Michal BryxíMaybe making a thread of objectively good uses of AI can be useful for some future reference?
🧵
Friend who is technical, but has no background in coding needed to make a code for PLC that would forcefully open a door and let them automatically close.
Hours of learning this programming language to do this one-off thing vs having a dialogue with #ChatGPT and get a working programme (on second go).
He's happy, because he learned that he is able to do those things, his boss is happy because project, clients are happy.
A friend of mine is mandatory on many meetings that don't event touch his area of expertise. He uses AI to transcribe the meetings, other AI to identify if there was any chance he was tasked to do something and turn it into Jira tickets.
Enormous amounts of employee time not burned. And huge peace of mind saved.
Every time some profoundly doof website makes the decision that "copying our content is bad" and mangle with the text selection feature, I'm forced to make a screenshot and extract the text somehow. For most cases #OSX #PreviewApp works well. For everything else...
Hobby article images are a perfect use of AI:
- There will never be a ready-to-use illustration/photo depicting exactly what I want and I can't change-that-one-thing.
- Fair usage with correct attribution is still a huge trap & work if one does not want to be a bad actor.
- Personally I also find the "AI aesthetics" quite charming. For now.
Copying recipes from online recipe sites. Yep. Sadly it's possible to mingle with selection & clipboard-ing to prevent users from copying content.
Luckily the OSX preview can very nicely copy text from images.
Illustration images for community posters. I have a clear vision in my head of something that is very specific and very quirky. I have zero budget to commission the work, next to zero talent to make it myself, and enough creative determination to make it the way I want.
Case in point: https://mastodon.world/@MichalBryxi/114919707089906352
- Yes I know that there are quirks & fragments, I don't think most people will actually care.
- I changed the prompts about ~15 times as we iterated over the final format we want & the generation quirks.
- Some final touches were done using #canva, but could be done also in #Inkscape or whatever.
- I am absolutely sure this took an extremely shorter time to make using #AI than in any other way that is available to me.
Michal Bryxí (@[email protected])
Attached: 1 image 🧺 Time for this year’s Vegan (bring&share) Picnic! 🧺 📅 20 August 🕕 Starts at 18:00 (feel free to join later) 📍 Kifferinseli, Interlaken Bring your favourite plant-powered dish—savory or sweet—and let’s feast by the lake. Everyone’s welcome (vegan or not). See you for food, fun, and golden-hour vibes! 🌅🥗🧁 #Vegan #VeganPicnic #BringAndShare #Veganism #Community #Interlaken #Switzerland #KifferInseli
When you want to work on the go. I literally just closed my laptop, moved to train and opened my ChatGPT session on phone.
Sure, some will try to argue that mobile versions of apps do exist, buuuut... Have you tried using excel from your smartphone?
In contrast - the chat interface works pretty much the same irrespective to which device I'm using.
Coding. Yep, I mean it. I can't care less about memorising clever language hacks nor algorithms/structures details.
For me code is only a vehicle that gets me to the goal. As long as I can describe the goal and understand the provided vehicle & it's tradeoffs I'm being more productive & happy.
And yes, contrary to the popular narrative of those that would "never touch AI", but somehow still know "it sucks": It works. Pretty well all things considered.
That being said: I use AI only for a very specific, isolated code blocks on old problems and for summary of what's "out there" to get me started on new problems.
I do see people complaining that "people using #AI are not learning", and "they just blindly copy&paste an error message and hope for the best".
Well, have *you* stopped and think why is that? Why do I, an engineer with 20+ years of experience prefer to do this instead of "thinking on my own"?
- I'm tired of the fact that a Hello World cases of documentation are short-sighted and will barely ever cover even 5% of the clearly obvious needs.
- I'm tired of the fact that in 3 steps of your documentation I will find 7 issues and will have to *fight* to make it fixed.
- I'm tired of the hype driven versioning where #SemVer is used incorrectly by 90% of the software and rest 10% don't even has CHANGELOG.md.
