I think part of it though is that LLMs (especially in novice hands) create code that is more difficult to review and has subtler bugs. Code review is (for many) the least fun part of programming.
Like imagine if your hobby was cooking, but you hated the cleanup. If you buy a robot that automates the fun part at the cost of making twice the mess, that is not a good trade off no matter how many extra cookies you end up with.
Or imagine a club where everyone cooks together and then cleans dishes together. Then one day some people show up at a club meeting with cooking robots that create 10x the amount of dirty dishes and then don’t help clean up at all. I bet the other members wouldn’t be happy, even if the robots mostly cooked decent food.
Then robots begin cleaning dishes too. And the other members aren't happy, because what's the point, we are here not for the dishes, but for the process.
> Code review is (for many) the least fun part of programming.
I'm lurking around the OpenBSD mailing list and it seems that code review is where collaboration starts. You send a patch, explain its purpose, people try it out and comment. If it's OK (various factors decide that), it's imported inside the codebase.
The main issue in corporate settings is that the author of the PR wants to browbeat you into accepting his/her changes. Either directly or by appealing to the manager. They don't want collaboration, they just see you as the blocker between them and the "Done" status on the ticket.
I used to work at a company that did in-person pair review. I wasn’t an engineer there but the engineers all legitimately enjoyed the process. They hired very capable people as a default, so it was watching two passionate and talented engineers debate.
I think that also hits what you see universally with every AI product trying to pass worse results off as a benefit: if you’re a business, the prospect of cutting your cost per cookie is really attractive as long as the decline in quality doesn’t cause people to stop buying and that’s going to get the focus of everyone who sees the financial reports more than they talk with customers. Anyone who’s going it because they care about cookies or take pride in your craft will have the opposite weighting.
We’re kind of at the fast food point: before LLMs, bring successful at all in software required some level of skill (witness all of the companies getting very expensive write-offs from off-shoring or big consultancies trying to save on payroll vs. staff), kind of like how the small cafe/bakery model needed to have at least some level of craft to have something people would buy before the option of making food at a factory and shipping it became an option.
It is, although in this case the cookie-making robot also makes 10% of its cookies with sand instead of flour. So it's still not a very good trade if you have any pride in your business or consequences for bad product.
> ADHD wins over autism in the immediate future. It was balanced in the past for development to need both but now it’s a total hyperactive victory of general knowledge, master of none.
That’s an interesting framing. I’m not sure I agree with your strong conclusion but that’s mostly because I think there’s a second axis about maintenance – you can make a grand strategy game in a couple months but it’s harder to make it balanced, fun, and to keep shipping updates which players like. A prototype or hobby experiment doesn’t have that problem nearly as much and I think that’s where we see such bimodal reactions where some people think it’s a huge productivity boost and others see modest or even negative results.
That's right. Clearly juleiie has never actually tried to ship a game, and is just speculating without any actual experience, which is fine for fan fiction, given they have free speech and all, but totally detached from reality and not in any way useful to people actually interested in creating and shipping games. That kind of uninformed opinion is just a distraction, AI hype, and cheerleading, not anchored in any actual experience or practice.
This is juleiie's laughable comment that illustrates my point exactly, and is a textbook example of the Dunning–Kruger effect, making brash overconfident claims without sharing the title and URL of any games they've actually made and tuned and finished and shipped and supported:
>Good luck making a grand strategy game in two months holidays without an LLM. Now you can.
Yet you didn't, juleiie. The term for you is "Armchair Architect".
>One who can solve all the software world's problems from his armchair but considers actual implementation to be out of scope. Usually identified by a casual dismissal of real-world problems that many people have struggled with. As well as a dismissal of everything that anyone else has actually built as crap.
>They have lots of seemingly great ideas that will be all things to all people but they avoid testing them with anything as crude as reality or running code.
Chris Trottier, who actually has shipped games like The Sims, describes the concept of "Tuned Emergence" and "Design by Accretion":
DonHopkins on Jan 30, 2021 | parent | context | favorite | on: Factorio 1.1 stable
The Sims (the original version) started with around 4 developers during its first 6 or so years, and it took quite a long time to develop. And yes, the team was quite tight and focused, with excellent "high" quality communication (in that we smoked lots of weed together). More developers and artists and designers came in towards the end, after EA bought Maxis, but the original version had a much smaller team than any of the subsequent version.
