“Just challenge…” is doing a lot of work. I have seen skilled engineers do exactly what you are describing and I don’t doubt their effectiveness in this very specific case, but they are skilled engineers with years of habits pre-LLMs.
LLM have an addictive feedback loop of the likes of Youtube Shorts. It’s easy to rely too much on the convenience, and you might sacrifice long-term rewards you would get from sitting with the discomfort in the duration between the now, and the moment when you achieve what you want.
These same questions can be researched or asked on forums, Discord channel, acquaintances, it’s a little bit less convenient but a lot more intentional and social.
It factually is an isolating technology, refraining from it is generally a lot healthier long-term.
One can use it responsibly, the same way one can be fit while having a diet consisting of pasta, pancakes, pizza and energy drinks—I would trust Michael Phelps doing so, but it would be a lot riskier for a beginner, and they will likely develop life-long bad habits.
It’s very true. I have 25 years of experience and I developed some taste for BS in code. It’s quite tempting to get LLM’s output as a face value and needs a strict hygiene.
When I heard about Zigs move to Codeberg middlefinger up to Github and Big Tech, it was the one thing that made me switch.
Yeah, GitHub has been incredibly unreliable for me for the past few weeks now, to the point where I can’t even browse files through the webpage. So I’ve already made a Codeberg account, but I just haven’t ported my repos yet. I heard about GitHub making deals with ICE too, which is confusing because why does a terrorist group need a Git service? Never thought that sequence of words would leave my mouth, but I guess we’re really living in crazy times.
Codeberg has an easy migration service. Simply select the “plus” at the right side in the top bar and select migrating repo. At least for not too complicated repos this works fine. Though, some things might be excluded (CI/CD pipelines etc.)
I agree with most of this except for learning.
I think that using it to learn things, having it explain things is good. I hate that it just shoves me the code most of the time, and prints out subsequent questions that are not in line with what I want and ask for.
Honestly, I’d love to have a human answer me those questions but I don’t want to spam this forum with questions like “What is a VTable?”, “Should I use a switch statement here?”, “Is this approach better or that one?”. Sadly I don’t have people that can answer me those in my surroundings. I mean physically, as in where I live and I’m in contact with. Online, I just don’t feel comfortable bothering others with basics that I lack so I resort to a chatbot for them.
I ask for examples, best practices, comparisons, how-to’s and why’s. I think it’s a good tool for that. With my combination of being overly critical about answers
Until the recent past I sometimes used chatgpt for these kinds of things. 80% of the times it was wrong. 95% of the times code examples were wrong, slow or useless (not to mention other terrriblly irritating properties of the thing).
Better just ask here, is my opinion. There are quite some people on this forum with lots of knowledge.
Andrew is mostly right with his bold statement.
Alright, I guess I should write here more often then, I hope you guys don’t mind
Regarding the code examples of ChatGPT, I stopped using it. I did in the past for the logic, but almost all of the code was handwritten since it just was bad at it. Google’s AI thing is better now. I did use it for some math code functions cause my background in math is almost 0. Then I went and spent a few hours understanding the logic behind it. Everything else I care about is hand written by me. All the refactoring I was always doing manually.
As someone who’s not a student, I can’t say with certainty, but my advice would be just don’t use it all the time - have “no AI” days. There are (naturally) days where AI does 95% of my work, and there are days where I do everything myself.
Look at the code, ask questions. AI is only intractable if you let it be.
These are questions that, honestly, you can only really get an answer on with experience, and looking at concrete examples for answers to. Even at that, these answers will be subjective.
An LLM is absolutely not the way to get these answers, using a probabilistic machine is already not great for unambiguous questions, using it for ambiguous ones is even worse, because what are you going to say to back up your opinion? “Source: ChatGPT?”
The ways to get these answers existed before mainstream LLMs, and they still exist now.
There is of course, this forum. If i may suggest, whether you decide to go with Rust or Zig, take time to learn the underlying computer concepts that both of those languages are offering to you. Things like virtual memory, pointers, how operating systems work and what they are trying to solve. These concepts will carry forward across any language you learn. Of course you will need to learn the programming language, but understanding basic concepts will help you understand what is going on. I can recommend “The Linux Programming Interface” as a great book, that has some of those basics in there (plus a lot more). I can also shamelessly recommend some videos I’ve made that cover some of these topics.
Now for my opinions on Rust vs Zig. I originally was really into Rust. It’s not a bad language, and a lot of good programmers work with it. It is tough to work with, especially coming from Python (that was my experience), as there were concepts that I just didn’t have the grounding for to understand. When I learned, there was a whole lot written about Heap vs. Stack, but it really made no sense to me. It wasn’t until i learned Zig that it clicked, because I was using allocators. (Something that was hidden away from me in Rust). Zig is a far simpler language. As an example it took me 6-7 months before i felt comfortable writing a more than basic Rust Macro. My first zig project made heavy use of comptime, because it was just like writing Zig.
