT is able to generate relevant text. And these words are generated in one iteration. ChatGPT is not sitting there iteratively deciding on what its going to say as a whole, it just generates on the fly. I’m starting to steal from what I want to talk about in that “thinking” article (another plug!), but I hope you get the idea.

So, this means they’re good at hallucinating things that are relevant to the data that we provide. And since its not burdened by getting stuck on a thought, since a new instance can be generated instantly, there’s no way for it to suffer from any type of creative block!

AI is really good at helping you get over any type of creative block. You can ask it for help coming up with an idea (or 10!), and if you don’t like it you can ask for more or give it a correction to the prompt. This interactive iterative generation is what sets it apart from all of the idea websites that just list out things or name generators that combine a couple random words.

AI is also really good at interpreting things. Because its trained to predict text based on context, you can give it two contexts and have it merge them! This could be asking for a very specific example of an API you’re trying to implement, or asking it to talk about Augmented Reality in the voice of Jerry Seinfeld. The possibilities are endless.

I mentioned a few upcoming articles above. I’ve got a lot on my mind that I want to get down, so if you enjoyed what you’ve read make sure you enter your information below. I really won’t spam you, I have no interest and no time to do so 🙂