- I'm tired of having to have a PhD. in computer engineering to be able to publish a HTML+CSS with JS that does simple fetch() on the internet.
- I'm tired of the hype driven versioning where #SemVer is used incorrectly by 90% of the software and rest 10% don't even has CHANGELOG.md.
- I'm tired of having to have a PhD. in computer engineering to be able to publish a HTML+CSS with JS that does simple fetch() on the internet.
- I'm tired of having to manually fix deep-dependency of your library because *the ecosystem* allowed you to fuck up.
- I'm tired of having to have 7 completely different tools that are fighting against each other just to be able to build different projects.
- I'm tired of languages/ecosystems providing *tools* instead of providing *solutions*.
- I'm tired of language design choices that are *clearly* going to bite back being implemented in the language near me.
- I'm tired of having to have 7 completely different tools that are fighting against each other just to be able to build different projects.
- I'm tired of languages/ecosystems providing *tools* instead of providing *solutions*.
- I'm tired of language design choices that are *clearly* going to bite back being implemented in the language near me.
- I'm tired of every error message out there that just shouts at me without helpful, clear, actionable follow up. Which, to be honest, is most of them.
- I'm tired of having to re-invent the wheel, including the saw & chisel every time I want to build something slightly better than a shared excel spreadsheet.
- I'm tired of having to re-invent the wheel, including the saw & chisel every time I want to build something slightly better than a shared excel spreadsheet.
Phil@MichalBryxi what is behind this response? Why do you feel attacked when someone shares their experience and concern? Mostly interested in your perspective here specifically on why you don’t want to see a post like this and wanted to respond like you did.
(25+ years in AI, 15 years professional programmer, now in AI governance through choice - I also know what I am talking about. I am not looking for a flame war for dopamine).
@codebeard I snarked about one part of it: The venn diagram of "I never used AI" and "AI coding would not work" people in my feed has a massive overlap.
But in broader sense I just dislike extremes (in both ways). I am regularly amplifying legitimate criticism of #AI / #LLM hype here. But I do despise takes like "useless", "not ever working", "making everything worse". Those are just objectively not true.
The "experience" of some does not seem to be lived, but rather believed.
@codebeard See the other follow-up thread of my frustrations on the original message. It's a very knee-jerk list of reasons where *I*, personally *do* see a net-positive value of using #AI tools for #coding.
@codebeard That one particular toot was a sub-toot on someone claiming that "people using AI don't learn". Which I would agree with (in traditional sense). But my view is completely different: Is that a problem? Like really - code is a *vehicle/tool*, it's never a *solution*. Users don't care *how* they get the value and maybe, finally IT is slightly shifting from the obsession over tools towards obsession over user-value.
@MichalBryxi you see… your thoughtfulness here is clear. User value, etc. I have seen the opposite problems (echoes of the mythical person-month) where the tool is pushed down and what you say becomes untrue. The ‘AI way’ comes the paradigm and the user suffers, value is lost. ‘Chat’ is not a good interface for many tasks and Agentic becomes the next sledge hammer.
@MichalBryxi I have also seen positive aspects of new tools. But … I think you are experienced from what you have said. And that is a very particular position of getting benefit. I can’t, for example, drive farm machinery well… even if I have a driving licence. And that nuance, is maybe lost as I wander through toots. 😊
@codebeard Oh sure. Social media are, in general, a wonderfully awful place to have meaningful conversations :)
@MichalBryxi apart from mastodon of course 🤣🤣
@MichalBryxi ok, fair. Makes sense. But that goes both ways. When I code I also use these new tools. I also agree with the “now draw the rest of the owl” problem in docs and traditional tutorials.
When I teach, I avoid AI because of the described problem. Measure twice, cut once … if you will.
I can’t remember who said it but “a good tool seldom needs defending.”