In the end, if your goal is to learn, then you are going to have a good time working with either language.
The whole premise of this does make me laugh though. I’m not saying you’re wrong in your assessment on market expectations, but I suspect when you press people on this, the absurdity of being expected to know how to what amounts to using a glorified chatbot would be pretty evident.
Honestly, I don’t think there is much of a difference between using LLM or Wikipedia/search engine.
For the second one, I totally agree. Until I build enough experience at something I’m just listening and reading. I’m not using LLM to back up anything. That’s a red line for me I’m not crossing.
I mean most of my usage now for it comes in the form of “Explain this error message for me.”, and instead of opening man page in Linux for a specific tool if I’m doing something time sensitive, I try to use it to get the command that I need which sometimes results in me opening up and reading man page anyways. But at least I tried to be faster and more efficient. Even with explaining error messages, I see the slop, so yeah. It’s not perfect for that either.
Regarding programming, most of my time spent on it is reading this forum, and reading actual code. LLMs are just a fraction
Both search engines and Wikipedia are supposed to be tertiary sources. As LLMs actually scraped Wikipedia for their training data, I’d posit that that places them on a tier lower than a tertiary source in terms of reliability, and that’s even before you throw in all of the problems their probabilistic method of operation. You could probably save yourself some time by just going to Wikipedia or other sources and doing your due diligence there, as your path to validating the LLM’s output is probably going to send you that way anyway.
That’s a healthy attitude, and one I’d recommend keeping even after you have the experience.
Alot of your case hinges on AI being wrong. What will you say in the future when it’s correct? Plus beginners have ALOT of questions. I’m sure if we all posted 15+ threads a day here people wouldn’t like us. It’s ziggit not stack overflow (which is also basically dead right now).
I think that in the long term, the case against LLMs for beginners/students needs to be a moral one not a logical one. Most of my classmates would easily sacrifice some reliability to be able to cut through the noise and get a quick answer, unfortunately. Obviously the same AI that teaches them bites them back in the ass because it’ll steal all their junior positions too, but it’s not like we can stop the wave at this point.
The case against LLMs hinges upon LLM companies like Anthropic and OpenAI and their products like Claude and ChatGPT being massively unprofitable and likely to go out of business in the near future. Of course considering how exposed the rest of the global economy is to these LLM companies and related like cloud hyperscalers, Oracle, and Nvidia, this would probably destroy the global economy and send us into another Great Depression.
I help out in the Godot forums, I’ve kept up in C++ ircs. I love helping beginners, I would love to help beginners here too. Please do liven up our forums with any and all questions!
I’ve seen many times and been in many conferences when asking the room if they would eagerly help someone asking a question nearly everyone raised their hand, but when asked if they themselves feel confident asking for help almost no one would.
If you ask your questions here in public and open spaces then answers and be built upon, future readers can read previous answers and may find what they are looking for or specify an edge case. Even if one doesn’t look for their question answering the same question can be useful for zig (and any in-development software) as the language updates so must our questions and answers. I think stack overflow gamified itself too much with points and medals and got strangely competitive. While chatrooms like IRC and Discord lost history and blobbed all conversion together, repeated questions would interrupt longer conversations so hostility would face beginners in such a setting.
Discourse such as this forum ziggit, doesn’t change much from the stack overflow/bulletin board forumula, I just like the interface having an easy time finding and writing responses to questions. Most Help topics have a response within the working day, even in this admittedly low-volume forum. And in these contexts I’d argue it’s very okay and encouraged to fire-and-forget topics, come back tomorrow if need be. It’s instant messaging, not instant replying!
I appreciate your attitude, but from my personal experiences even if almost everyone in that room raised their hands, in reality the people who actually follow through would be much fewer, and the people that do so without overt/passive aggression even fewer. I think if we asked that question to a room full of blindfolded people it’d be different as well, there’s a lot of peer pressure in that room. Plus the people that show up to those conferences are way different than the “average” person in a forum.
Often the fastest people to respond to questions in online threads are the most agitated by the question in the first place as well. I think people’s fears here are very rational. They don’t come from nowhere. Sometimes when we try to help others we get annoyed and impatient as well.
I understand that ziggit may be the exception to this, it is a nicer community from my experience too, but any other mainstream website like reddit and it’s way worse. Also finding a good group of reliable mentors is the exception not the norm. People usually expect to be paid for personally training another guy. Really the true solution to this would be like a UBI to teachers and free classes or something. Much of their work benefits society as a positive externality and not directly, so their efforts go monetarily unappreciated.
