If you are a software engineer, this will output your productivity ten fold on the upcoming years. Now you don’t need to hire junior devs and can just build the product of your dreams with very limited capital.
In my opinion this technology will be as democratising as the YouTube’s early days.
Instead of worrying, learn to work with it. It will be harder for large companies/large teams to extract value from this compared to small companies/small teams.
It means competition between companies will increase but it isn’t necessarily bad for existing software engineers, especially solo founders.
This 10x productivity absense of a 10x expansion of programming industry (which is very unlikely) translates to less developers in general, including senior ones. Even more so in an economy like this...
>It means competition between companies will increase but it isn’t necessarily bad for existing software engineers, especially solo founders.
"Solo founders" is what? 1/10,000 of working programmers? And they're absolutely not the ones people worry about regarding GPT replacements...
I think I disagree. If software then becomes 10x cheaper, a lot of use cases that used to be too expensive to build now becomes affordable. At my own job, I think we could easily do 10x the business, because our customers need tons of tooling (for example for energy transition) but we don't have the people (among other problems).
Worst case scenario is that it gets SO good at writing code that software engineering teams are severely downsized or are made obsolete altogether, and I find myself out of a job. I’m not expecting UBI to start falling out of the sky any time soon, especially while there are still manual labor jobs that robots can’t do.
Alternative scenario is that individual developers get somewhere around a 2x-5x productivity increase, but why would I want that? That doesn’t give me more free time - that just means I’ll be expected to do more work. Non-technical management already expects ridiculous delivery timelines; now I’ll have to deal with them asking “why can’t you have the whole project done by tomorrow? Why can’t you just have the robot do it?”
It’s a lose-lose situation and none of us asked for this.
If you're writing react/python/angular or something popular it seems to do amazing things and spit out entire websites (per demos).
Unfortunately, when I try to put together C++, Rust, or even C# using recent libraries like Blazor it chokes up. I fully understand at least one reason why (libraries and language features not being in the training data from 2021) but that makes me feel that perhaps software engineering at the cutting edge or niche is safe and still requires human reasoning. Not to mention things like properly understanding when and why to use certain data structures, real-world impact of coding choices, pricing, esoteric speed/efficiency improvements, etc.
I think there's still a broad general area where good, great, and amazing+ developers can operate without much threat and in fact using their knowledge and experience to leverage GPT-4 (or others) as a force multiplier.
It'll be interesting to see what happens when AI truly surpasses human level intelligence, as in, being able to completely replace human jobs, but we're not there yet. It's likely that when we reach that stage, the world will change dramatically and we will either live lives of abundance and leisure or face extinction :)
As for the latter...I'd say GPT has increased my productivity and therefore allowed me to focus on the more interesting aspects of my work, rather than writing annoying boilerplate code and doing boring tasks where I don't learn anything. I almost never write my own boilerplate anymore.
More productivity doesn't necesarily mean more work. It does mean more focus on interesting work.
>...and I find myself out of a job.
Tell me you are the problem in the industry without telling me you are the problem in the industry.
You are overestimating the vast amount of "software engineers" in the world. The overwhelming majority of us are just programmers, we are just gluing together CRUD spaghetti in the random language we grew up with. We don't care too much about work or a career. And most of us don't want to do more, we want to get a decent salary for our boring work. And we certainly do not want to be "solo founders", build products of our dreams or increase our productivity.
This way of living feels threatened now.
Like sibling commenters, I love the idea of building something new with greater leverage. On an individual level, I'm looking forward to leveling up and finding new ways to be effective in my work.
Unlike sibling commenters, I don't think that should be our only option in life. It saddens me greatly that, given a new option to increase the effective output of a unit of time, we repeatedly choose as a society to profit monetarily (and with vast disparity in who benefits) rather than to give people more options in life than drilling on their jobs.
The industrial revolution promised people lives of relative leisure by replacing the need for much physical labor, but instead we concentrated the benefit to the few—and we keep making that same choice over and over.
