ChatGPT was introduced to me by a junior at university as an AI that can write code.
Hearing that, the first programming language I had it generate was a shell script.
You can program in natural language
→ You can control a computer with natural language
→ Computers can do anything
→ You can do anything with natural language!!!
I quickly built an app where I give instructions in natural language, the AI generates a shell script, and the script is executed as-is via subprocess.
“Create an HTML file that displays ‘Hello World’ and open it in the browser.”
After a short while, the browser launched and “Hello World” appeared.
“I just controlled a computer using nothing but natural language...!!”
Following CUI/GUI, I named this DUI (Dialog User Interface).
Back then, the GPT-3 API was so unreliable that it even failed at simple shell scripts, but apparently there is something called the scaling law, and this AI is expected to keep getting smarter and smarter.
Humans think in language, so this AI should be capable of logical reasoning.
Thinking usually takes a lot of time and money, but that will probably stop being an issue.
After all, GPT-3.5 Turbo is smarter than GPT-3, yet cheaper and faster.
Whenever GPT-10 is born someday, it will surely be far more intelligent and cheaper than today.
In that moment, I had a clear gut feeling that an incredible era was coming.