Yeah, but "reason" is not a well-defined term. It means different things to different people in different contexts. It's just marketing speech. You can easily argue that all ML models, even those from 50 years ago, can reason to some extent.
I am sympathetic to that however this is also a company that consistently promises the moon. Their CEO recently put out a blog post about AGI in a few “thousand days”. They clearly do have a hype issue and it’s fine to point that out.
Honestly... SWE management already is kind of already out-of-date and I say that as someone who is a SWE manager.
Most software teams can pretty much run themselves and the only real use a SWE manager has imo is running interference because the rest of the company are morons and/or chaotic and being the gopher for things the team needs. I think a manager who isn't also participating in the work should be removed as they are a net-negative.
If the rest of the company is calm and sane, in my opinion, you could have a team with senior+ ICs that take turns doing planning or doing a collaborative planning.
Damn honestly why is this downvoted at all. Very anecdotal but I tried out the new o1-preview available to me, and it did not produce anything noticeably better for my domain. To claim this is “reasoning” to me seems infuriating
As others pointed out he seems to be talking more about about automation which... sure that's a fine discussion to have, but what bugged me more than anything is the overselling of Waymo/FSD. I understand a lot of this is also on a spectrum but it seems a bit irresponsible of Karpathy not to mention crashes Waymo has faced or other problems FSD systems have faced. It's not just an issue of scaling up, sensors, etc. There is more engineering work that needs to be put in clearly. It's fine to bring it up in his example of reactions to economical forces, but let's be completely honest about the whole thing.
It's not a god given right, but I worked really hard to get to where I am in life my visa is tied to my profession and I have a feeling it's going a way soon. Which means my visa might go away soon. I've gone from contemplate suicide to how to carry on. What you might think is awesomeness is dread for others. Just to point out there are 2 sides to this coin.
I actually don’t know what I should do. How I should prepare, what the future holds for me or us. I worked really hard to get where I am and it feels like it’s going to all crumble soon, and people who are more positive about things tell me either UBI will somehow fix things, and humans will persuade what they want or that we will all be in pain together so I won’t be alone. That doesn’t help. It’s given me actual nightmares.
Humans will not tolerate the condition I describe. Inequality can only get so bad before it becomes intolerable. Learn to do something with your hands, it may well be valued in our lifetime -- and if the future is brighter than my dour prediction, the hobby will have been a salve which still has its own value.
Every time I see these articles I get more and more worried about my employability to be frank... with no backup plan and way too much time poured into learning software engineering. It's not looking good gang.
I'm genuinely not worried about that. As others have observed, us programmers will be in trouble when regular people learn to produce a product spec that's detailed enough that an LLM can create the working software that they need.
That sounds a lot like programming to me!
I expect our work will change: we'll be able to spend more time thinking about what we are building and less time typing code on our keyboards. But if anything, we're going to become more valuable - because we'll be able to get a whole lot more done.
Not having a detailed product spec is a much, much bigger problem when you can't change the product to fit your needs on a whim. Beyond that, there's a few layers between "software developer" and "regular people" that complicate your prediction.
While most developers don't hold interface designers in particularly high regard, honestly ask average people what they think of most developer-made interfaces and you'll find out the skill is a lot trickier than it seems. Most developers only think they know what developers do, but they're like many corporate workers see everyone from the head network architect to the community college desktop support intern as "computer people who can fix your email." Most think interface designers primarily work on aesthetics, but most probably aren't even invited to meetings about branding/visual aesthetic/etc...) Going further, many of those designers are savvy enough to write some basic code and likely bodge something into place, especially if they're using some kind of purpose-built interface that can handle things like data model consistency between builds.
It's a very unpopular opinion around here, but I think designers using the next generation of no-code tools will eat front-end and simple app developers' lunches and I think it will happen really soon. I'll bet teams at Wix, Webflow and other no-code authoring tools are working like mad to develop these tools right now, and I'll bet that's a hair's breadth from automatically generating electron apps from whatever users make there. If your specialty is code, and there are people with whole other useful skillsets that could passably approximate that capability with a few occasional hours from a contractor, the developer isn't going to be the one that still has a job.
