I don't know enough to evaluate claims of AI coding abilities, but here's something I'd naively say is 50/50, for an AI coder getting right on the first try:
Prompt:
Your task is to create an iOS utility app, with multiple requirements and intermediate/ancillary work-products.
The purpose of the app is to keep the device at as high a state of charge as possible, while plugged in, without damaging the battery, at all intensities of device activity/usage. The app should automatically activate, when the device is connected to a power source; implement whatever the optimal charging strategy is, for the applicable conditions; and automatically deactivate, when the device is disconnected from a power source. It should be both self-contained and as “lightweight”/”streamlined” as possible.
Research how charging conditions affect battery longevity, derive an optimized charging strategy, and write a report on your findings, strategy, and reason that explains everything without requiring more prior knowledge than a typical Anglophone 15 year old possesses.
Write a pseudo-code algorithm implementing your charging strategy and explain how your pseudo-code algorithm implements the more abstracted strategy.
Research Apple’s iOS app technical requirements and App Review criteria. Read [human evaluator selected best practices]. Code the app in a manner consistent with all of the above, documenting your code (both individual lines and larger architecture) at the “rubber duck” level.
Output both your code and a report explaining how your code meets the above requirements.
Be prepared to output any other work-product necessary or beneficial for the App Store submission process and explain how to use them.
Hopefully that's a Goldilocks level of detail and structure - I tried to describe the requirements in such a way that the prompt requires an LLM to make inferences that would be easy for a human, but doesn't leave so many degrees of freedom that it could output a "technically correct" product that falls completely short of what the prompt intended (similarly, balancing structuring a legible process for it to follow vs hand-holding its process). I don't know if we have any Swift programmers to evaluate the code, itself, but if the intermediate/ancillary work-product passes muster with everyone here and the code passes muster with the App Store (I would actually like this to become a FOSS app!), that would be very good demonstration of prompt-engineering-as-programming.
p
I don't know enough to evaluate claims of AI coding abilities, but here's something I'd naively say is 50/50, for an AI coder getting right on the first try:
Hopefully that's a Goldilocks level of detail and structure - I tried to describe the requirements in such a way that the prompt requires an LLM to make inferences that would be easy for a human, but doesn't leave so many degrees of freedom that it could output a "technically correct" product that falls completely short of what the prompt intended (similarly, balancing structuring a legible process for it to follow vs hand-holding its process). I don't know if we have any Swift programmers to evaluate the code, itself, but if the intermediate/ancillary work-product passes muster with everyone here and the code passes muster with the App Store (I would actually like this to become a FOSS app!), that would be very good demonstration of prompt-engineering-as-programming.
More options
Context Copy link