Recently I gave my first ever presentation at iO. It was about what it actually feels like to be a junior developer in a world where AI is just always there. This is a write-up of that talk, including what the room said back.
I’m not sure humans can do anything continuously without learning. It’s just a question of what your brain is filtering down to habit and experience. If your enemy is the compiler saying ‘error on line 314’ and line 314 is blank, you know to take a deep breath and start checking semicolons and braces. If your enemy is Copilot emitting ‘assert true’ tests, you know to grumble and move down a level in the specificity of your instructions.
At the end of the day - does the project work?
I’ve never used a LLM willingly, and otherwise avoid interaction with them where possible.
My base impression from interacting with those who actively choose to use LLMs, and unironically call them ‘AI’, is they’re a competency and credibility chipper.
It’s one thing to leverage a tool, where it makes sense for a task, and can otherwise be justified, but there’s a carcinogenic effect on cognition at play with prolonged exposure to whatever form LLMs are in today.
Also, happy 29th anniversary of Judgement Day.
I’m critical about a few points of LLMs (such as data stealing and licencing issues, big companies controlling online only and destroying environments and buy all RAM and so on), but not entirely against it.
Just wanted to say that the current LLMs we have are the closest thing to an Ai we ever had. Now what does the word artificial intelligence mean? And that’s it, the word is something different based on definition and context. There is no single correct definition for this word. We also used the term Ai way long before LLMs. And they were Ai too, let alone we use it still. In example in videogames we have Ai in form of NPCs or enemies. That is a completely different definition and technology of the word, but its the same word as used by LLMs.
Maybe you are thinking off general AGI? Because that is a whole lot of different conversation, AGI and self awareness. If you think in those terms, sure, then its laughable to talk about Ai being somewhat intelligent. I personally see LLMs as auto complete with very complex algorithms and huge databases behind it. And that is one form of Ai.
For me, a core requirement for intelligence is comprehension, something a LLM is intrinsically incapable of possessing nor performing.
‘Artificial Intelligence’ has been used for decades, but it has never functionally been anything more than a marketing term; originally conjured for a grant proposal, and subsequently oozed into entertainment and politics.
No one who is impressed by the function, or output, of a LLM to the level they consider it to be in any way adjacent to intelligence, should be involved in a decision making process which affects anyone not similarly cognitively impaired.
I think the way the word Ai we used in the past for computer programs and games, and with the modern LLM is, we actually mean “the illusion of an intelligence”. Because that is what the term Ai in most cases mean. Ai is as much of an illusion like VR is not a real reality. Maybe we should have called it Vi, for Virtual Intelligence. But that is just yapping about the words, where their meaning doesn’t change just because you use a different word.
Yes, there are lot of people who don’t understand the technology and think Ai is comparable to the way how “real” intelligence work. But in reality it is just perceived to be intelligent, making it a virtual illusion of it.
That there’s a gap between “it works” and “I understand why it works” and that gap is getting wider, not smaller.
20 years from now, important shit will go sideways, and no one will be able to fix it. That’s the world you’re building and that’s the world your kids will blame you for.
US CEOs experience imposter syndrome
It’s not imposter syndrome if you’re really an imposter.
A short read, but I really got a lot out of it. One thing I’ve started losing with the rise of AI and didn’t realize until now is the communication with coworkers about current problems we’re working through. I reach out to people in the git-blame log, and others used to reach out to me. But now everyone asks the AI, so we’re even more siloed than before.
It feels less like we’re working together and more like we’re working next to each other.
The workflow was: Google it, find Stack Overflow, maybe actually read the documentation (said as a joke, but not really).
Anyone that says this is not good at their job in the first place. I (and other actually competent devs I’ve talked to about it) have always gone straight to official documentation first. Stack overflow has its place, but it is not the first place to research.
I have met few colleagues in my career who bothered to read the official documentation, even when failing to find the answer on Google. Most would either just give up and say the problem is unsolvable (it never was) or try hack after hack until something stuck, never knowing why it worked.
