As a manager, I don’t give a rats ass if you use AI. If you’re performing on par with your peers and you know and own your stuff, you’re safe.
Honestly, that last part is crucial and I’ve hd to remind some of my employees that using Claude is cool and all, but at the end of the day, you own it and I’m going to ask and assume you know it or we’re going to have a serious performance discussion. I can hire anyone to prompt Claude for me, what you bring to the table is what sets you apart.
As a manager, I want all my reports to be familiar with AI. I want them to know how to use it, how it will fail, but most importantly, how to recognize if something they’re working on has been touched by AI and what processes need to be followed if it has.
Because the thing that’s worse than junior devs not knowing what their code does is when they slip it into production and someone attempting to make updates doesn’t realize the code doesn’t follow human logic and may not do what they think it does.
When I ask the younger devs at work about their code they often tell me what they asked Claude to do instead of telling me what their code does. It’s so frustrating.
That is correct, but it’s also kinda beside the point. A compiler doesn’t “understand” anything either, it’s just a program, but it’s still very useful. So useful in fact that most programmers nowadays trust it unconditionally to the point where we say “a programmer doesn’t need to understand the code the compiler outputs”.
But AI is just really not trustworthy, and the question is whether they will ever be trustworthy enough that we can say “The programmer doesn’t need to understand the code the AI outputs”.
It sounds like your boss needs to stop being anyone’s boss. That kind of view is going to lead to piles of tech debt and everything that comes with it.
But, you know, my boss can now finally code himself. He could never get his head around coding, but now he finally can code. Who knows, maybe he won’t need us expensive software developers anymore when AI gets better, then he can do it all himself.
AI seems like it will help the already bad bosses do more - which seems like a bad thing.
It’s better if the bad ones just wank off in the toilet all day.
I think how much difference it will make in the long run will depend on who gets sacked and who survives.
Does a dev really need to understand the code if the AI understands it?
That if is doing a lot of lifting here, because AI doesn’t understand anything, it only finds mathematical responses to your question. There’s a reason the AI can’t give you the same script twice (or it constantly rewrites the fucking thing instead of just fixing the thing I told it to)
A compiler also doesn’t “understand” like a human does, but it’s so accurate and reliable, that collectively we decided that a programmer doesn’t need to understand the output of the compiler any more. We can trust it, because it works perfectly reliable.
So the question isn’t so much whether AI “understands” the way a human understands things, but instead whether it will become reliable enough to be trusted. And I don’t see any evidence towards that yet.
These models do have some form of understanding though. There are features for bugs and typos, and general features that map descriptions and pieces of code. It understands the code in so far it helps with next token prediction.
The bigger problem is that these language models are inherently unreliable and stochastic in how they generate. You request a feature - and it destroys something else in the process - because a single incorrect prediction caused it to diverge and skip a portion of the original code. You request a small modification - and it decides to restart from scratch - because random sampling made a different way of doing something more likely rather than what was there already. Errors compound and the model has no way of fixing or correcting them.
While I agree on the second paragraph, I’m gonna argue about the first, partially because I think the second invalidates the first.
These models do have some form of understanding though. There are features for bugs and typos, and general features that map descriptions and pieces of code.
The models don’t understand anything, they have rules that allow for finding tokens that don’t belong and fuzzy match to correct tokens (typos) and the ability to find code that breaks known rules for a language. That is no more understanding the problem than my spelling or grammar checking understands the comment I’m writing. ‘Understanding’ something requires intelligence and the ability to learn something and incorporate that knowledge into itself and use it to better process that information, not just finding tokens that break rules.
It understands the code in so far it helps with next token prediction.
And this is the crux of my beef, I think, because stochastic pattern matching is not understanding, it’s a mathematical representation of how the model processes your input tokens. The fact that it has to start over every time you provide it input, and uses the previous input/output tokens as context is why this is not ‘understanding’, it’s just fancy token prediction that gives a middling-to-passable facsimile to intelligence and understanding things.
The problems you note in your second paragraph fundamentally undermine the argument that there is any form of understanding to the AI, because those are basic mistakes that a trivial understanding of the problem would prevent.
That is no more understanding the problem than my spelling or grammar checking understands the comment I’m writing.
