Here is a thing I've been wondering about:
Let's for the moment say that generative AI is Absolute Evil. No wiggle room, it's just straight up bad for EVERY use case.
Can all you smart AI hating people out there come up with good tools to help fill the gap?
Like, if using AI to generate code is bad, can we make programming languages or paradigms that lower the bar to entry and make it possible for more people to be empowered to create their own programs?
I feel like there are smart people making some very good points all around, but I can't help but wonder if all this negative energy is being mis-directed.
I feel like more people USED to have that vision. Remember Hypercard? Or Visual BASIC?
Where have all the tools like this that enable and empower gone?
Digital Mark λ ☕️ 🕹 👽 reshared this.
Isaac Ji Kuo
in reply to Feoh • • •Apple dropped the ball with Hypercard, sure, but Visual BASIC spawned numerous clones including within Microsoft, which would get smoothly transitioned into a combined product, Visual Studio.
Which still exists and is still super popular. I'm not entirely a fan, especially not a fan of Microsoft, but it's ... it's still a thing.
Is it too complicated to jump into maybe? Sure, but Python has stepped up to be the "easy" way in.
But anyway, I do wish "man" would "upgrade" to use examples.
Isaac Ji Kuo
in reply to Isaac Ji Kuo • • •Back in the day, "man" had to be terse just to be able to not consume an excess amount of disk space. But that's not an issue now.
Is it so much to ask, for the documentation to include examples of use? I mean ... I get by with internet searches but that really shouldn't be necessary I think.
thomastc | frozenfractal
in reply to Isaac Ji Kuo • • •tldr pages
tldr.shMichael Weiss
in reply to Isaac Ji Kuo • • •@isaackuo I enjoyed programming in VBA back in the day. When I was Chief Architect for IFX, I used it to generate XML DTDs from the specification documentation, guaranteeing that the two were in sync. Since we always worked from the spec, it just made sense to go that route.
I've never heard of anyone else doing that before or since.
-dsr- (hypoparenthetically)
in reply to Feoh • • •If you have a syntax good enough to convey your intent to a computer, you have a programming language.
As far as I know, the two big barriers to learning to code (after willingness, time and opportunity, of course) are the concept of variables and the concept of pointers. There are languages which hide pointers, but you still need to understand them to know what's going on -- it's just delayed until later in your education.
"Low code" programming tools are hip, but they are almost always front-ends to domain specific languages -- great until you hit the wall.
You've already heard one of my big takes: corporations are dropping in LLMs as gatekeepers to multiple databases, and almost always this becomes a security and performance issue. The utility, of course, is undeniable -- but if your LLM can be replaced by a tutorial to a SQL frontend, it probably should be.
How different is a really good LSP-enabled IDE with a great not-in-this-codebase search function from an LLM code plagiarist?
Isaac Ji Kuo
in reply to -dsr- (hypoparenthetically) • • •@dashdsrdash It would be so nice if we could have nice things ... and by that I mean one specific nice thing - exposed SQL functionality instead of just APIs or worse, web scraping.
Every time some asks something like, "How can I query my list of followees for X characterstics?" I think of what a world it could be if we all exposed SQL (along with properly maintained SQL level security, of course).
iogrt
in reply to Isaac Ji Kuo • • •Feoh
in reply to -dsr- (hypoparenthetically) • • •@dashdsrdash This whole thread and your response in particular are making my point for me in 40 foot tall blinking neon.
You're basically saying that humans can't create programs if they don't understand pointers.
That's provably false. See my initial thesis about tools like Hypercard and Visual BASIC.
Neither had anything like pointers.
I think we've cargo culted ourselves into a world view that's so narrow we can't see beyond gcc, and that's kinda sad IMO.
-dsr- (hypoparenthetically)
in reply to Feoh • • •I may have not expressed myself properly.
People have problems understanding variables. That's the first cut in CS100 courses.
People have problems understanding pointers. They end up needing to understand them when they get sufficiently advanced; there's a lot that can be done in between "I understand variables" and "I don't understand pointers".
Isaac Ji Kuo
in reply to -dsr- (hypoparenthetically) • • •@dashdsrdash FWIW, I understood what you mean.
Like, a back end programmer can do great with just a solid understanding of SQL and a passing understanding of whatever preferred "glue" language is being used to tie to the front end.
But if you use SQL at scale, you've got to understand SQL optimization, and that requires some understand of indexes. What are indexes? A bunch of pointers.
Hypolite Petovan
in reply to Feoh • • •@Feoh What do you think more people creating their own programs would achieve, realistically?
