What is Generative AI?
Generative AI is AI capable of generating text, images, sounds, music, videos, source code, or other types of data. The term is often abbreviated GenAI.
Examples of text generative AI include GPT's and LLM's, such as ChatGPT and Claudi; for image generative AI we can cite Stable Diffusion, Dall-E, and Midjourney.
Generative AI doesn't include AI whose purpose is only to generate classifications. For example, an AI trained to tell whether a photo contains a bird or not is generating a "yes" or "no" answer, but we don't call that a generative AI.
Generally speaking, generative AI are complex autocomplete machines that predict what would be the next word of a document or step of a work starting with the user's input. GPT's predict the next "token" which represents a word, part of a word, or a character of text, while Stable Diffusion predicts a denoised image in a multi-step denoising process.
Generative AI models are trained with datasets processed to guide them into becoming predictors for user input. In general, the generative AI will tend to generate works that match the most common elements found in the dataset it was trained with, since they are fundamentally pattern recognition machines. Patterns that are often repeated in the dataset will be often reproduced by the generative AI model, giving the model a certain consistency and sameness on its generations.
This sameness means that people who have seen AI many times before can recognize something is AI so long as it was generated by someone who doesn't know how to hide the fact they're generating something with AI. Although there are infinite ways to hide it, the average person using generative AI is just too lazy to bother.
Generative AI, in particular LLM's, can not understand English. What these models generate is what they guess to be the most probable continuation of the input they have received. When ChatGPT starts making words appear one by one, that's just not a funny animation to improve UX. That's literally how it works. It guesses the next token, then the token after that, iteratively.
Generative AI can not reason. There are formal ways to formulate reasoning, such as entailment, and while they could theoretically be implemented in some algorithm, generative AI has none of that. It's purely guessing.
If the AI gives you an answer, it's not because the answer is correct, it's because it was the most frequent answer found in its dataset considering the current context with which the AI is working.
Generative AI works like magic. It's hard to fathom how it can do the things it can do, but it isn't doing anything "intelligent" at all. Artificial intelligence is as intelligent as your smartphone is smart.
With that in consideration, generative AI is neither reliable as it can't give you the right answers, nor is it creative as it tends to stick to the most common elements rather than being original. However, that doesn't mean generative AI is without merits.
Generative AI is a great tool for prototyping and brainstorming, as it can dish out lots of random ideas very quickly. For example, you can tell ChatGPT to talk like a pirate, and it will. So if you ever need to consult how write stereotypical pirate speech, and you don't have a stereotypical pirate friend around to consult with, you can just use ChatGPT for that.
However, do not let ChatGPT replace your ability to write, think, or create. As a programmer, whenever I read the word "technology" referring to software, I feel a great sense of unreliability. Twenty years using computers and I still haven't found a single software that can do everything a pencil and a piece of paper can do. In 20 years, ChatGPT will be gone, just like Bard, but pencil and paper will still exist, because you don't need Internet to use pencil and paper, you don't need electricity to charge your pencil, you don't get a sudden update that breaks all your paper, a change in terms of service that locks a pencil feature behind a premium tier, nor will the CEO of paper wake up one day and decide to ruin paper's reputation by getting into scandal after scandal in a childish, juvenile, completely immature manner. You can depend on pencil and paper. You can't depend on software, so you can't depend on AI.