The recent introduction of LLM (Large Language Model) — based tools, creates a lot of hype and excitement. Writing a sentence and creating almost anything suddenly seems possible — without ever gathering the know-how, practice, and resources needed not long ago. The result: 100+ million users and domination of the word “AI” everywhere seems unstoppable — you either join or are left behind.
Sure, I have joined the ranks and started using ChatGPT and its siblings to alleviate some of my writing tasks. However, a basic fact about the tool and the basic principle of its operation needs to be addressed in public discussions. The old, hated command line is back!
Yes, we are back to DOS after decades of working hard to create GUIs and eliminate the blinking cursor. Sure, DevOps and programmers continue to like the command line offering them quick access to a zillion of operations — but these guys have strange communication skills with computers anyhow — see the strange languages and hundreds of syntax rules they know by heart. We, all the rest, got accustomed to GUIs, which have dominated human-computer interactions for decades. Suddenly command line is back. “Prompt Engineer” — haven’t we heard this term when looking for guys capable of spitting long text “commands” that Windows and Linux could understand?
Current prompts are a bit more friendly, but if you track YouTubers and articles about “prompt engineering,” the complexity increases daily, and the number of “tricks and secrets” would not leave any Linux expert unsatisfied.
So what happened? How did we regress to the point that we thought we had overcome yet still feel enthusiastic about it? One simple reason is that LLMs function this way — they were trained with text input, and they can only output another set of text lines. Sure, some smart guys created “forms” that can be used to assemble these prompts, but it is still a text-in-text-out process.
Wait, what about Generative AI for Images, Video?
Well, this is even worse. A domain in which brush, mouse, and digital pen are all you need to create a masterpiece suddenly uses a sentence to create the future of aesthetics. The connection between our hands, eyes, and minds is put aside, and the command line reigns again. Well, not so fast: The impressive huge compilation of aesthetics created over time, converted into elements that can be combined using sophisticated statistical methods, makes astonishing output, but still a replication of previous generation achievements.
Is it useful? Certainly yes. Only some pieces of graphics should be a work of art, original and never seen before. Commercial graphics and certain types of video are used to color our digital communications. Generative AI is as legitimate and relevant as the well-known and acceptable image libraries - just a bit faster and sometimes more accurate. So using the command line to create such graphics and even video is valid and, to some, the only practical method. Others who enjoy filling the pressure of a pen and its motion and those in love with the touch of a camera will continue to use these forms and create genuinely new and unique forms.
So Are You One of the Nay Sayers?
Not at all, but amused by the sudden adoption of the command line after years of condemning it (rightfully) as the most nonintuitive, boring tool possible.
I greatly support synthesis: Combining tools to get the best possible result. One area that is doing that is the No-Code app development. Developers of these tools have recently added to the visual design environment AI (read: “Command Line…”) that one can use to automatically create specific combinations /parts of the system. It can accelerate the creation of portions of an app's GUIs and logic elements and then use the standard tools to complete the mission. So you can still place elements and select colors and shapes using the mouse, but certain highly structured parts can be defined in a short line of text.
Another excellent example is Adobe’s recent enhancements to Photoshop, in which you combine graphical selection with a command line telling the AI how to combine two processes (i.e., transparent and complement a selected area). This is definitely helping, and while Generative AI is the exciting new kid on the block, its addition does not seem to me that different from previous generations of Photoshop’s unique filters and algorithms. Just a bit more complex and statistical.
I hope this will be the dominant form of development so that we can enjoy the merits of both worlds.