I'm a programmer by training and by trade, and I didn't create the overwhelming majority of programs I use daily.
Let's take your premise around and reverse it: let's say that generative AIs are absolutely good. They don't draw from a corpus of stolen work nor draw hazardous amounts of electricity and water and produce correct programs. What good would it be for most people?
Jonathan Lamothe
in reply to Hypolite Petovan • •@Hypolite Petovan @Feoh The majority of the code I've written never sees the light of day because it was some niche utility I wrote for a weirdly specific task that nobody else is ever going to give a crap about (highly customised scripts and stuff that wouldn't even make sense outside of my own personal system).
Programming is a useful tool to have in your toolbox, but most people can (and do) also generally get along quite well without it.
like this
Feoh and iogrt like this.
Hypolite Petovan
in reply to Jonathan Lamothe • • •Jonathan Lamothe
in reply to Hypolite Petovan • •@Hypolite Petovan @Feoh Ohh... I misread your question. I thought you were arguing in favour of LLM-generated code... which honestly surprised me.
It turns out I just need to actually pay attention to what I'm reading. 🙃
like this
Feoh and Hypolite Petovan like this.
Feoh
in reply to Jonathan Lamothe • • •@me @hypolite No worries a LOT of people are doing that.
Just look at some of the (not yours!) super smug, heavy handed replies.
Feoh
in reply to Hypolite Petovan • • •@hypolite @me You're ignoring my question and trying to pin me into a corner so you can cudgel me with your superior anti-AI "facts".
Thanks, but no.
Hypolite Petovan
in reply to Feoh • • •Feoh
in reply to Hypolite Petovan • • •Hypolite Petovan likes this.
Feoh
in reply to Hypolite Petovan • • •@hypolite @me So, I want to apologize for my extreme response.
I'll admit I'm a bit frustrated with:
Re-reading your post it seems I over-reacted and you weren't necessarily doing any of those things.
Hypolite Petovan likes this.
Hypolite Petovan
in reply to Feoh • • •@Feoh @Jonathan Lamothe Thank you for this. I am frequently equally frustrated in conversations about AI, specifically generative AIs based on Large Language models because the people who make any sort of claims about what they can or could do also are people with the least understanding of how it works technically.
The truth is that this crop of AI is engineered to fool us humans, including about their capabilities. Because that's the target model trainers have set for them. And it turns out machine learning systems are uncannily good at reaching their set goals, regardless of any other consideration.
And so you have people who use AIs casually who are rightfully bewildered by their apparent capabilities, while experts in their respective fields who try to use AIs to enhance their workflow end up dropping them for a variety of reasons (inaccuracy, lack of underlying understanding of the subject matter, loss of ownership of output, etc...).
Does this mean a fooling machine can't produce an accurate answers? Absolutely not, but it will make figuring out the inevitable inaccurate answers harder because it's already been so good at fooling the people who trained the model.
Even without considering the ethics (or mostly the lack thereof) of this current crop of AI, it cannot answer any need that isn't about fooling people at scale.
Feoh
in reply to Hypolite Petovan • • •@hypolite @me I couldn't agree more.
There are so many potential harms that come from unleashing prompt based conversational AI on the masses, you'd exhaust yourself just trying to enumerate them.
As you say, code generation as well poses some very real dangers.
It's actually why I asked the initial question "Hey, let's assume for the moment this stuff is straight up BAD. I remember there being more tools to enable people to scratch their own itch in the past. Where did they go?"
So far we have Python (Duh, can't believe I overlooked my favorite programming language!), Excel, Access and maybe Scratch?
Hypolite Petovan likes this.
Hypolite Petovan
in reply to Feoh • • •@Feoh @Jonathan Lamothe Then I guess it's my turn to apologize for misunderstanding your original post. I thought you meant to claim that whatever harm AI agents are doing, they would somehow be the only way for more people to access programming because accessible programming solutions disappeared a long time ago.
The "let's assume generative AI is straight up BAD" part led me to believe you ascribed some good to the technology as it's currently developed because in my view this isn't a hypothetical, it's a reality, including poisoning the way we talk about it for the reasons I spelled earlier about fooling us about its capabilities.
queselevaaser
in reply to Hypolite Petovan • • •In this sense, the main issue with gen-AI is that it cannot conceptualize. It has no sense of what it's "saying", no understanding. As prev said, it's a fooling machine: an algorithm that generates a response based on statistical data with no understanding of what it's saying or what it's answering to.
