Mastodawn

I’ve been testing a theory: many people who are high on #AI and #LLMs are just new to automation and don’t realize you can automate processes with simple programming, if/then conditions, and API calls with zero AI involved.

So far it’s been working!

Whenever I’ve been asked to make an AI flow or find a way to implement AI in our work with a client, I’ve returned back with an automation flow that uses 0 AI.

Things like “when a new document is added here, add a link to it in this spreadsheet and then create a task in our project management software assigned to X with label Y”.

And the people who were frothing at the mouth at how I must change my mind on AI have (so far) all responded with resounding enthusiasm and excitement.

They think it’s the same thing. They just don’t understand how much automation is possible without any generative tools.

Jimmothy Baggins Nov 14

@mariyadelano it’s the new buzz word. Like big data and cloud before it. We’ve utilized technologies like it for years and the marquee companies come in and heavily market it. Like those bubbles this will burst to. Just hope it happens before most of the industry is laid off.

Isaac Ji Kuo Nov 14

@mariyadelano It's because they aren't computer programmers and I define "computer programmers" to be "the laziest people on Planet Earth."

No one else puts so much effort into automating something, to avoid a small amount of effort.

Antonius Marie ⚧Nov 14

@isaackuo @mariyadelano Meme of person turning away at option A, then pointing happily at option B:

A: Spend three hours Doing The Thing
B: Spend four hours writing a script to automate Doing The Thing

Isaac Ji Kuo Nov 14

@melindrea @mariyadelano Yep. But from my point of view, it makes sense.

The thing is ... it's not just about the time and the tedium. It's about actually doing the task correctly. If I'm doing the same thing over and over for three hours, how likely is it I'm going to do it 100% correctly with no mistakes?

I'm more confident of the script getting it done right.

Antonius Marie ⚧Nov 14

@isaackuo @mariyadelano Yes, agreed on that too. It's a lot likelier that my script to, say, "copy 500 bits of data from here to there" won't make a brain fart, whereas when I'm getting to bit 100 *my* brain will be looking for any distraction.

MidgePhoto Nov 14

@melindrea @isaackuo @mariyadelano
There's an xkcd of course.

A more measured response is option C:
Given a 3 hour job likely to be recurrent spend an hour writing and testing a script to do half of it.

Next time, spend an hour writing a script to do half the remainder.

@melindrea @isaackuo @mariyadelano but you learn so much while doing option B, and the next week you can apply that knowledge to something else...

literally happened to me this week. I had to investigate as many as several dozen rows of a user activity audit log, the relevant parts of which are in an XML column. rather than spend several minutes looking at raw XML I learned how to parse the relevant bits into columns and rows in T-SQL and show the relevant data without surrounding boilerplate in a single view. then I used the same technique to analyse a bunch of data in a different table for a completely different purpose

Isaac Ji Kuo Nov 15

@jackeric @melindrea @mariyadelano I gets a sads whenever I see XML in a SQL table field, but ... oh well ... I've been there, were it's the most sensible way to deal with the data.

@isaackuo @melindrea @mariyadelano oh this database is full of things to make you feel sad... the designer made a hobby of naming columns and tables with SQL keywords, and seemed to consider database normalisation to be a passing fad

@melindrea @isaackuo @mariyadelano https://xkcd.com/1205/

Is It Worth the Time?

xkcd

Regan King Nov 16

@_rowdy @melindrea @isaackuo @mariyadelano well, the one day a year saved creating a snowflake automation saves me the 30 minute slack back and forth + addtl time getting back to the interrupted work because I mis-copied a data point which happens a couple times in 56 weeks.

Checkmite Nov 15

@melindrea @isaackuo @mariyadelano

Okay fine - if we're talking about people who don't want to learn to program or don't want to spend the time, so they ask ChumpXYZ to write a script for them to use on their own computer, I don't even care. Go nuts.

The real problem IMO happens when these folks, who aren't programmers, who don't want to be programmers, who are completely uninterested in programming, nevertheless insist on being allowed to contribute to projects.

Antonius Marie ⚧Nov 15

@checkmite @isaackuo @mariyadelano I'm not entirely sure how that relates to this particular part of the thread? All three of us that were in it are people who don't use generative AI for our automation needs, we write the actual scripts ourselves.

Checkmite Nov 15

@melindrea @isaackuo @mariyadelano

It's not really refuting you, moreso the meme in certain situations.