It pains me to recall how crude and primitive it was even a year and a half before the release (check out the awful flamingo and telephone and fish tank graphics, not to mention the characters and animations themselves, like pot-bellied Archie Bunker always holding a cigar in his hand, and the naughty scenario description text @ 40:44) -- a whole lot of content and user interface and tuning was done at the very last moment:
The Sims Steering Committee - June 4 1998. A demo of an early pre-release version of The Sims for The Sims Steering Committee at EA, developed June 4 1998.
>Chris' honest analysis of how and why "the gameplay didn't come together until the months before the ship" is right on the mark, and that's the secret to the success of games like The Sims and SimCity.
>The essential element that was missing until the last minute was tuning: The approach to game design that Maxis brought to the table is called "Tuned Emergence" and "Design by Accretion". Before it was tuned, The Sims wasn't missing any structure or content, but it just wasn't balanced yet. But it's OK, because that's how it's supposed to work!
>In justifying their approach to The Sims, Maxis had to explain to EA that SimCity 2000 was not fun until 6 weeks before it shipped. But EA was not comfortable with that approach, which went against every rule in their play book. It required Will Wright's tremendous stamina to convince EA not to cancel The Sims, because according to EA's formula, it would never work.
>If a game isn't tuned, it's a drag, and you can't stand to play it for an hour. The Sims and SimCity were "designed by accretion": incrementally assembled together out of "a mass of separate components", like a planet forming out of a cloud of dust orbiting around star. They had to reach critical mass first, before they could even start down the road towards "Tuned Emergence", like life finally taking hold on the planet surface. Even then, they weren't fun until they were carefully tuned just before they shipped, like the renaissance of civilization suddenly developing science and technology. Before it was properly tuned, The Sims was called "the toilet game", for the obvious reason that there wasn't much else to do!
I think we’re on opposite sides but this is the first time I’ve seen someone in the wild share my suspicions that the LLM conflict is part of some vast ADHD vs autism hyperwar. At one point I was collecting posts I found where LLM boosters explicitly mentioned how it interacted with their ADHD.
I'm pretty skeptical of people who are ADHD and claim to be getting boosted by LLMs.
It's hell for my ADHD, personally. It's an infinite rabbit hole generator and distraction machine. I have to use browser extensions to filter out stuff like Stack Exchange suggestions for interesting topics, too. I have had to block Wikipedia in the past to avoid rabbit holes there
I'm better now than I used to be but I am skeptical of people who claim to both be good at multi-tasking and also have ADHD. My experience with ADHD is that multi -tasking is not a strength, it's a sign of distraction
> ... multi-tasking is not a strength, it's a sign of distraction
Interesting point of view from one who's been there. Intuitively the comment makes a lot of sense.
In a way human multitasking resembles how single-threaded computer programs work. That is, we focus on one task at a time but tasks can be interleaved to give an illusion of doing several tasks at the same time. Managing several tasks requires remembering the next step to do when leaving one task to attend to another.
That suggests ADHD people lose track of where they left off when performing several tasks. Distraction plays a role, but I think problematic working memory is the major issue. Among ADHD researchers there's a saying, "people with ADHD know what to do but don't do what they know." ADHD task performance may improve with proper treatment.
> That suggests ADHD people lose track of where they left off when performing several tasks
Not quite. At least not for me. My problem is hyperfocus. If I switch tasks I might go down a rabbit hole and forget to change back. It is similar to being distracted, except the distraction can still be legitimate work stuff. My focus is like a freight train, it's hard to change tracks
Which can seem like an attention problem when you're trying to change me from focused code mode to "jump on a call" mode
Yes, hyperfocus is indeed a frequent problem for people with ADHD. One version is getting overly involved in irrelevant details, due to interest or having difficulty pulling away. The latter takes much effort and a good measure of courage. Maybe "rabbit holes" are kind of like that.
For many with ADHD starting tasks is hard and resuming a task can be just as hard. Constant experience of "starting from scratch" is burdensome, and contributes to inertia and inefficiency. More than not, a tough problem to solve.
I've long thought disability associated with ADHD is generally underappreciated despite decades of study proving the point.
Self-observation isn't easy. I always considered it a sign of growth when people shared what they've seen.
