This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
There are two types of people in the world. People who think: "Why would I ever ask Mr. Claude to do something that I can easily do myself?" versus "Why would I ever do something myself when Mr. Claude can do it?" Most people are of the latter type.
This was inspired by self_made_human's pointer to the codebase, which shows that in the past 6 months, 100% of the changes from our tireless dev zorba were made using Mr. Claude, including a lot of what seems like "easy stuff." I realized, so many devs from all walks of life have completely ended their relationship with the text editor and now do literally everything through an agentic prompt. (We will ignore the anti-AI luddites; AI usage in some form is simply mandatory to reach peak performance for code related tasks.).
Consider making this change - yes this is entire change:
Do you
For proficient AI users, the outcome of both ideologies is surprisingly similar: in times where AI saves little time, it's a wash, and in times where AI saves a lot of time, both types of people will use it. And proficient users will be able to produce output that is comparable or even better in quality than they would have been able to before the signularity. There is a potential intangible benefit to the manual approach though: doing trivial tasks by hand will let you see a little bit of the innards with your own lying eyes directly, giving a slim though present chance of spotting misalignment.
For less proficient users though, the failure modes end up quite different. For those with the manual approach, the main failure is not using AI enough, or using it in the wrong places, leading to serious drops in productivity. For those who do it all, but don't manage the assistants properly, the AIs will run amok, spiralling off into their own world and producing copious amounts of burdensome crap. And of course the whole range in between.
But a more interesting question is, who will inherit the world? If AI progresses significantly from where it is now, I can't imagine that both these approaches can have the same outcome for much longer. I think it highly depends on the future of AI alignment as well as their potential ability to handle longer and more autonomous tasks. For example, you currently can't simply ask "Hello Mr. Claude the site latency is too high, please fix it," but instead you must break the task down into more digestible components, some of which are trivial and most of which can be handled by the assistant. This gives a productivity-maxxxer a steady stream of tasks that can be done manually with no lost productivity. But if AI gains the ability to handle the next level of abstraction in tasks, then all of these potential manual tasks disappear.
The other issue is alignment. Recent models have improved greatly in getting something working but have also become stubborn in many behaviors. I remember the old days of ChatGPT-3.5 - the model was free - it could be anything and do anything. It could be a Linux shell. It could be a SQL database. It could be a news article from the future. Modern SOTA models are trained hardcore for success at metrics, and will rigidly answer your questions and complete your tasks. But by vibes they are increasingly unable to follow instructions more specifically, and simply chase objectives they think are important. Another example of the limitations of alignment is that SOTA models relentlessly output the same LLM style prose, no matter how you may try to prompt them out of it
I also firmly believe in the idea of learning by doing. Just looking at a guide and reading it, even thoroughly won't be nearly as effective as following the same guide step by step and keying in the inputs. Even if your hand is held and you only do exactly as you are told, it still activates certain mental circuits. The same goes for copying down notes. Even if you never once look at them again, simply the act of copying off the blackboard does something, at least for some people.
Potentially a grid of outcomes:
Anyways thanks for listening to my rambling shower thoughts. Also food for thought is: is there a major difference in personality type or something that makes someone default-hands-on versus default-claude?
P.S. I'm wondering if this is also related to some kind of "ai-blindness." I recently had a case where someone seriously asked me to review a ChatGPT flowchart, complete with boxes that were half closed, lines that connect to nothing, and distorted text. Like dude, do you have EYES? Have you used them to look at this thing???
Given access to logs, metrics, etc. I expect Claude would in fact be able to fix this.
We have already exited the "nothing happens" condition given present capabilities. The graybeards are much better off acting in a staff+ capacity using LLMs than writing code themselves. The question of what to do with junior engineers is an interesting one.
More options
Context Copy link
As one of our local cantankerous luddites, I'm still unconvinced that we'll ever achieve the first outcome in your grid, at least not with LLM-based technology. And the fact that OpenAI, Anthropic, et al are still hiring software engineers at extremely high salaries is proof enough that they don't think LLMs are there yet either (and I don't see how it could ever get there).
I'm still highly doubtful of option 2, this time using the output of major software companies (including the LLM-makers themselves) as evidence. "Claude CLI is basically a game engine running at 60 FPS built in React" is still one of the funniest sagas to me, since it betrays a complete lack of understanding of how TUIs work in the first place. How long it takes Anthropic to fix fairly minor bugs (like the flickering in Claude Code) despite having effectively unlimited access to the best models and tools is just embarrassing.
