Curiously I just “asked” ChaptGPT to write me a Python script to generate random numbers and to include an infinite loop in the code. You would think it could surely generate the correct response to that? After all some people think that AI will replace Developers.

But strangely it didn’t reply with “what upper and lower limits would you like?” or even “would you like integers or decimals Ed? If the later how many places?” I would expect even a lightweight Developer to come back with those responses.

A more experienced Developer might have replied by asking how random or repetitious could the numbers be. But not ChatGPT.

But even the least experienced Developer, I’m talking just finished a Comp. Sci. Degree and so wet behind the ears they actually drip when they walk… Would have replied with “why an infinite loop and are you REALLY sure about that???”. But again not our new friend ChatGPT.

In fact the reply that ChatGPT did come up with was conceivably the only wrong reply possible. It wrote code that produced (some level) of random number in an infinite loop. Oh dear.

I hear people talk about so called “prompt engineering” in fact they have missed the point (nearly as badly as ChatGPT did in its interview to be my newest Developer). What we learn from the above is that the current level of “AI” is in fact not “intelligent” at all and is wholly dependant on humans asking perfect QUESTIONS (not prompts). Apparently the Intelligent Human has to know the subject matter perfectly, the scope and context of what they want to know and deal with any nuisance or implicit knowledge required. They also need to know a whole lot about how the “AI” model was built, but that is for another post. Yes we need Humans that ask perfect questions… That’s going to go well then…

Great.

But don’t worry, “AI” is “amazing” at generating pictures of dancing elephants in pink tutu’s. And for some Humans that appears to be enough.

#ai #software #developers #strategy #getagrip