My general point would be that even a grammar checking can have some form of understanding of the text, no matter how shallow. The checker probably has a rule for when a is used versus an, if this rule generalizes across new words that rule that were previously unseen is a form of ‘understanding’ of the language being used in my view, despite being overly simplistic, while rote memorization - having a list of words that are followed by an - may not be.
LLMs are a weird case, because their internal representations for many concepts generalize even across new settings / inputs - in that sense the model has a form of understanding of what is being given, while for many other concepts the patterns break down even in the simplest of cases. It may ‘know’ that the preceding text is the writing of an essay - and it should autocomplete accounting for that -as is defined in its weights, but this understanding is shallow - it does not know why it knows, or how it knows - it cannot self reflect as it does not see and understand its own internal workings, or account for that. Yet the internal representations represent a form of text understanding that can be useful nonetheless - it is a language model after all.
My comment was intended to show this duality, hence the duality between the two paragraphs.
And this is the crux of my beef, I think, because stochastic pattern matching is not understanding, it’s a mathematical representation of how the model processes your input tokens. The fact that it has to start over every time you provide it input, and uses the previous input/output tokens as context is why this is not ‘understanding’, it’s just fancy token prediction that gives a middling-to-passable facsimile to intelligence and understanding things.
The problems you note in your second paragraph fundamentally undermine the argument that there is any form of understanding to the AI, because those are basic mistakes that a trivial understanding of the problem would prevent.
I am not entirely grasping the point you are trying to make here. I am certainly not arguing that it is conscious, self-aware, or in any way not a mechanical procedure that is being performed (I would not argue for that!). My key point is that it is not a simple black or white it understands / it does not understand - it may have internal representations that relate many concepts together, allowing it to draw upon these links when generating text, giving it a certain semantic understanding of the language and text it is using, while simultaneously not having a bit of self-awareness.
My point is that saying an LLM understands anything is anthropomorphizing the LLM and leads people into thought patterns that give it an inordinate amount of authority because people equate the simulacrum of understanding/comprehension with actual understanding.
I think we just fundamentally disagree on the concept of llms a being able to understand a topic rather than it being a shallow statistical prediction if the correct answer, and I just can’t equate understanding with statistical predictions. The fact that the underlying math is able to generalize the prediction in novel ways lends weight to the misbelief that it understands concepts, but the decoherence that happens over long conversations should shatter the illusion.
That’s fair. I actually don’t think we disagree that much - I just think I have trouble conveying what I am trying to say. Whenever someone talks about ‘shallow statistical predictions’, I think about older techniques like Statistical Machine Translation which even had trouble with things like word order, LLMs handle text on a higher level of abstraction (which I described as a form of textual understanding) - and hence handle things like word order better - but are still inherently statistical predictors. The model stores info about how words interact and relate to one another, but it does not ‘understand’ what the words actually (physically?) represent beyond these interactions nor does it ‘understand’ what it is doing. Albeit, those interactions are modeled well enough to give a convincing replica of doing so.
i have a contrast fpr you - there is a younger guy at work who transitioned from more of an analyst to a dev. i work with helping all the devs but because he doesnt have the background he needs extra help. he has used ai more to help him. i have helped him use guardrails and to be precise so he doesnt go down rabbit holes. BUT… after ai helps him get there he takes the time to learn and understand what he has done so he can get there on his own in the future
this is how it should be. it drives me nuts that actual developers at your job have no idea what they are doing
I recently proposed a solution to my boss (who’s older than I am, but not by much) regarding a technical problem we have been having the last few days. He said “that won’t get you anywhere. Ask Gemini.”
Reluctantly, I did. Besides asking it how to disable it, it was the first time I (sort of) voluntarily interacted with it.
It made three suggestions. Two didn’t work. The other one was the solution I proposed to my boss. I told him that (more politely) and he said he wasn’t really interested in addressing the problem anyway and that I should move on.
I’ve gotten this shit from all sorts of people (note: I am not in software dev)
I just repeat my question until they either answer it or fuck off without me doing whatever they wanted me to do, because no I’m not going to waste my time with your half finished work
Uhh the “performing on par with your peers” part is doing a lot of work here. It’s like saying, I don’t care if you’re doing speed at work, as long as you’re working as fast as everyone else. Except AI affects everyone on the planet, not just yourself.
As a manager, I don’t give a rats ass if you use AI. If you’re performing on par with your peers and you know and own your stuff, you’re safe.