Ken Butler Nov 14

@isaackuo @mariyadelano Larry Wall (the inventor of Perl) said that programmers should have "laziness, impatience, and hubris".

Cite: https://thethreevirtues.com

The Three Virtues of a GREAT Programmer

Woodstock Nov 14

@nxskok @isaackuo @mariyadelano

Well crap, 3 for 3.

Today I learned that, because of my genetics, I have doomed my children to a life of consistently over complicating things, and printer repair. Sorry kids, I didn't know.

Mark Hughes Nov 15

I don't know if that's true, but I score higher than it is comfortable to admit and had a great career. But perhaps not one for the CV!
@isaackuo @mariyadelano

@isaackuo @mariyadelano I have lost track of the number of times I thought about automating a throwaway task, did not, and then found myself repeating the throwaway task over and over and over...

Guy LeCharles Gonzalez Nov 14

@mariyadelano I'm in the middle of evaluating Hubspot and told the rep to avoid mentioning any AI features. He stumbled at one point and I reminded him that pattern matching and automation are things Hubspot was good at long before everything had to be called AI.

Mariya Delano Nov 14

@glecharles yeah sounds about right!
in all seriousness, though if you want an opinion from somebody that has been using HubSpot very heavily for multiple companies, I am happy to talk to you.

I use HubSpot basically every day and I have been for the past two years, I’ve also avoid their AI features so I mostly use all the other things so I think I hopefully would be able to provide a good perspective.

Guy LeCharles Gonzalez Nov 14

@mariyadelano I've been a fan for years, but we couldn't afford when we first launched in 2020 so we chose Zoho instead. We've outgrown that now so, if we do make the move next year, I'll definitely hit you up for some tips!

Gabriel Pettier Nov 14

@mariyadelano dunno if i’m high on AI, but i use it a lot at work.

I know i can automate things, have been programming for 20+ years, i write shell scripts, i have my own vim plugins, etc, but:
- LLM can write the automation code, too, and it’s often boring code, frequently throwaway, too, so that helps.
- Some things are a pain to automate, too many corner cases, etc, and the LLM itself can help with that.

But it’s good to teach what doesn’t need it.

Scarlet Phoenix Collective Nov 14

@tshirtman @mariyadelano Not to mention, for those of us who are not computer, programmers, like myself, it is a hell of a lot easier to put the conditions into an AI, and then have it regularly automation. Yes, I would have it, checked by an actual programmer, but since I am not in that market right now, that is not needed.

sidereal Nov 14

@the_spc @tshirtman @mariyadelano Honestly? That actually sounds so much more arduous and frustrating than just learning how to program (I say this as someone who taught myself to program. I'll never be a software engineer but I can do plenty of my own stuff).

If you actually learn how to program, you can fix your own mistakes. If the AI thing doesn't work, what do you do? Keep trying with different prompts? Then that's no longer faster.

Scarlet Phoenix Collective Nov 14

@sidereal @tshirtman @mariyadelano No, but it is somewhat less frustrating. I’m totally blind, and have no clue where to even start with learning how to program. I don’t even know the difference between the languages that programmers use. Not to mention, if there is math and/or other things involved, I would need a braille display. While I have one, I’m not quite sure it would do the trick, especially since it has a Perkins style rail, keyboard.

Gabriel Pettier Nov 14

@the_spc @sidereal @mariyadelano i admit i never really though about the usage of LLMs for blind users, it must have a lot of interesting use cases for accessibility, from description of things (on your terms), to delegating and automating interactions, though they produce very copious amounts of texts, that we tend to scan over looking for what we need, i assume you have to listen through a lot of it, even at high speed.

Scarlet Phoenix Collective Nov 14

@tshirtman @sidereal @mariyadelano We do, but it's a helluva l ot better t han bombardding youru sighhted family with hquestions.

The sticky part comees when we want to use itt t o make images b describing what we want it to prodduce.

Some say it sshoulddn't be used in crreative sppaces becauuse theft. While I won't deny thhatt part of it. It's finally excitinng for ust to be able to describe what we want, and have itt be put out. Accurately in many cases.

While I'd like to ppay an artisst, I'd still have to trust that they did my vision justice, and not did somethhing completely different.

Scarlet Phoenix Collective Nov 14

@tshirtman @sidereal @mariyadelano Smart glasses are our firends. Identifying thhings gust got got way easier and less time comsuming.

