@Bart_Mol Unpopular opinion but I dislike how labs train models to oneshot things because it implies direction and opinion to fill the gaps. It's great for marketing, horrible for any professional usage.
Half my work now is removing clutter I never asked for.
@synthwavedd I think it's a psychological thing that we just start to get used to something and then notice its shortcoming. Goes from 'new - shiny' to 'used to it - meh' really quick
@aisalesguy@synthwavedd I had the exact same thing in mind. Taste might be great if you want it to create a quick prototype but when I work in established systems I don't want my clanker to bring in his taste to go against the system that was designed to do and behave a certain way
I am feeling the same right now as I did when I started diving into music production and exploring sample packs. You start to hear it everywhere
It's no different now with Agents. Everyone who uses AI a lot β will start to notice those Β· patterns; everywhere they show upπ
@jsedlacekjr@jazzychad Assuming Anthropic does the same as OpenAI, the model now asks questions even without plan mode, so the gaps that plan mode filled are now working natively
I went to school with, and have worked with hundreds of different devs in my 15 years of doing this. I could not recall a single developer who was super arrogant about his own skills. I think it's at the core of an engineers mindset to be curious and try to improve. We all fuck up every single day, then we research, we learn, and we improve. And everyone who has niche skillset understands how long it took to get there, and that generally also makes them understand how long other people took to get where they are. It feels really alienating to read this.
I got more and more annoyed by agents messing up UI so I did what we all do when agents annoy us.. Writing guardrails and validation. But for that I needed a system with rules and guidance to even understand what to validate against and what errors to throw.
So I installed storybook, extracted my whole presentation layer into it and started going use case.. by use case.. layouts/patterns/components/primitives covering all possible states something could have including loading from or away from it.
I'm not gonna lie, I really have new found respect for everyone doing UI/UX design. It's such an underrated skill because we see and use things that just feel simple, but they do because someone has done the work to think about every little detail, every interaction, every state and translate it into guidelines on what, how and where to apply what for which use case so it CAN feel simple!
Also it makes working with agents a lot easier.. When clankers try to write their own components because they haven't had the design system loaded into their context. Well now they do because validation will fail for not importing presentation from the design system!
I'm far from finished. There's way too many details to define and understand, and even while defining them I'm constantly thinking about whether or not to abstract something further.
A good design system is really not much different from architecting software. So many decisions π€―
@thsottiaux Voice has been such an integral part of my workflow it's actually crazy cause I've never talked to my computer before. Now I'm getting tired just at the thought of having to type things out π
The amount of cognitive load caused by interrupting someones daily workflow is so true
There's so many things I'd like to change in my software that make it not feel like every other app, but then I'm like.. standards exist for a reason and this is what people are used to π So it really is a balancing act of taking established principles and tweaking them just a little bit at a time
It also reminds me of every time theres a redesign everyone complains, but after a few weeks no one talks about it anymore because everyone has already gotten used to it!
In the obsession with AI, too many founders, leaders, managers seem to have forgotten that an engineer with motivation + drive + focus + AI >> hundreds of AI agents without any of those
The best products and companies are built by people and teams like that, and will continue to be so
Ask yourself, again, why the AI labs hire the best of the best (some of the most driven people in the industry; founders, CTOs etc) if people aren't making the differrence when everyone has access to similar tools already
It's like swimming against the current. I have all sorts of guidance and even workflows to clean up tests and frontend clutter and while they understand it as I prompt and ask about it, they still always fall back to the same behavior.
It's as if you don't supply examples where it can base wrongdoings of, it doesn't notice wrongdoings. And it seems really difficult to supply examples without having a human in the loop.
The only way right now for you own codebase is to establish a system and deterministic guardrails to catch it where you can, it's not always possible though.
@aiivakhuman@animeupdates@grok I remember being in the same spot during Whole cake arc. You'll eventually convert to reading the manga, whether you like it or not π
@jerometoby3@nikitabier I remember threads on black hat forums where people did this manually, making hundreds of mails to vendors. Then got free shit from them until the corpos caught up. Was like 10+ years ago. Larger corporations are equipped to deal with this, just maybe not at scale π
@GergelyOrosz Feels like now that we have AI and can do so much with it we notice more and more things to do, and everything that needs to be done still requires someone to understand and maintain it π
@GergelyOrosz Isn't Apple THE consumer supplier right now if you want to run your local models?
Or are you saying Apple is missing out on providing their on models on their notebooks?
@ImLunaHey I just remembered when the whole switch from master to main was happening I'd read OSS issue trackers because introducing this discourse to open source maintainers was just hilarious to read