Not diagnosed, on online “tests” I score just shy of the ADHD threshold, and I find myself in this. If only I properly remembered and agreed on what I wanted to do yesterday it would be fantastic. Even if I write it down and happen to remember to check the notes, it’s almost like I was a different person yesterday, todays priorities are all new
> Not diagnosed, on online “tests” I score just shy of the ADHD threshold, and I find myself in this.
Don't know the tests you refer to, but it's recognized that many adults have ADHD symptoms not quite meeting DSM-5 criteria for ADHD. It's called "subsyndromal" ADHD which causes as much disability and responds to treatment just as well as "full" ADHD cases. IOW it's absurd to regard someone with 6 symptoms as belonging in a different category vs. a person having "only" 5 symptoms. Anyway, ADHD is a condition that usually responds to treatment, no reason not to check into it.
Thanks, I know I could probably get a diagnosis and treatment for it. However, to be honest I kind of like it!
I think it keeps me on a different path wrt most people, so until it becomes a serious problem (financially) I’ll keep riding the waves of focus and distraction ;)
Of course, it's always entirely up to you. But wise to stay alert to documented ADHD effects, e.g., on social functioning, general health, increased risk of accidental injury, depression, anxiety, to name a few. In any case, treatment won't undo the uniqueness of your path, if anything likely to augment reaching goals your path leads to.
For me it's a dopamine machine. I get my doses of lacking dopamine from completing things.
Agents let me complete things in parallel all the time.
They also let me build tools that manage all the attention-grabbing things in my life. Something I've been meaning to do "tomorrow" for a good decade.
I've already built custom notification tools for all the "I'll just quickly check hn/lobste.rs/tildes/whatever" so that they'll tell me if there's something actionable, closing down the rabbit hole for me. My email is pre-filtered by an agent that pings me if there's something urgent. My IRC client is custom-built for the specific way I use it etc etc.
The contest you described (people with ADHD vs. autism) would have an enormously negative effect. But there are reasons to think it's unlikely. FWIW at least 25% of ADHD adults manifest autistic traits. Hardly surprising, clinicians regard overlapping neurodevelopmental disorders as the rule and not the exception.
An online community works when participants learn and share knowledge gained. Contests, with self or others, disrupt learning and sharing, hence by definition constitute "bad behavior". If it exists, the hyperwar you allude to will harm communities more than all LLMs combined could do.
No, it felt too invasive and creepy so I abandoned the idea. Or my lack of follow-through is a sign that I’m actually on the other side of the “war”. Reader decides! But suffice it to say that it was, to me, a significant trend especially around the “Claude Code moment” back in February/March is this year.
Autism should not be framed as being "opposted" to ADHD. The correct axis is "Autistic" (Neurosis) as opposed to Psychosis. Opposite poles but with a little bit of horseshoe theory dynamics
> This way one person can implement a whole new hobby project instead of only writing a single specialized part in an existing project.
I still do start from bottom up to refactor everything into workspace members and this way, parallelism is way easier
> LLMs reward seeing forest instead of trees.
Not just forest. LLMs reward seeing structures and designs, and it still requires a lot of creativity to design it well. I have to do a lot of architect works for that, like package layout, testing strategies (BDD vs TDD vs E2E), setting up access to platforms and digging out what strategies to apply...etc.
> People focused on the trees will however lag behind as implementation details matter less and less.
Trees like demonstrations and example code are even more important. This is because those are literally post-training materials for the LLMs
> Specialists will be fed to the model weights
I don't think so. New labor laws are certainly going to be introduced by the politicans to tackle this.
> generalists will inherit the world.
Now the problem is how far do you want to stretch the systemic thinking, and you also get burned out quicker.
Funny you should mention autism. I'm of the belief that if I really wanted to, my autism could help me become a very effective AI-assisted programmer, because autistic communication traits that actual humans find annoying (and which most people hence train out of themselves), like infodumping and using precise language, are things AI actually loves and produces better results with.
But I don't want to use AI to code, in large part because I iterate on ideas by working directly with code. Attempting to program through an LLM intermediary becomes like attempting to eat using soft silicone chopsticks.
Like imagine if your hobby was cooking, but you hated the cleanup. If you buy a robot that automates the fun part at the cost of making twice the mess, that is not a good trade off no matter how many extra cookies you end up with.