Scarlet Phoenix Collective Nov 14

@tshirtman @sidereal @mariyadelano Also, thank you. A lot of ppeople are just lik,ke.

"There's no use case that justtifies burning up the pplanet!!!!

The fact that your engaging with me, and admitting that tthere *are* interesting use cases is really very refreshing.

Gabriel Pettier Nov 14

@the_spc @sidereal @mariyadelano I think a lot of people are overestimating the footprint. Sure, it's not negligible, but it varies a lot depending on what you do, and i'm convinced many cases are legitimate in that regard. I'm probably a bit self indulgent myself, with using it to produce code, but that pales in comparison to the production of videos that people are doing these days, and it has useful output, rather than a joke forgotten seconds later.

Gabriel Pettier Nov 14

@the_spc @sidereal @mariyadelano But yeah, i think accessibility is one underappreciated consequence, having a translation layer between images and text, that has enough "understanding" of the world to allow useful interaction with it, has deep consequences. Sure, sometimes human help would be better, but there is a finite amount of that, and having agency is also very important.

So i'm glad it helps you.

Scarlet Phoenix Collective Nov 14

@tshirtman @sidereal @mariyadelano It is. Especially when we don't have to ask people seemingly every time we turn around goodd.

They're trying to develop an AI mobility thing, too. I wishh I could afford that thing, thouugh.

Scarlet Phoenix Collective Nov 14

@tshirtman @sidereal @mariyadelano Well, and that's just it. I'd love to learn how to program, at least on some level, but other than stuff like HTML/a tiny bit of markdown, which I don't' really consider programming as much as coding, I wouldn't know where thhe hell to start.

Not to mention the Braille display issuues we mentioned upthread.

Gabriel Pettier Nov 14

@the_spc @sidereal @mariyadelano html is programming, don't let people tell you otherwise, it's a designed for specific things, it's not general purpose, but it's a way to tell computer to do stuff, using code, even, it's programming.

General purpose languages are a bit different, but you don't need to learn them all, one is enough, like Python (though the meaningful indentation might be in the way for a blind user).

Scarlet Phoenix Collective Nov 14

@tshirtman @sidereal @mariyadelano HHuhh, good to know.

May see which one is easier. Althouugh pithom means that I might be able to make aaddons for NVDA.

Gabriel Pettier Nov 14

@the_spc @sidereal @mariyadelano i was having a look at
https://www.youtube.com/watch?v=ShDBkTe9fgQ which shows some html and python programming for beginners that is oriented for a blind audience, so the host explains the concepts but also the way to work and the pitfalls of it.

A Look at Programming Languages and Concepts for Blind and Low Vision Coders Part 2

YouTube

David Culley Nov 14

@mariyadelano This is exactly my experience. They just ask for some automation. And naturally, due to their cluelessness, their first thought is AI.

What they ask for can almost always be delivered with zero AI:

- find all occurrences of A in a text and replace them with B
- transform the CamelCase words in all this text to snake_case words
- auto-indent this text according to our style guide

All you need is a basic grasp of your text editor and/or the ecosystem of the programming language of your choice.

I understand that a non-technical manager doesn't understand the difference between this and AI. But when I see senior engineers who don't understand this, I wonder how they ever earned the Senior in their title.

Years since graduation from university isn't the same as experience. Having survived multiple years in the same job without growth isn't the same as being senior.

None of the mentioned "use cases for AI" are sophisticated. If you want to auto-format your code in a way that is always correct, just execute prettier, black, rustfmt or sqlfmt, depending on whether your code is written in JavaScript, Python, Rust or SQL. I will never understand people that write up a Confluence document, jotting down their personal preferences and pet peeves, and then ask Copilot to format the codebase according to that Confluence page, and each time get different results and/or merge conflicts with the non-deterministic LLM messing up half the time.

Also, LLMs are popular since about 3 years. How the fuck did managers ask their engineers to replace all occurrences of A with B, four years ago? Before they could say, "Use AI to do it."

You S Blues Nov 14

@davidculley @mariyadelano To your point, I was reviewing what is referred to as the “Machine Learning Lifecycle” yesterday and this is the first step:
“Step 1: Problem Definition
The first step is identifying and clearly defining the business problem. A well-framed problem provides the foundation for the entire lifecycle. Important things like project objectives, desired outcomes and the scope of the task are carefully designed during this stage.