Yikes. Productive work is not just a way to earn a living but also a way to achieve personal fulfillment and happiness. It's a means of creating value and contributing to society. A person who works just for the salary and does not find any meaning in his work is not living up to his full potential.
The only problem is that we live in a system that directs the gains upward and any costs downwards, and in so doing creates perverse incentives against people welcoming their redundancy.
I’ve worked with tons of programmers like you describe. I’ve continued to tell them that simple UIs and CRUd interfaces to dbs are solved problems we should not be fighting with.
You mean the same YouTube that routinely ruins people's livelihoods when it closes their accounts with no recourse? Because I'm totally looking forward to the day when that happens to my development tools.
"We detected that you are using our code to kill vulnerable children (aka orphans). This is against our TOS and we have permanently disabled your account. If you believe this was in error please log into your account and talk to our ChatGPT-powered tech support".
Millions of creators grinding for pennies while the lucky ones that got in early and made it rake in the profits.
I think success in tech is going to become extremely pyramidal in the coming years. This is a huge shame, as this was one of the only fields out there where you could make a really good living without going to the "right" school for years and years and years.
10 years from now we might have the equivalent of what today costs 10 million dollars today. Automated farming means what today we consider high end and expensive produce becomes almost free. Automated transportation means that food gets delivered to you for almost nothing. Imagine you had a 95% off coupon on Uber Eats. Does that sound terrible? If so why? Because it also means that Jeff bezos gets a 2000 foot yacht?
Edit:---------
I'm getting a lot of doom and gloom respones. And you all are right, there are a lot of people who do not have food/shelter/cheap colleges. But what you all probably are not aware of is that 100 million people have risen out of poverty in India over the past 15 years. Your word view is being warped by the doom and gloom media. I would suggest reading just the beginning of the book factfulness. It will totally change your view of the world and probably make you much happier.
In my niche(s), I still see new Youtubers pop up all the time that gain large followings and turn Youtube into a full-time job. Sure, they don't all become rich, but many have started earning enough to drive Teslas, so it's definitely not pennies.
Software development, as an employment opportunity, does not have these same dynamics.
But you’re right, the more level the playing field, the greater the competition.
> I don’t get the overall doom and gloom towards LLMs on the software field.
From the second line of your comment:
> Now you don’t need to hire junior devs
Do you need GPT to put the two together? I think it's pretty obvious why folks are freaking out.
Date Weakly General AI is Publicly Known https://www.metaculus.com/questions/3479/date-weakly-general...
Date of Artificial General Intelligence https://www.metaculus.com/questions/5121/date-of-artificial-...
The latter includes this criterion: "Able to get top-1 strict accuracy of at least 90.0% on interview-level problems found in the APPS benchmark introduced by Dan Hendrycks, Steven Basart et al."
The APPS benchmark: https://arxiv.org/abs/2105.09938
Note that the predicted date of "stronger" AGI has moved quite a lot since GPT-4 is revealed, from late 2030s to 2033 at this moment.
And so-called "senior engineer" salaries will now be brought down and deflated since they were inflated and unjustifiably high in the first place and are the main reason why these tech startups run themselves into the ground with little to no path to profitability.
I guarantee you that so far, the only winner in this is OpenAI. Not the 'senior engineers' building on top of someone else's AI API.
In fact, why hire 3 over-priced seniors when one junior with ChatGPT is significantly much cheaper? I quite find it funny that somehow, all hope is instantly lost because of a "AI" spitting out code will replace them. It just shows that the majority of these tech startups were just good at losing money and being solely dependent on VC cash.
If you are a CTO, this will output your productivity ten fold on the upcoming years. Now you don’t need to hire managers and can just build the product of your dreams with very limited capital.
If you are a VC, this will output your productivity ten fold on the upcoming years. Now you don’t need to hire anyone and can just build the product of your dreams with very limited capital.
Agree it'll definitely be amazing for creatives and solo founders, but how many ideas are really out there to be had compared to the reduction in workforce?