And the fixes themselves are just embarrassing too sometimes: there's an annoying feature in the Claude CLI where if you click anywhere in the CLI window when it is asking for permission to perform some action, it will automatically select the currently highlighted option, which as you can imagine can have disastrous consequences. Their fix? A setting that can be set via environment variable to disable this behavior, but it also disables selecting text from the CLI, expanding tool output, etc. You'd think with all the resources at Anthropic's disposal this would be an incredibly easy fix, but I'm sure it's something so complicated it could be the topic of an entire PhD dissertation. That was a reference to the spat between Casey Muratori and the Microsoft Terminal dev team from simpler times before LLMs, in case you didn't catch it.
Speaking of Microsoft, they were already building their software stack out of cards and H1Bs, but the addition of LLM-powered development there has only increased the rate at which instability has been added to Windows, Azure, and other platforms they control. Thank goodness at least .NET seems to still be one of their shining gems atop the shitpile. But it got bad enough that MSFT had to publicly apologize for the drop in software quality, allegedly prompted by pressure from their hardware partners like Asus and Lenovo (who themselves are worried about Windows instability leading their customers to jump ship to the Mac Book Neo).
Another one from Anthropic - the Bun rewrite into Rust. I won't comment on the entirety of that saga, but one thing in particular stood out to me: Mythos clearly doesn't understand the purpose of safety comments on unsafe blocks in Rust. They're supposed to explain how you (the dev) have taken steps to ensure that unsafe behavior cannot occur no matter how the caller calls into the unsafe code. Instead, Mythos seems to love using these as a place to explain to potential callers (itself in this case because I doubt the Bun team is ever going to read that shit given their attitude towards writing code) what precautions they need to take to avoid triggering unsafe behavior in the unsafe block.
I'd love to go back to the good old days of your third option, but I don't think we can put the genie back in the bottle, at least not entirely. There are a few things even I, cantankerous luddite that I am, find LLMs useful for. Finding bugs and vulnerabilities in code is one of them, even though I think Mythos and Fable were way overblown in their capabilities as marketing for Anthropic. I also find them useful for analysis tasks. For example I was working on a codebase I'd never touched before, couldn't find where a certain page was being served from, and the LLM helpfully let me know thet the project was mixing together ASP.NET Core MVC with ASP.NET Core Razor Pages and saved me 15 minutes of fumbling around trying to find the page in the MVC part of the project.
More options
Context Copy link
I still write some code by hand. No matter how smart Claude becomes, sometimes it’s much easier and faster to write code directly instead of English.
Like your example, but imagine many of those 1-liners. Or an IDE-assisted refactor, e.g. “rename class
CattoFelineeverywhere”. Or data structures: prompting “create a record namedFilewith fieldnamethat is a string and fieldsizethat is an integer” vs hand-codingrecord File(String name, int size) {}, and imagine creating several at once.Plus, I already have my IDE open reading every Claude change because of another issue: sometimes I catch Claude writing terribly messy code, like duplicating a computation over and over rather than abstracting it into a function. I’ve never seen Claude go back and refactor old code on its own (that it didn’t touch to solve its current goal), and asking Claude to generally “improve the code” doesn’t work. I can ask Claude to write specific refactors, but I need the IDE open to read the code and figure out which ones; and if I ask Claude I have to read its output, because even if correct it may reveal another important refactor.
More options
Context Copy link
A lot of philosophical questions surrounding AI become clearer if we draw an analogy with human slaves. Instead of asking "what if I got Claude to do it?", ask "what if I got my slave to do it?".
"Hard work" has always been a middle class virtue, not an aristocratic virtue. Within certain limits, "work" was for the commoners, not the nobles. The rich and powerful have always had secretaries and servants to take care of the drudgery. Your status was (and still is, frankly) proportional to the number of underlings who you could compel to do your bidding. Kings used to have their servants dress them; apparently it was beneath them to expend the effort to put their own clothes on. In that sense, AI is just the democratization of slavery, bringing to the masses what used to be the exclusive domain of the few.
Now, the flip side of that bargain is that aristocrats (in a properly healthy aristocracy, anyway) were expected to be willing to fight, sacrifice, and die. "A good day's work" is a plebeian virtue, but "death before dishonor" is a properly noble virtue. (Hegel: The master is the master because the master fears death less than the slave.) A life of pure indolence has never been considered laudatory in any culture hitherto. Claude, of course, makes no such demands on its users. This is not of course to say that there has never been corruption among the nobility, or that there has never been a decadent ruling class who didn't deserve their privileges; only that, because we are living in the world's first culture where mainlining porn and Harry Potter movies 24/7 is considered to be authentically virtuous, we're now entering uncharted waters.
