My personal journey into the wondrous world of machine learning-generated AIs began when DALL-E 2 was released in mid-2022. The concept is simple: I type in any text, and the AI conjures an image from it. But the ease of use and the quality of the results were immediately absolutely astounding.
–
Felix Schittig
Code
AI
Arbeiten
Spielen
20.3.2023
... or how I learned to love the bomb.
The world of images
A nose cast in bronze as a sculpture in a public space? The last selfie ever snapped on Earth? A colorful stained glass window with a pizza saint? Super Mario on a Time magazine cover? The AI will draw any photo, painting, or for that matter any visual medium you can imagine, with any content, in seconds. Wow - this really is as revolutionary as everyone says.
ChatCPT enters the conversation
But when a few months later the free demo version ChatGPT was released, it was clear: Now the playtime is over. This one is serious - and incredibly useful for me. ChatGPT - a chatbot from OpenAI, based on GPT-3 - is able to coherently respond to my questions - and remembers our conversation. So you can keep refining an answer by asking more questions. And it really knows its stuff: Medicine, law, math, statics, particle physics, love letters, gift ideas, recipes, history, orchid care.
The AI is fed with so much incredible data that it actually knows the answer to everything. Except for current events, because the data only goes up to 2021. Admittedly, you can never be one hundred percent sure whether the answer is just made up or based on fact - ChatGPT would rather fabricate than not answer something. A little basic knowledge in the area concerned is therefore helpful. Also, depending on the question, the answers may serve stereotypes and ingrained thought patterns.
… By the way, I always say "thank you" and "please" in our conversations - that's how my mom taught me, and that's the only way it feels right …
No more fun - ChatGPT likes to code
For me as a UX designer and programmer, ChatGPT is especially interesting because of its knowledge of diverse programming languages. I can describe to the AI in words what a particular function should do, and it then implements that in any programming language. Handy, especially for things I simply can't remember and before ChatGPT had to look up in every project, like RegEx codes or redirects in htaccess. In the meantime it is enough to formulate the desired result and check the finished source code again.
Solutions instead of problems
From the examples shown you can already see: ChatGPT doesn't just slap the results in front of my feet - it explains every step and tells me exactly why and how it did something - or why it would rather have taken a different approach than I suggested. This also works great when I have questions about programming or errors in my source code. A bug with the javascript tool 'nodemon' has been with me in a project for almost a whole year and I just couldn't find a solution - neither in the official description nor on Google or Stackoverflow. When I asked ChatGPT - to challenge it, even deliberately flippant and vague - about it, it immediately spit out the correct answer and explained the context and why. How it could find this answer, which remained hidden from me, I don't know until today.
AI revolution has long since started
These AI tools are undeniably revolutionary, and yet all we see right now is just the very first (publicly available) generation. The AIs that can already be glimpsed on the horizon have a lower probability of error, understand logical relationships even better, and work on the copyright issues that cannot be dismissed, especially with image-generating AIs. After all, the bot feeds its "intelligence" from millions of images that it probably found "somewhere on the Internet" - tracing back on which data a generated image is based is not possible today.
Problems with AI-content or why I am worried
Since photos and videos are still considered proof of something in our minds, but they are already falsifiable, we need to rethink. On social media right now, for example, the AFD member of the German Bundestag Norbert Kleinwächter is creating a mood with AI-generated images. These show, for example, angry non-white people. The comments prove that the images are convincing and seen as real.
Text-generating AIs also bring problems. The sheer volume of text that can now be generated in minutes with just a few instructions is sure to cause some word floods. SEO websites with endlessly boring and keyword-optimized texts, which until today are still painstakingly written "by hand" and cost corresponding money, will be created in the future at the push of a button. On Amazon you can already buy numerous e-books written by ChatGPT. An AI can sound very smart, while it invents or replicates any "facts", the quality of content in general across all media will probably decrease with the increasing use of AIs.
Challenge accepted - The future
So these AIs are neither mature nor safe from abuse - but useful all the same. And most importantly, unstoppable. So in the future, instead of thinking up, structuring and painstakingly typing polished blog articles, will we just list the facts in bullet points, and let an AI do the rest?
And will the reader take the trouble to read such a long article oneself? Or will an AI compile a quick key point list of facts to save time here as well?
I honestly think so.