Although the icon suggests it, shipping an App Clip isn’t as easy as clipping a coupon. Imagine trying to fit three toys from a toddler into a shoebox - it’s more like that. Here’s a checklist along with some tips and tricks to help navigate the journey. https://t.co/KHoCc1qDSj
@BowTiedTrance “And the moment we ask whether AI is ahead of us, we have already accepted that we are measuring the same thing.”
They’ve got us thinking past the sale.
@rolypolyistaken Excellent post. Thank you for putting this into words.
First time I recognized the sticky confirmation bias trap was with “Republicans will be hunted”. It was fascinating watching him work.
@BowTiedTrance I read the whole second paragraph before reading the second part of your post - only because you were highlighting it.
Took me about 6 tries and I still don’t understand it. I think you nailed the 97% figure.
@Nicolascole77 I’m having a hard time understanding this from first principles. Did you get the concept of open/closed loops from a particular field of study?
#Android engineers, rejoice! We have an incredibly helpful how-to video - on customizing your API docs with Dokka. In this video you'll find examples, code snippets, etc. Improved API documentation is something we can all get excited about! https://t.co/K6wgmso3cj
I’ve thought about this a lot, and at the moment I disagree. I can’t tell if it’s only because I have a financial interest in not having my job disappear. Here’s why I disagree:
When a new feature is added to a programming language (ex. Swift concurrency), how will AI be trained how to use the APIs if programmers don’t write the training code?
Ok, new language features, like concurrency, are irrelevant because that feature is for humans.
Would we start creating programming languages that only AI’s understand? Will AI programs contain hallucinations or be completely flawless? If they’re not flawless, who will debug the program that’s written in a language that only AI understands?
Ok, debugging is useless because you just regenerate the entire program from scratch every time.
Will you not discover the AI has chosen different tradeoffs that you find unacceptable in each new generation of source code? Size of the program, speed, maximizing CPU efficiency, power consumption, etc.
How many tokens would be required to create/maintain the Facebook website/mobile app? There must be an upper bound on how many tokens a future AI will be able to handle. Will it be enough to maintain a large codebase?
What happens after the initial launch of a startup that decides to pivot, but wants to maintain 1/4 of the working features? Will AI be able to reliably regenerate a whole new program while maintaining the 1/4 that was working flawlessly?
I’m open to the possibility that we can answer all these questions, but at the moment, I’m skeptical.
By the way, the folks who made copilot have explicitly said there’s a reason it’s called copilot and not “pilot”.
I think the most likely future is one where AI is a tool used by programmers to be more efficient.