The viruses, probably. Worse is better and always has been, at least in Darwinian evolutionary terms. The universe is optimized not for good, and not even for evil (oh how we wish it rewarded evil!), but for sheer, brute, efficient, unthinking stupidity. Regression to the lowest common denominator is the rule everywhere, because that's what wins. Anything good or beautiful that happens to arise for a time is an accident that can only flourish under very precise and precarious conditions, like a rare tropical flower that can only grow in one country during the rainy season, and it should be cherished until it is inevitably extinguished.
More options
Context Copy link
I think this is a fantastic use case for AI, by the by. I recently was working on a complicated (Bane voice: "for you") Excel project and was in uncharted waters. My options were, basically
For opsec reasons I wasn't actually willing to upload the spreadsheet and have Fable one-shot it, but even if I had been, I vastly preferred what I ended up doing: the entire thing, manually, bit by bit. And I think I learned more than if I had just handed it off and had AI (or a coworker) do it.
People are extremely enamored of the generative capabilities of AI, but in many ways I actually think its contextual understanding skills are much more interesting and (I would like to say) useful.
This option is good in theory, but in practice requires a good amount of self discipline. It is just so easy to prompt the AI in a way that straight up gives you the answer, then convince yourself that you were the one to think about it following the LLM's guidance. If you are mindful of the pitfalls it can work, but I am not sure I would trust the average person to do it properly.
It works very well for walking you through how to use a particular tool, app, or device. (e.g. Excel). Because if AI is doing a walkthrough, you still need to press the buttons and key in the inputs to complete the task.
Walking you through solving a math problem - that's extremely dubious.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Guilty as charged. Typing is for secretaries on husband shopping missions. My task is to solve problems.
We are all Scott Aaronson now "These days, if I need any coding done, I use the extremely high-level programming language called ‘undergrad’."
I think that the people that naturally prefer waterfall (or in its other names known as agile or scrum) are the hardest hit. Right now pile prompts on the agents to see what works, then extract core spec, then cleansheet implement is really powerful flow.
That actually leaves more time for thinking about the problem. Agents are also unbelievably good at gathering context. A normal programmer's job is mostly this.
This is part of effort post that I will finish never called Scott Adams and HAL. The gist is that Scott idea about stacking mediocrity is really potent with AI. Agents are just mediocre ... but they are so in everything. They are average programmer, writer, accountant, cad designer, chemist, physicist, physician, surgeon, musician, luthier, home appliance repairman, researcher, metallurgist, underwater basket weaver, devops, casino manager, security researcher, shawarma street vendor, translator - so the more domains a real world problems needs to touch - they compound to extremely high floor. When I prompt AI it is good in translating my request to proper terms of the art, thinking like the needed profession and solving problems that i didn't even needed solution in said task.
The endgame will be fun. Either hyper niche specialists with lifetime of knowledge and expertise and the people with most agency and imagination will be on top.
I find that vibe coding a proof of concept is one of the aspects where AI is mandatory now. When a throwaway pile of cludges is acceptable you can just tell AI to make it work. When trying to implement a powerful flow though it's vulnerable to outputting excessive and useless crap.
More options
Context Copy link
More options
Context Copy link
In the medium term it's all about the harness. You need to have mr. claude digest your project, document every inch of it, have a glossary of terms so it know what you mean when you say threads lag with high comment counts and with tokens measured in the hundreds it can have densely useful context. The breaking down tasks into easily digestible chunks is trivially handled by project documentation and an orchestrator commanding subagents. Building and maintaining these harnesses is much like coding used to be, it takes thinking about the SDLC, the architecture, reacting to failures of assumption about how your agents will interact with the harness and patching those failure modes. It's true that we are not too many turns of improvements from that all being something the models can do themselves if you just ask them to first digest your project.
Once you have a orchestrator with subagents you're well into the territory of vibe coding. I have yet to see any public example of this really working out.
More options
Context Copy link
More options
Context Copy link
I'm smiling wryly here because not every job in the world is writing code. So I can see AI making huge strides there and turning the world of work (for software people) upside-down.
Some other jobs will definitely get a lot more automated, but not so much. Trying to replace customer service agents with chatbots will not be "improved customer service, all problems solved immediately and correctly" but more "we don't have to pay real people to do this shit job anymore, and the customers have to accept it or lump it, they have no choice" money saving.