While the demand for developers is still large, it's not infinite, and I look askance at assumptions that there won't be a really painful 'adjustment' for a lot of working professionals.
> It's a very unpopular opinion around here, but I think designers using the next generation of no-code tools will eat front-end and simple app developers' lunches and I think it will happen really soon. I'll bet teams at Wix, Webflow and other no-code authoring tools are working like mad to develop these tools right now, and I'll bet that's a hair's breadth from automatically generating electron apps from whatever users make there.
I'm sure Wix et al are indeed working on that. Looking at their website, they've been at it for 17 years and have a whole bunch of high-level things already available — meaning "blog" and "eCommerce" and "scheduling", not merely "fancy checkbox" or "date widget" — which will cover the needs of a lot of small businesses who neither want nor could afford a single full-time web dev for even just 3 months.
Nevertheless, I disagree with your core claim: I learned how HTML worked partly by reading the output of Dreamweaver[0], followed by teens working with REALbasic (now Xojo) and then VisualBasic. I also spent most of the last decade using Interface Builder in Xcode (with 1.5 years of "why did you do this in C++?" and 1.5 years of "why won't Xcode render this SwiftUI like it's supposed to?").
I don't understand why the visual editors aren't more popular in our profession — they seem like they should be perfect; SwiftUI and React shouldn't need to even exist — and it's that gap between how I think the world ought to work and how I observe it actually working that contains the real problem that means so many of the jobs we do, aren't with the no-code tools we already had 25 years ago.
(Of course, one possibility is that Upton Sinclair quote: “it is difficult to get a man to understand something, when his salary depends on his not understanding it.”, and perhaps we're all just in a bubble of belief that can collapse at any time…)
[0] I think. Something like that anyway, free demo on a Mac[Format|World|User] magazine cover CD.
I'm not totally sure, but you seem to be conflating knowing how to design interfaces and being able to code them? They're not even in the same career trajectory. I could be misreading what you meant.
That’s assuming frontends need to be ”developed” at all in the future. There’s a possibility intelligent chat bots able to generate microfrontends on the fly is going to eat our lunch, but what is really going to eat the lunches is a failing capitalistic system due too lack of demand for labour. I guess overall society is going to balance things out somehow, curious if it’s going to be a in a civil manner or brutal disruption.
Well, no the point to my comment was that they won't need to be developed, and that the real skill will be in knowing how people interact with things so you can tell the machine what to make, and that is absolutely not a technical job. Even in the interim, incidentally gained technical skills will suffice rendering the technical people in the chain redundant. Surely the human understanding component will be whittled down at some point, too, but it's a much much further goal than automating essentially mechanical processes.
And, of course-- all of the "unpleasant" side effects of massive innovation would be totally avoidable if we didn't treat people as disposable labor units worth nothing more than their market value.
>Surely the human understanding component will be whittled down at some point, too, but it's a much much further goal than automating essentially mechanical processes.
Why do you think that?
People used to think that making paintings and poems would be one of the most difficult things for AI, and look how that turned out.
I don't think the above comment disagrees that much with you. As far as paintings and poems go those are also pulling off of patterns that humans naturally have given off too right? An LLM is a bit more than a stochastic parrot sure, but it's also a large part of how it functions. The real surprise, or maybe not a surprise, is how formulaic everything is.
Because interfaces require situation-specific reasoning that purely expressive art can be imitated without. And frankly, I think the slick looking images that AI spews out from actual artists' munged up work is a far cry from being equivalent. Like interfaces, the people who wouldn't pay an artist or designer to begin with probably don't care enough about the quality of the art or interface for it to matter. Cookie cutter applications and stock images are going to suffer, but developers making basic crud functionality or simple API interactions will suffer soon after.
I can only see brutality personally. I’m not prepared. Once I lose my way of making a living I think it’s done basically. Time will tell, but it’s never been kind before, why now?