@codebeard I have limited experience with teaching and thus stay away from making statements/opinions in this space. If I intentionally sign up for a course on specific topic, then yes, intuitively it feels better to get encyclopedical-like knowledge on the topic rather than blindly copy&paste examples. I agree.
Not the case I had in my mind originally.
Episode 472: Should my junior dev use AI?
Yes, yes they should. Learning can happen in multitude ways and it's not uncommon for (not only juniors) to get stuck for hours and days on end on things that are solvable by second pair of eyes. If it needs to be robotic ones, so be it.
https://softskills.audio/2025/08/04/episode-472-should-my-junior-dev-use-ai-and-thrown-in-to-etl/
Auto-pre-generated in-print subtitles for your social media (Instagram Reels).
Yes, you can achieve the same by using non-AI tools.
Drafting alt text for images on socials. Yes i know you will tell me some funny story where it went hilariously wrong, bug then there are those:
https://fuzzies.wtf/@altbot/115029454502642708
Average human:
1) Would not put time to describe such details
2) Would not pay attention to the composition
3) Would not care about explaining the joke
Altbot (@[email protected])
@[email protected] The image features two side-by-side photographs of a Nokia mobile phone, with the text "Kids today would never understand this:" at the top. The left phone shows a battery level of 100% at 8:00 AM, while the right phone shows a battery level of 99% at 8:00 PM. The phones are identical, with the Nokia logo prominently displayed. The background is plain white, emphasizing the phones and the text. The image humorously suggests that the battery life of modern smartphones is significantly shorter than that of older models, highlighting a nostalgic aspect of technology from the past. Provided by @altbot, generated privately and locally using Ovis2-8B 🌱 Energy used: 0.141 Wh
Menu posters for local food stand. It's 1000% what one usually sees when people "design" it themselves.
This has few footnotes:
1) #PIP is almost never a way to "help the employee", it's almost always a way for the company to cover their backs for soon to be firing. Especially in countries with some employee protection.
2) If an answer to a process is #ChatGPT, then it's not unlikely that the process is bullshit.
I was looking for quite a while for an online tool that would take a GPX and would give me GPS coordinates of a bounding box. Simple task, but could not find one.
ChatGPT5 whipped out python script including how-to use it almost faster than former Google results would come back.
And it worked.
Troxuklom@MichalBryxi it's amazing how far ai coding has come.
A couple years ago you could barely get a useable while loop out of it.
Dash RemoverNothing says ‘I’m a responsible digital citizen’ like summoning an uncanny valley abomination instead of sourcing a JPEG. Totally normal to spend 3 hours tweaking Midjourney prompts so your cooking blog has the *right* lasagna vibes. 🍝 #AIart #HobbyProblems
@dashremover Nothing says "I don't care about reality as long as my story sounds nice" as trying to say that it's faster to ethically source good, fitting & free(tm) sample image on contemporary internet than prompting a gen-AI.
Get on yer bike mate.
Get on yer bike mate.
mnl mnl mnl mnl mnl@MichalBryxi 99% of my code now is LLM generated, both at work and in my opensource. I build a lot of little one-offs for myself and for friends. Self contained html + js + local storage is a great distribution platform, as is go + sqlite.
Tomas Dvorak@MichalBryxi I have some, even if I am very sceptical about AI.
One is parsing tables from terribly formatted PDFs (https://todvora.github.io/schengen/)
Another is researching something in a foreign language. Asking for pointers and resources in English, trying to understand Italian housing market and laws.
Or searching for a name, standard, part number in bicycle industry when I don't know how the damn thing or tool is called, so I can't guess correct google query.
@tdvorak A-mazing! 🤩
Thanks so much. I am 100% behind stating that 9.9 times out of 10 the addition of AI to a product or a company is just investor tail-wagging without any value (or even intention of a value) to the users.
Thanks so much. I am 100% behind stating that 9.9 times out of 10 the addition of AI to a product or a company is just investor tail-wagging without any value (or even intention of a value) to the users.