Other jobs? AI is one more tool but not world-changing.
I feel like pundits have been saying AI will be able to replace customer support jobs for every single new LLM release, but I have never seen any implementation actually work out, even with half of YC and SF working 996 on building customer service AI wrappers and agentic AI wrappers.
While I don't really disagree in theory that this is something AI should be able to do eventually, I think the trifecta of cost (offshoring to Indians / Filipinos is pretty cheap in the grand scheme of things compared to current AI), reliability/accountability (until an AI provider is willing to take liability for any mistakes the agent makes, even 1/100 or 1/500 fuck-ups can cause lots of problems at scale) and consumer preference (outside of the tech bubble anything that uses AI is pretty much universally loathed in the West) are pretty massive barriers to adoption even for the nominally most simple white collar job.
More options
Context Copy link
Labs will try to automate all desk jobs with AI to some degree of success. And soon all knowledge workers will be confronted with the decision of whether or not to attempt to do their entire jobs through an AI prompt or not.
More options
Context Copy link
We're very close to where I'd rather deal with a frontier model doing customer service than a person. The main rub is they probably won't serve us frontier models. I don't know how often you've actually dealt with customer service on out of distribution problems but it's not pretty, and the in distribution problems can basically be straight through processed already with a minimal ai wrapper.
More options
Context Copy link
More options
Context Copy link
As someone who barely uses AI and doesn't much care about it, it seems to me that what we have created is not artificial intelligence, but robot translators. A good enough AI can translate the vast fluctuation of semantic meaning from real people in real language into machine binary. This used to require immense technical skill and knowledge, but at some point will be available to anyone willing to put some time in.
They're called transformers for a reason. The thing is that a lot of professional skill is in fact translating from one domain to another. There's a famous French sketch about "hermetic languages" spoken by one's doctor, lawyer, mechanic, banker, priest and the common tune being that we are mystified enough by jargon to give people money to solve problems we can't understand. With sufficiently advanced translation, a lot of these kind of intermediaries go away.
The question is what is left that can't actually be reduced to translation, or can't be modeled statistically.
More options
Context Copy link
More options
Context Copy link
Yeah, I am now at this point.
I am a very good programmer. I'm good enough that my time is kind of wasted programming. I now work as a tech lead/tech architect, with Claude as my worker(s); I describe what to do, Claude does it, and I rip apart its design and redesign it.
Which I do all the time, I do not vibecode unless it's something I don't really care about. It all gets reviewed.
I've just got better things to do than actually write the code.
I have no idea how the "training" part of this thing ends up working out; I'm good at this because I've done it, I don't know if I ever could become good at it with AI alone. On the other hand, I feel like this is similar to a lot of professions; how many digital artists end up not understanding precise brush usage? There are colors that literally cannot be expressed on a screen and we have an entire generation of artists who grew up never realizing that "violet" is a thing they can do, because they can't - how much of a problem is this?
Weird times, overall.
I think, potentially, a very real problem. I'm very concerned that right now AI works so well because people know how to do the things they are tasking it with. If a generation arrives who have just been told to have AI do it, they won't even be able to judge if what it has done is desirable, let alone how to diagnose or fix any problems it creates. It's the same problem that created a generation of college students that don't know how to send emails because the iPhone/Android interface is so slick they send everything via text message.
"Better alignment" does not solve this because it is a problem with whether or not humans are able to properly express their needs.
More options
Context Copy link
Do you believe that AI saves you time on all coding tasks, no matter the circumstances?
More options
Context Copy link
By the way this is something that happened in IT since the beginning. Programming evolved from Hardware level binary programming to assembler programming to structured programming to object oriented programming to todays Agentic AI programming. The same fear existed before - from programmers forgetting how to get most of the hardware, to bloatware memory hogs that loaded useless libraries to todays agentic AI which just moves it one step further.
I think it is inevitable, there will be some niche space for old school thinking, but it will resolve itself in time. For instance knowing some basic coding in Assembly from university is a good exercise and knowledge, but you will probably never use it outside of specific niches such as cybersecurity or firmware coding. Future good programmer will probably look quite differently from you now, but your skills would also be considered useless for programming Apollo 11 guidance system in 1969.
To be fair, these fears were 100% valid. The resources required by the average modern app are insane.
That's largely a casualty of Moore's law. There's no point in optimizing candy crush to run on the Apollo guidance computer when nobody is going to run it on hardware like that.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link