I don't know. But I don't see why you might not be able to ask GPT-6 or GPT-7 to enumerate (and patent and implement) all of them for you. Why do you think "founders" or "creatives" are special?
In the end, something like that is "amazing" only for the person who owns the most GPUs or manages to figure out the first effective meta-prompt.
> If you are a software engineer, this will output your productivity ten fold on the upcoming years. Now you don’t need to hire junior devs and can just build the product of your dreams with very limited capital.
And if you're a junior software engineer? Fuck you and be unemployed.*
Do you get it now?
* Until you can climb up the ladder where each rung is now 20 feet apart.
No white collar job will be valued the same since GPT will basically be doing most of the work and we will maybe review it and steer it. We will just keep feeding it and it will know everything at the cutting edge of all fields.
Is this really true? I may be missing something (I probably am), but I didn't find much use for AI tools in my itsec/programming work. It's a nice tool to have, but I don't write that much boilerplate. I've tried to use it as a better Google, but it kept replying with made up nonsense (things I have problem with are usually niche technologies OpenAI is not good at - I expect it will get better in the future). So I find it dubious it will "10x my productivity" in the "upcoming years". Decades, maybe.
But maybe the future really is now, and I'm just being an old-timer who can't adapt.
If you want to do anything new or - god forbid - know of a better way to do things than what 90% of the population is doing (htmx?). Good luck.
So the future is anyone with that model access (the 8k tokens could have 20kb of docs which is still useful) who wants to really try.
Also what happens to Europe? All these companies behind LLMs are from US, and Europe is nowhere to be found. This seems like it will dramatically accelerate the wealth different between the US and the EU.
Europe in itself is (together) the single biggest market/economy in the world by the way, and the US is actually falling behind into developing-country territory when you look at the population and their access to basic services. And just because right now it is convenient to rely on the US companies, and we're deep allies btw, doesn't mean europeans couldn't spin up the same tech if really needed.
Yet here we are.
It’s a human-political issue, it is not a technology issue.
What’s the difference now?
OK, put your money where your mouth is and send me 10% of your pay check.
The fact is that anyone who understands even at a basic level what the computer is actually doing and isn’t afraid to look at it at a low level can’t be replaced by an AI trained on stack overflow.
It may be that I will spend more of my time on code review of LLM generated code, or make my money in the new kinds of legacy code created by copy pasting ChatGPT snippets together instead of SEO optimized stack overflow scrapes.
For me the outcome is the same. The skills I need to be more effective than the machine are the exact same as they were decade, century or even millennium ago. I still don’t see these LLMs do any synthesis of knowledge, and they don’t seem to have a grasp of logic or grammar at the level I expect a bright middle school student to have.
Lol, I was thinking about this the other day. Eventually most devs will essentially just be praying to the Machine spirit to make the computer do what they want. A small few high clerics will bother to learn how computers actually work. The rest will simply be cargo culting to the maximum extent possible.
If the technology pans out the way the techno-enthusiasts hope it will, upward social mobility will be nearly eliminated... unless there's some kind of successful Luddite revolution against the technology and the people that own it. But that's not going to happen: there are all kinds of social pressure against revolution, as well as strict gun control in most places. Anyone who tries to resist their obsolescence will soon find themselves either ridiculed and condemned or in jail.
Of course, downward social mobility will accelerate, and be celebrated by idiot technologists who just want to build tech, and don't really care to think about the consequences of the technologies they build on real people.
Ideally, defund the police and so on, so that every state worker also is keen on getting that wealth redistribution done.
Those businesses would not be around for very long, so who cares?
As you might have noticed, the AI boom will decimate the code writing jobs as well, something that the EU is behind on. Europe missed the "tech" age, but notice how the EU is not any poorer than the USA. Sure, some countries are poorer than others, but not everywhere in the US is Silicon Valley. Why? Because despite the EU missing out on "tech", actually the EU is very technologically advanced. Tech doesn't mean only low-touch high-scale computer-based businesses. There are chemists, biologists, anthropologists out there who don't know how to write a single line of JS and are paid like 1/5th of a junior JS developer, but the work they do is very valuable to society. Guess they don't need to learn JS anymore.