Honestly, that last part is crucial and I’ve hd to remind some of my employees that using Claude is cool and all, but at the end of the day, you own it and I’m going to ask and assume you know it or we’re going to have a serious performance discussion. I can hire anyone to prompt Claude for me, what you bring to the table is what sets you apart.
As a manager, I want all my reports to be familiar with AI. I want them to know how to use it, how it will fail, but most importantly, how to recognize if something they’re working on has been touched by AI and what processes need to be followed if it has.
Because the thing that’s worse than junior devs not knowing what their code does is when they slip it into production and someone attempting to make updates doesn’t realize the code doesn’t follow human logic and may not do what they think it does.
When I ask the younger devs at work about their code they often tell me what they asked Claude to do instead of telling me what their code does. It’s so frustrating.
And yet, that’s all my boss wants of me. “Does a dev really need to understand the code if the AI understands it?”
That’s frustrating.
(The AI does not, in fact, understand it. The AI doesn’t understand anything at all. It’s a statistical text prediction machine.)
That is correct, but it’s also kinda beside the point. A compiler doesn’t “understand” anything either, it’s just a program, but it’s still very useful. So useful in fact that most programmers nowadays trust it unconditionally to the point where we say “a programmer doesn’t need to understand the code the compiler outputs”.
But AI is just really not trustworthy, and the question is whether they will ever be trustworthy enough that we can say “The programmer doesn’t need to understand the code the AI outputs”.
It sounds like your boss needs to stop being anyone’s boss. That kind of view is going to lead to piles of tech debt and everything that comes with it.
But, you know, my boss can now finally code himself. He could never get his head around coding, but now he finally can code. Who knows, maybe he won’t need us expensive software developers anymore when AI gets better, then he can do it all himself.
I mean, I haven’t worked at any company that doesn’t have a decade or more of tech debt all human-made.
Is this any worse such a regard?
I do not advocate for more bullshit, but I don’t really think that this will make a big difference either way.
AI seems like it will help the already bad bosses do more - which seems like a bad thing. It’s better if the bad ones just wank off in the toilet all day.
I think how much difference it will make in the long run will depend on who gets sacked and who survives.
By the time that’s a problem, he will have failed upwards because of his ‘forward-thinking’ AI push.
Yeah, he likely will. The whole department management is in a gold rush here.
That if is doing a lot of lifting here, because AI doesn’t understand anything, it only finds mathematical responses to your question. There’s a reason the AI can’t give you the same script twice (or it constantly rewrites the fucking thing instead of just fixing the thing I told it to)
A compiler also doesn’t “understand” like a human does, but it’s so accurate and reliable, that collectively we decided that a programmer doesn’t need to understand the output of the compiler any more. We can trust it, because it works perfectly reliable.
So the question isn’t so much whether AI “understands” the way a human understands things, but instead whether it will become reliable enough to be trusted. And I don’t see any evidence towards that yet.
These models do have some form of understanding though. There are features for bugs and typos, and general features that map descriptions and pieces of code. It understands the code in so far it helps with next token prediction.
The bigger problem is that these language models are inherently unreliable and stochastic in how they generate. You request a feature - and it destroys something else in the process - because a single incorrect prediction caused it to diverge and skip a portion of the original code. You request a small modification - and it decides to restart from scratch - because random sampling made a different way of doing something more likely rather than what was there already. Errors compound and the model has no way of fixing or correcting them.
While I agree on the second paragraph, I’m gonna argue about the first, partially because I think the second invalidates the first.
The models don’t understand anything, they have rules that allow for finding tokens that don’t belong and fuzzy match to correct tokens (typos) and the ability to find code that breaks known rules for a language. That is no more understanding the problem than my spelling or grammar checking understands the comment I’m writing. ‘Understanding’ something requires intelligence and the ability to learn something and incorporate that knowledge into itself and use it to better process that information, not just finding tokens that break rules.
And this is the crux of my beef, I think, because stochastic pattern matching is not understanding, it’s a mathematical representation of how the model processes your input tokens. The fact that it has to start over every time you provide it input, and uses the previous input/output tokens as context is why this is not ‘understanding’, it’s just fancy token prediction that gives a middling-to-passable facsimile to intelligence and understanding things.
