Профиль

Didn't expect my grousing this morning to blow up as much as it did. I mean, it's not @socio-steve.bsky.social numbers, but it kept my phone notifications busy today.

... if Apple didn't force upgrades through Xcode* updates. * Hardly a sterling example of code quality itself, but its issues predate the LLM era.

As someone who has been an AVID MacOS user and developer for the past 25 years (going back to 10.1 on a Titanium PowerBook), the last few MacOS releases are prime examples of the degradation of the platform. I'd go back to releases 5 years ago (or earlier - Snow Leopard was the sweet spot)...

I don't have to worry about my work being declared in the public domain (even though most of it is with the Apache licenses I use).

I'm not the one doing the prompting. I'm getting the outputs from others' prompts and having to refactor them. My understanding is that the feedback I've provided has been incorporated into their system prompts. I don't use AI to write code myself. It's in writing code myself that I think about it.

That kind of abstract thinking doesn't seem like something that would come naturally to LLMs given their underlying next-token prediction fundamental mechanics.

... The LLM is happy to add one-off watchdogs around places that need them, but doesn't seem capable of recognizing that this might be a common feature worth building some actual new infrastructure around, and suggesting the changes needed to make clients compatible with the newly-standardized API.

The benchmark I'm waiting for it to pass is to be able to recognize when different parts of a system are trying to do something similar and it makes sense to abstract out some infrastructure that results in net-less code. Today's example was a watchdog timer for a Chrome extension...

The Framework Laptop 13 Pro and the Steam Machine are canaries for a society wide problem which feels comically unfair because they both would be two of the most interesting pieces of hardware in years at their original price points.

This is one area where I'd be happy to be proven wrong, in terms of more robust software that's well constructed, efficient, and usable. What I fear is that our standards will simply drop to accommodate the new reality in the same way I have a hard time finding good shoes that last.

Stopped taking meeting notes due to AI notetakers only to be reminded almost immediately again that writing is thinking

I think Oracle is the canary in the coal mine. Keep an eye on them.

I think the best analogy I've seen is the one comparing the changes to software to the changes we saw in clothing with the "fast fashion" movement.

Yeah, at some level completely avoiding LLM code is impossible. I'm thinking of all of my Node and Python dependencies, in addition to the "regular" apps I'm using. That said, I'm also seeing code quality decrease across the board (as an end-user), and wouldn't be stunned if LLMs are partly why.

I'm quite excited to receive the X-Men series!

I'm glad it worked out for you and I hope you reviewed that code carefully, since it's connected to a heater element.

I use all of the above. I should have been clearer with respect to "touch". What I meant by that is code that I have to integrate into systems that I'm responsible for and maintain with a time horizon of at least 5 years into the future.

I'm largely talking about adding new functionality to an existing system. It has no sense of architecture and seems to be brute-forcing a bunch of quick fixes on top of each other.

Let me know when you figure something out. Lots of people have been thinking on that exact same problem in all of the decades I've been writing software.

💯 And as someone *responsible* for deciding if the floor's been met, AI is making my job harder in the code it generates. It's like reading a book where the author switches between drastically different writing styles mid-paragraph. It's difficult to get into that state of reading Flow.