Also, notice how despite the thousands of layoffs, the US job data keeps coming out very positive - there's no unemployment problem. This is because of the markets, but AI will have similar effects. The world no longer needs that many CSS experts and React gurus who pull in $200K; the world apparently needs more hard-tech engineers and retail workers.
The AI thingy is devastating just for a subset of the "tech" workers and creative industries. It will enable other types of people and industries.
Startups who are trying to solve food production issues, for example, might finally outshine the next grocery delivery startup.
EU is significantly poorer than the US. Lots of different ways to measure it, but it’s a factor of roughly 1.5-2x in purchasing power parity.
And so on. Maybe the answer is in fact "yes", or even "yes, and it would have done these things even better than humans did". But so far it seems to be amazingly good at doing things that we showed it how to do.
If we stop creating actually new things, will it do that for us also?
Why would it care to do so? What interest does it have in creating new things on its own?
If the amount of people that will have social mobility opportunities will be equivalent to the amount of people who could have invented the MPG format or something comparable, then my point is made.
Would you?
Could you?
This is exactly the opposite of what you would expect given the increased efficiencies that come from adopting computer systems and automation.
I see AI as a continuation of this trend and I don’t expect it to put people out of work, bureaucracy will always find new ways to justify itself.
Like, I could technically have a newbie running commands on a production router for a script that I wrote out...but even if I let them do that there's no way I wouldn't at least supervise. I don't think most companies are even remotely comfortable with the idea of having an AI system running code on their systems no matter how smart it is.
And at the very end, it will reduce to capital only, with no need for labor at all. Most people will be unemployed, and whatever capital they've amassed is unlikely to be enough to sustain themselves and their families for the long term. They (you) will end up as little more as impotent ants to AI-fueled Elon Musks, neglected until the infestation needs to be cleared to make way for some project.
People who are in tech just to "climb social ladder" i.e. only for the pay check are going to be pushed out by LLMs and people who are actually passionate about tech will remain. This will cause less and less shitty code to be written (of course for next few years even more bloated shit code will be written with ChatGPT and Copilot by noobs who have no idea what they are doing)
So far technology has enabled use to increase economic output which means rising standards of living. Even if 99% of people subsist from selling their labor, the tools they use are a force multiplier that (in theory) drives wages up.
When you can spin up a bunch of Von Neumann level intelligence LLM-powered agents and have them run your company for you, there is no more labor to sell. You can either pay the former laborers to exist, or just let them starve.
So our two options are social ownership of all AI capital, or letting everyone without AI capital die, and let a handful of people live in the resulting AI-powered society.
* Although ChatGPT is pretty good at generating code, it kept making simple mistakes such as calling non-existing APIs or introducing bugs. Some of them it could fix itself, some I had to fix.
* The code provided worked well for the "happy path" but failed miserably for some corner cases. I had to fix that manually.
* The code was working, but I wouldn't consider it production ready. It required some cleanup, unit tests, etc. Again, some of this with ChatGPT, some without.
* Not to mention that I was the one with the knowledge about the domain, what problem to solve, a vague idea of how...
Not to pick on OP but extracting a few seconds of video from a file is a pretty straightforward task, you can essentially do it with a bash one liner [1]. My biggest question is how ChatGPT performs with a large codebase, contributed over time by different authors, with complex domain logic and layers of abstraction.
I also had a brief existential crisis, but I just shrugged it off and got back to work.
[1] https://askubuntu.com/questions/59383/extract-part-of-a-vide...
I saw a bunch of people talking about how GPT helped them code stuff on Twitter, so I thought I'd give it a try. Right now I'm building a sort of simple, mock version of the type of software that integrates with my company's APIs. I've successfully managed to create a simple web application that creates a new object, hits my company's API endpoint to create a corresponding object on our software, allows me to upload a document locally and then allows me to upload that document to our software via API as well. It's all a little messy and clearly not production-ready, but it works. It would've taken me probably a few months on nights and weekends do this (mostly refreshing myself on JS and Python). Instead I've done it in <24h (would've been shorter except for GPT-4's message limit).