The problems you note in your second paragraph fundamentally undermine the argument that there is any form of understanding to the AI, because those are basic mistakes that a trivial understanding of the problem would prevent.
My general point would be that even a grammar checking can have some form of understanding of the text, no matter how shallow. The checker probably has a rule for when
ais used versusan, if this rule generalizes across new words that rule that were previously unseen is a form of ‘understanding’ of the language being used in my view, despite being overly simplistic, while rote memorization - having a list of words that are followed by an - may not be.LLMs are a weird case, because their internal representations for many concepts generalize even across new settings / inputs - in that sense the model has a form of understanding of what is being given, while for many other concepts the patterns break down even in the simplest of cases. It may ‘know’ that the preceding text is the writing of an essay - and it should autocomplete accounting for that -as is defined in its weights, but this understanding is shallow - it does not know why it knows, or how it knows - it cannot self reflect as it does not see and understand its own internal workings, or account for that. Yet the internal representations represent a form of text understanding that can be useful nonetheless - it is a language model after all.
My comment was intended to show this duality, hence the duality between the two paragraphs.
I am not entirely grasping the point you are trying to make here. I am certainly not arguing that it is conscious, self-aware, or in any way not a mechanical procedure that is being performed (I would not argue for that!). My key point is that it is not a simple black or white it understands / it does not understand - it may have internal representations that relate many concepts together, allowing it to draw upon these links when generating text, giving it a certain semantic understanding of the language and text it is using, while simultaneously not having a bit of self-awareness.
My point is that saying an LLM understands anything is anthropomorphizing the LLM and leads people into thought patterns that give it an inordinate amount of authority because people equate the simulacrum of understanding/comprehension with actual understanding.
I think we just fundamentally disagree on the concept of llms a being able to understand a topic rather than it being a shallow statistical prediction if the correct answer, and I just can’t equate understanding with statistical predictions. The fact that the underlying math is able to generalize the prediction in novel ways lends weight to the misbelief that it understands concepts, but the decoherence that happens over long conversations should shatter the illusion.
That’s fair. I actually don’t think we disagree that much - I just think I have trouble conveying what I am trying to say. Whenever someone talks about ‘shallow statistical predictions’, I think about older techniques like Statistical Machine Translation which even had trouble with things like word order, LLMs handle text on a higher level of abstraction (which I described as a form of textual understanding) - and hence handle things like word order better - but are still inherently statistical predictors. The model stores info about how words interact and relate to one another, but it does not ‘understand’ what the words actually (physically?) represent beyond these interactions nor does it ‘understand’ what it is doing. Albeit, those interactions are modeled well enough to give a convincing replica of doing so.
@Passerby6497 @8uurg #imho people believe #computers are #god like perfect machines, free of any #errors nothing could be further from the truth, basically every #cpu because of its complexity has #errors or even #security #flaws that need to be corrected afterwards via #software #microcode #updates yes current #llm #ai does not understand anything? It is just very good at guessing the next token to output?
thats crazy yo me.
i have a contrast fpr you - there is a younger guy at work who transitioned from more of an analyst to a dev. i work with helping all the devs but because he doesnt have the background he needs extra help. he has used ai more to help him. i have helped him use guardrails and to be precise so he doesnt go down rabbit holes. BUT… after ai helps him get there he takes the time to learn and understand what he has done so he can get there on his own in the future
this is how it should be. it drives me nuts that actual developers at your job have no idea what they are doing
I recently proposed a solution to my boss (who’s older than I am, but not by much) regarding a technical problem we have been having the last few days. He said “that won’t get you anywhere. Ask Gemini.”
Reluctantly, I did. Besides asking it how to disable it, it was the first time I (sort of) voluntarily interacted with it.
It made three suggestions. Two didn’t work. The other one was the solution I proposed to my boss. I told him that (more politely) and he said he wasn’t really interested in addressing the problem anyway and that I should move on.
I’ve gotten this shit from all sorts of people (note: I am not in software dev)
I just repeat my question until they either answer it or fuck off without me doing whatever they wanted me to do, because no I’m not going to waste my time with your half finished work
Uhh the “performing on par with your peers” part is doing a lot of work here. It’s like saying, I don’t care if you’re doing speed at work, as long as you’re working as fast as everyone else. Except AI affects everyone on the planet, not just yourself.
Except doing drugs is illegal, whereas using AI is not. Fairly important distinction.