Collaborate with stakeholders to understand business goals
Define project objectives, scope and success criteria
Ensure clarity in desired outcomes”

I’m thinking, isn’t that always the first step?

https://www.geeksforgeeks.org/machine-learning/machine-learning-lifecycle/

Machine Learning Lifecycle - GeeksforGeeks

Your All-in-One Learning Portal: GeeksforGeeks is a comprehensive educational platform that empowers learners across domains-spanning computer science and programming, school education, upskilling, commerce, software tools, competitive exams, and more.

GeeksforGeeks

David Culley Nov 14

@patmikemid Yeah, if you don't want to waste time working on the wrong problem, you should first think through what the problem even is, what its boundaries are, what you want to specify as in scope and what as out of scope, and how success would look like.

You S Blues Nov 14

@davidculley Back in the time of the dinosaur (the '80s), I took an Introduction to System Analysis course taught by Jack Gibson at UVA. One of his tenets was "The client does not understand their own problem." The problem definition section in that link is pretty similar.

Define the system as is (descriptive scenario)
Document how it should be (normative scenario)
Define objectives to get from descriptive to normative
Define measures (indices of performance) to evaluate your solution alternatives
Model

Charlie Stross Nov 14

@patmikemid @davidculley @Flux This is true: usually, understand a problem and you're already halfway to solving it.

You S Blues Nov 14

@cstross @davidculley @Flux I always chuckled when a younger developer or admin would pop over to my desk and dive right into the details of their issue and I'd stop them and ask, "Can you describe in words the problem you are trying to solve?"

They'd take a breath and start again and about two sentences in to it, they'd say, "Oh never mind, I've got it now" and they'd walk away.

"I'm a genius, I'd muse." Ha.

Paul Lalonde Nov 14

@cstross @patmikemid @davidculley candidly, I believe this is the skill that Ph.D. programs try to teach: how to ask meaningful questions so well that they nearly answer themselves.
But you don't need a doctorate to practice this!

:EA DATA. SF:Nov 14

@cstross @patmikemid @davidculley @Flux I solved a problem yesterday by clearly writing up a support ticket outlining all the things I'd already tried with examples. By the time I was done, I saw the hole with the things I hadn't tried and therein was the answer. I never had to submit the ticket.

You S Blues Nov 14

@jrdepriest 🎯

@davidculley @patmikemid https://i.pinimg.com/originals/58/dd/1c/58dd1cd6957dc487c960c4d51aef300b.jpg

cognitively accessible math Nov 14

@mariyadelano Yes.
It works the other way too. THere's this reading software where when they've done all the levels, they automatically can't do anything any more. They *can't* say "I want to go practice the vocab. in level K and L."
I said to tech support it felt like an invasion of AI and they said that oh, no, they didn't use AI!!!!!
Okay, algoruthm abuse :( :(

@mariyadelano Ans-Ible!

Let's start a unicorn...

none thanks left giving Nov 14

@mariyadelano parsing with regexes is deterministic, debuggable, efficient, and hard. parsing with LLMs seems easy, until you try to debug it…

Greg Johnston Nov 14

@mariyadelano Was reading a scientist say he saves lots of time by using AI to search reams of data. I'm not knowledgeable about programming, but isn't that what computer programs have always done? And so that's not really AI?

Beeker, in Prescott ON Nov 14

@GPJohnston @mariyadelano

I think for a lot of people, the term AI is replacing the word Software, with no change in meaning.

Bytebro 🇬🇧 🇺🇦Nov 14

Scripting languages can be your friend here, of course. My 'automated' stuff is now almost always a Perl script, usually with a bash script (or similar) front-end to invoke the correct version of the Perl interpreter and pass arguments in.

I'm down to 2 systemd scripts for backup stuff - everything else is a Perl script. This is on Garuda Linux.

George B Nov 14

@bytebro @mariyadelano

I've rewritten backup bash scripts so many times before finding rsnapshot and realizing it just does everything I want it to

Phil Thane ✅Nov 14

@mariyadelano
People need to learn some history of technology. About 300 years ago stream pumps needed a boy to open and close stream, water and drain taps in a sequence. Then a lad figured out that by trying string to different moving parts of the engine he could make it control its own valves. From that we evolved control systems engineering. Now we often do it with electrons rather than string but it's still just a logical sequence of events.

Alerta! Alerta!Nov 14

@mariyadelano Also what people often mean when they say "AI" is "chat-interface"