I'm sometimes able to spot and fix GPT's bugs, but even when I'm not, it walks me through adding more logging and successfully debugs issues. Sometimes it takes a few tries and a little direction as to what I suspect the issue is, but so far it's fixed everything that's come up. I don't think this would be doable for a totally non-technical person, but I do think it'll get there pretty soon.
I'm just absolutely blown away.
I think we're going to see a lot of programmers who are going to trust GPT a little too much, and I think that's sort of scary. For the most part that is going to work out just fine. Often the quality of your programming isn't actually going to matter that much, because as long as it solves the business needs okish, then it's frankly great. That's not always the case, however, imagine someone using GPT to get your healthcare software wrong.
I'm still impressed with it in other areas. I think it'll do wonders in the world of office automation because it seems to have the ability to succeed at this much better than any previous "no-code" attempt where the logic would almost always end up requiring people who are basically programmers for it to work. I think GPT will help here, requiring less "superusers" for a department to move their data flows into automation. Especially in areas, where efficiency and stability aren't necessarily that important if the automation-tools mean you don't need three full time employees moving data from one system to another.
Modern languages (and tools like autocomplete) have already helped that a lot compared to assembly code or binary, this looks like the biggest jump in a long time. The path of programming so far has been moving from "describe how to do something" to "describe what to do" which this is certainly in line with.
It’s easy to get amazed by something that can halfway do something you can’t do at all automatically. But as others have pointed out, it’s not that great at it and not knowing enough to do it yourself means you don’t know enough to catch and fix bugs.
So this move to using chatgpt and similar in production by people who otherwise wouldn’t be able to do things in production is worrisome, imo.
The availability of a button inside an IDE doesn't make this a fundamental change in how we work
I don't feel that my job is at risk of disappearing. Instead I think we'll be using LLMs as tools to do our job better.
Therefore the inverse can be safely inferred by nondisclosure.
Yesterday I got a complex data structure out of it in 1h that we'd been talking about but not implementing because it would have taken a couple of days to get right.
In all cases it made mistakes and I had to rely on my experience as an engineer to ask the right questions and fix things. But god damn it made me insanely more productive.
Don't shrug this off and go back to work. You'll get left behind, and may not have a work to go back to.
This is supposed to take programming jobs?
HN is incredible.
Not being able to write code without it might be bad but it's a valuable resource and you should use it when it's available to you (for both)
I have no doubt that ChatGPT will become even better than StackOverflow at answering questions. Is this really going to make us better programmers?
Consider this classic: https://stackoverflow.com/questions/12122159/how-to-do-a-htt...
I was discussing a bug with a colleague, so for curiosity's sake I decided to plug a similar question into ChatGPT. I was quite impressed with the solution it gave, and interestingly, it had the same subtle bug that our code had. What blew me away is that when I pointed out the bug, ChatGPT fixed the code by itself. On one hand I felt "phew, at least it needed me to point out the bug", but then I thought "I just (perhaps stupidly) provided training data so that down the road ChatGPT would get it right the first time."
The other day I was working with the Cisco Meraki API...I knew exactly what the script needed to do, but the calls were tedious and I didn't feel like learning the names of all the JSON columns, so I just had ChatGPT do it. I had to fix a couple mistakes, but the 20 minutes it took was better than having to read all the documentation.
This quote highlights the challenges of accepting new information or ideas when they might jeopardize one's livelihood or status quo.
Let's wait for "And now... long term memory is all you need" paper.
And afterwards it cleans the cup and puts it back in the cabinet.
If you want it to solve arbitrarily complex problems, you need to set up some sort of loop. People are already feeding the outputs back in as input in various primitive ways, but I suspect the real breakthrough will come when someone trains some sort of recursive transformer from scratch. (Assuming the current networks waste neurons in unrolling loops, we might possibly even see smaller models).