Bc of this, inaccuracies, mistakes, and outright falsehoods happen all the time, and because it was made to imitate our patterns of speech*, it's damn hard to spot them. The time it takes to clean up gen-AI generated content is ultimately more laborious than just doing it yourself (seriously. it's the bane of translators everywhere, even worse than MT, and i figure programmers are somewhat on the same boat). Plus, due to the same, it also reproduces any societal bias that are unintentionally (or intentionally) introduced into it's training data, and possibly exacerbated.
Beyond gen-AIs inherent issues (and i'm not getting into the environmental side of things, that would require too big a sidetrack into the environmental impact of everyday internet usage which im not rlly willing to get into rn), there's also the fact that gen-AI is a tool, and the way it's being used….. it's downright dangerous. Between data collection, job loss, hiring third worlders under the table to do the work the gen-AIs can't, or can't do correctly, the fact that you can't really know what "ideology" or biases are being introduced into the AI (well, you can guess, we've all seen Grok), not to mention the use of AI in general in the war machine, from actual "security" tech to the propaganda machine? Yeah, you can see why so many ppl are fucking rabidly against AI. And no, I'm not getting into the copyright issue, which also pisses people off a lot, but like, fuck property anyways. I will give it to you that using tech to fake an actor's likeness is beyond fucked up tho.
*specifically referring to LLMs here onwards, and their Chat-GPT type agents or however you wanna call it, since it's what im most familiar with.
anyways, sorry if this response is kinda ehh, it's 4am and i'm insomnic (:
Hypolite Petovan likes this.
Jonathan Lamothe
in reply to Jonathan Lamothe • •@Hypolite Petovan @Feoh The more I think about this the more I feel I should point something out:
The reason I only tend to have to write super custom stuff is because for everything else there tends to almost always be some FOSS tool that does what I want. That is only the case because someone decided to create and maintain it.
It's important to ensure that there remains a pool of people with the skillset and time to sustain this ecosystem. I don't want software development to become a lost art because nobody feels any reason to bother with it.
Hypolite Petovan likes this.
ɗ𐐩ʃƕρʋ
in reply to Feoh • • •Isaac Ji Kuo
in reply to ɗ𐐩ʃƕρʋ • • •ɗ𐐩ʃƕρʋ
in reply to Isaac Ji Kuo • • •Isaac Ji Kuo
in reply to ɗ𐐩ʃƕρʋ • • •@deshipu Well, one Hypercard inspired development environment was already mentioned - Visual Basic. There were also others like Foxpro (which included Hypercard inspired "visual" form editing and code snippets before being bought by Microsoft).
The basic development paradigm of Hypercard didn't go away, it thrived. Apple could have been riding high on it. Instead, others took the baton and ran with it.
Very ironic, if you look at what Steve Jobs was doing at NeXT...
Jonathan Lamothe
in reply to Feoh • •@Feoh I feel that Python tries to do this. I wish there were more programming tools built with the aim to being accessible to novice programmers.
I am probably a bad person to ask this. I've been programming for so long that there are many things that feel obvious to me that aren't at all to a novice.
My biggest gripe with so-called AI is that whether it's useful or not, a lot of it is not even remotely sustainable.
like this
Isaac Ji Kuo, Feoh and Dan (he/him) like this.
Feoh
in reply to Jonathan Lamothe • • •Isaac Ji Kuo
in reply to Feoh • • •Oh, I forgot to mention the ACTUAL runaway "success" story for making programming accessible.
Excel.
Like it or not, there are absolutely oodles of people who think of themselves as not able to "program" a computer, but they do complex computing tasks in Excel.
This has particularly become a problem in science, where many scientists use Excel to do complex calculations because that's what they know how to use.
It becomes problematic because it's difficult/impossible to replicate.
Jonathan Lamothe likes this.
Jonathan Lamothe
in reply to Isaac Ji Kuo • •@Isaac Ji Kuo @Feoh God I hate Excel, but also... yes.
Also, just because I don't like it means I think there's anything wrong with others using it.
like this
Isaac Ji Kuo and Feoh like this.
Hypolite Petovan
in reply to Isaac Ji Kuo • • •Digital Mark λ ☕️ 🕹 👽
in reply to Feoh • • •There's tons of them. The most popular programming medium is the web, you can literally write crap in an editor and reload a page to run a program. We often complicate it with "frameworks" but those are supposed to make it easy to do high-level tasks.
I live in game dev, where PyGame is trivial. Lua toys like pico-8, and engine scripting are super common. Unity is a fucking awful thing, but Godot has a tolerable scripting language, or a drag-blocks-around mode.