Many here seem to think anyone "smart enough" or "hardworking enough" could simply pivot into something else after their entire category of employment was decimated. I assume that:
a: These people are very young and don't understand what it takes to invest 3, 4, or 5 decades in a career and/or despite their assumptions about their life experiences, have never actually experienced significant hardship.
b: They're neurologically or emotionally incapable of empathy or lack a usable theory of mind.
or
c: They've read too much inspirational linkedin hustle porn about people pulling themselves up by their own bootstraps (which is actually supposed to be a joke-- it's obviously impossible to pull youself up by your own bootstraps, but for some reason people repeat it without considering its true meaning) and think if they are tough and ruthless enough that they'll be one of the ones on top. Which is kind of sad.
Or a variant of a: they're forgetting how little their first jobs paid.
Even in most lucky case, pivoting your career into an entirely different category means reverting to entry-level pay, while your age, health and obligations remain the same. Imagine just the paycut alone happening to you, out of a sudden. And that's the best case for what people displaced by AI will experience.
As someone who spent quite some time as a young person with the fringe end of society, I can assure you that saying you'll pivot into crime and drugs is like saying you'll pivot into being a plumber. It takes time to build a career that pays more then entry level money consistently, and good luck developing a network of connections to make it happen. It's not like walking down the street to pick up an application at Chipotle. And AI is probably going to take individual computer criminals' jobs even before most other people's.
Yeah, but a lot depends on where they live. From what I heard about the US penal system, someone living there would be better off being homeless than jailed.
That will never happen because business changes requirements on an hourly basis :)
Jokes aside, I think you kind of need to know how to problem solve in code to even know in which direction to go.
If I tell it, create a Facebook - the limit will kick in. But most people do not konw how to destructure Facebook in terms of functionality and what components are.
That's fair and I hope that is the case. That might even be fun, and I mean I actually have a lot of fun with gpt-4 now. My bigger complaint there then is that you need more money to be competitive in this industry now? Until LLMs become smaller more efficient? Like back in university the best student was a Russian immigrant who learned back in Moscow on pen and paper, no compiler to mostly handhold.
It depends if the 'skill' of programming/creating shifts more to the LLMs vs programmers. If so market capitalism will see programming commodified.
If not, and LLMs just add leverage to what programmers/creatives do, it'll move more towards a sports/entertainment model (winner takes all, or most), with probably far fewer professional participants.
I think most utility software projects won't even require this three step process. More like "define problem, use an AI-assisted interface to whip up a few solutions, spit shine your favorite, if necessary."
LLMs seem very well-suited to generating written specifications, like they do with writing the code itself. They can even help with the "define problem" part.
Steps 1 and 2 are more reliant on human-to-human communication if you are developing an application to be used by some end user and not something that just ingests data and processes it. It takes skill to translate non-technical speak into a clear goal and specifications required to achieve that goal, to probe the third party and ask specific questions to tease out their expectations of the application. That, LLMs can't do yet. I have no doubt that they'll get better at that in the near future though.
Ah, requirements capture.. yeah, not straight ChatGPT but "having a directed conversation" doesn't seem like a stretch.
EDIT: for example, Character.ai as an example of character prompting. You can prime an LLM with a prompt so that it asks you certain types of questions.
> I expect our work will change: we'll be able to spend more time thinking about what we are building and less time typing code on our keyboards. But if anything, we're going to become more valuable - because we'll be able to get a whole lot more done.
That would be lovely, but not how efficiency works in a capitalistic society. We can produce cars much faster in a factory then we could in the past due to various factors including automation. This doesn't mean the workers get to take their time and a do a better job. Instead it means they are expected to have a higher volume.
This is what I worry about with the rise of LLM programming (especially as someone who is less than impressed with the actual output I've seen). It's not that I as a software engineer will be replaced by someone using GPT, but rather it's I'll be replaced by another software engineer who doesn't care, using GPT to produce 3x as much "software" at half the quality because the only metric that some will look at is volume.