[0] Try the following family of prompts: "_ is an example of _, which is an example of _, which is an example of _...." etc to a depth of your choosing. At some point it bottoms out and you can't get any more levels out of it.
Basically we need to equate "safety" in LLMs to mean "being open-source".
OpenAI keeps talking about "safety" as the most important goal. If we define it to mean "open-source" then they will be pushed into a corner.
Another way to put it is to make it more accessible to everyone, right?
The opposite of that is happening to nuclear power. They're actually trying to stop any more countries to have the technology at their disposal. So no, make it "open source" doesn't make it safe by any stretch of imagination.
reactor blueprints have been accessible to IAEA members for something like 50 years
I think open source is a reasonable component to safety, but I wouldn't want to make them equal. Open source may be necessary for safety, but I wouldn't call it sufficient.
For example, assume the source code, the model, the training data, and all the model weights are open source. How do you know that the model was actually trained using that training data? Very few organizations have the capacity to train models at this scale themselves.
Whenever I see this I simply think "monopoly". It smells of anti-competitiveness and is a kind of open forum lobbying to restrict who gets to lead the AI wave (and make a shit tonne of money in the process).
Spot on. It's a good time for existential reflection: Who would you have been hundreds or thousands of years ago? Who will you be now that technology is radically changing again?
There will always be interesting, creative challenges like programming, whatever form they take.
Given how I grew up ingesting science and science fiction alike, literally attributing half my personality to Star Trek: The Next Generation being on TV during my formative years? It's really hard to tell. I have very little connection to things which were possible before late 19th / early 20th century.
In my mind, being thrown back centuries in time, I'd spend my life trying to use everything I remember from present day to give everyone a head start on science and technology. Being thrown back centuries in time, but without the memory of specific things I've learned in present day? That sounds like a particularly sadistic death sentence.
Obviously you won’t be able to tell for sure, but I’d guess that 1000 years ago I’d probably be a serf, and 100 years ago, likely would have fought in a large war and likely doing some form of physical labor or subsistence farming afterwards, based on what my family was doing then.
I'll just use this opportunity to recommend the video game "Ancestors: The Humankind Odyssey". It's a game where you start as an early hominid and have to gradually discover how to make and use rudimentary tools in order to take control of your environment, literally evolving in the process. It's weird and unforgiving, and it made me really think.
Do you invest in a college education is that field is obliterated by the time you get out.
What about your debts if you lose your job and companies aren't hiring because they can just use AI for a 10th the cost in 6 months.
the instructions for configuring google auth were off. I tried a number of different ways to get gpt to give me the right instructions, but to no avail.
so it was back to the old way, of spending a few hours reading google's documentation (which I'm doing today) to figure it out.
once I'm there, I feel confident I could better coach chatgpt to instruct me. though I wouldn't necessarily need the help at that point.
on the code side, staring at the google auth api code it had generated, I was faced with a hard truth. I didn't understand this code. to iterate with it, essentially to develop it, I would continue to be dependent on GPT. Even if there was a one liner needed, I wouldn't be able to come up with it on my own. I'd always have to rely on this outside "brain". How can that be more efficient than a tight REPL loop conducted by me, an evolving master of this API?
And how will we humans even maintain knowledge of these API surfaces if we are not putting in our hours and hours of repetitive usage of them? We become ignorant of the evolving capabilities of the computing platform. And chatgpt becomes useless without humans who understand what's out there, what's needed.
Wasn't this the final objective of the programming languages abstraction evolution? From Binary/Assembly to Natural Language Programming? I think it is awesome that more people will be able to create software/products as this accelerates innovation cycles a lot.
And, for now, I believe devs that don't rely solely on copy/paste coding from stack exchange don't need to worry about their job stability no?
I envy the people who are bottlenecked on their typing speed and benefit 10 times more from the chat bot than I do.
When I need to ask for boilerplate code for fetching a web resource or using a well-defined API, ChatGPT is great.