Docs rarely give you information about anything except the happy path. StackOverflow did.
I (and other actually competent devs I’ve talked to about it) have always gone straight to official documentation first.
This is silly posturing. Anyone that is troubleshooting a problem tends to start by googling the error message, and more often than not Google will return SO hits, not official documentation.
Is that because you didn’t bother to read the documentation before using product X, and now you encountered some issue because you used X incorrectly?
Your comment shows some ignorance mixed with lack of experience. Errors do happen due to many reasons, and there is no project on earth that is nearly as documented as you pretend them all to be. This is one of the reasons why googling for error messages ended up being a very basic and mundane technique in software engineering.
What makes your comment stand out is the fact that you are completely unaware that some projects even use Stack Overflow to help users troubleshoot issues, with project maintainers taking the lead in providing answers. But here you are, pretending that reading docs is a silver bullet
After years of using SO, then years of going to docs first, I’ve started going back to SO. It just tends to be faster. You can find a question that is 90% similar to yours, rather than piece together scattered bits of information across pages and pages of docs.
I agree you can find an answer quickly on SO. And you can find it even faster by asking an AI. But you will not have learned much, if anything, by copy/pasting the answer. As opposed to reading the documentation, which might have taught you how to solve similar issues in the future.
But you will not have learned much, if anything, by copy/pasting the answer.
Your ignorance is showing. The whole point of using SO to help troubleshoot problems is avoiding the troubleshooting stage and all the iterative work of cross-checking docs with parametric tests. The main deliverable of SO is not a magical code snippet but a root cause analysis on a particular failure. You end up with a nice summary of someone else’s prior work troubleshooting the same issue, and a clear indication that X causes Y because of Foo.
And you do not get that by reading official docs documenting the happy flow. You’d know that if you had any relevant experience developing software.
Well I can also tell you that from years of trying to go straight to documentation: you rarely learn anything. Often the docs are written badly and missing usage examples, so it’s unclear how certain options work. Often the tool you’re having trouble with is one you only use once a month so by the next time you use it you’ve forgotten everything you learned.
Speaking from my experience, in most cases it’s just not worth it. In the few cases where it’s something you use every day, then sure invest some time in the docs.
Problem is docs are often out of date. Either old, or too new if you have to use an older version. So you can spend hours chasing your tail because what the docs and examples say works, doesn’t.
I do agree to start there, but reading the raw code if available is often times the best, and AI can do that much faster if its source available. And ofc it can be wrong in how it “thinks” it should work but so can humans when skimming quickly.
Problem is docs are often out of date. Either old, or too new if you have to use an older version. So you can spend hours chasing your tail because what the docs and examples say works, doesn’t.
I did say “first”, as in the first place you check and not saying you never have to go elsewhere.
“often” is doing some heavy lifting here. Stack overflow and yes looking at source code are useful tools, but the vast majority of the time, the “stack overflow coders” completely eschewed the official documentation when reading it would have not only provided then the information they needed but also helped then develop an intuition for how a library worked that would have served them much better.
Of course I mostly take issue with the blog presenting the experience of a “stack overflow coder” as a universal experience when it was just the experience of incompetents.
Are we human, or are we Dancer?
Is a boss just delegating work or leading?
A boss is not learning from other’s people work
… yes, thats how all of us learn. what a weird comment. do you really think you just magically know everything? we all benefit from the people who came before us.
EDIT: i saw the downvotes and actuslly had to verify the community i am in. i really am curious. to the people who do think we learn from others… how do you think you obtained the knowledge? how did you learn to program?
It’s just jibing. I don’t think the comment was meant in earnest. Many, including me have had bad bosses and felt unheard and left due to it.
is that what he is saying? maybe it flew over my head? i honestly didnt know what poknt ge was trying to make. thank you fir taking the time to explain it.