ChatGPT has made the mundane plumbing a lot easier. It is a threat to plumbers at this point. Many of those plumbers are now freed up to do more valuable work. I am happy to have it, so I can focus on higher value work.
If your only skill is at this kind of low level plumbing, you are in danger. But I doubt this is the case for most.
Today.
What happens when it understands computational geometry and can calculate an optimal strategy to apply it to a dataset and end goal you provide?
Until someone starts testing this and finds a bug. And then AI will say, hey, there is no bug, I don't make mistakes. So you need a human to look on the code, a huge pile of spaghetti code with cryptic names and conventions, code patterns that fell out of fashion years ago but, since there is a lot of code that uses them, AI thinks they are ok.
How long it will take to fix anything, how long it will take to extend the code?
The code it generates is by no measure "a huge pile of spaghetti code with cryptic names and conventions".
I was sceptical myself before trying GPT4. I asked it to change the Python C internals for a new feature, and googled to ensure the description doesn't exist anywhere. It came up with very good changes and explanations.
And this is all not even mentioning the pace of improvements. It didn't take too long to go from GPT3 to GPT4. Even if the pace slows down, it is still huge.
> it does not need good variable names or functions/methods/class names,
It's the exact opposite. It's too good at naming things. It insists to use variable/function names that make sense in plain english, and often make mistakes when the API has inconsistent naming, or consistent but unusual naming.
For example, it makes mistakes when writing code that use "Loop" in Blender API. And the reason is quite obvious to me: because Blender's "Loop" is not what loop means in plain english.
Did OpenAI just commit a trillion dollar mistake?
I don’t think convert it in and out of proprietary standard is that difficult?
There is little to no vendor lock-in effect
I don't see this
What insight!
My first reaction was to be afraid for my money-making skills. My second reaction was fear about us ourselves making ourselves irrelevant--that fear still lingers.
My third wave of fright, cemented by days burning my eyes looking at a screen parsing logs and trying to figure out bugs for my corporate master, was, "when did my imagination go for a vacation? Old boy, don't tell me now that you have run out of ideas of things to make, of things to have an AI army to help you build." And now I dread that all of this AI is just hype, that it will never be good enough to come for our jobs without also coming for our jugulars, or that we will make it too damn expensive to matter[^1].
-------
[^1]: Capitalism has a way of leveraging economies of scale to make certain goods cheaper. But there are physical limits--what if Moore's law with regard to power consumption is really dead, and we as a collective really decide to spare power?
It's either my imagination that has gone for a vacation, or yours is running wild, but that is the one thing I really can't see at all. Reducing power consumption? I don't think that's happening any time soon, or ever really.
Some day it will be. Not those ones, those ones are only hype. Also whether or not they'll come for our jugulars depends on what they are commanded to do. But we will get them eventually, and they will be as good as articles like this pretend the hyped ones are.
The funny thing is that nobody will use the current panic to prepare. And everybody will use the current panic as an excuse to avoid preparing once the real AIs come. So they'll get us completely unprepared.
Interesting! Somehow I missed this. https://spec.openapis.org/oas/latest.html
90% of programming is communicating with other people - chatgpt can't talk to people.
It also can connect to your Notion, Slack or whatever
Have we already solved AI safety problems? It seems like LLMs can now execute shell commands on our computers.
[0]: https://openai.com/blog/chatgpt-plugins#code-interpreter
There will be no Post-GPT computing world, just the Turing police and console cowgirls.
There's so much of talk about what these models can generate, which is cool in relation to plugins, but there's still a lot of interesting code to write, companies to build, and ideas to formulate, that an LLM cannot do on its own. If you're terrified of your software engineering job becoming at risk, I urge you to just take a beat.
There was a paper by someone @ Microsoft who tried to train a boardgame playing AI like this. The "best" models started losing to beginner level players from some point onwards.
I'm processing this news in realtime like many of you and forming a plan:
1. Understand how LLMs work. I've heard the Wolfram paper is good; open to more suggestions here.
2. Continue to practice using real implementations of LLMs including ChatGPT and co-pilot.
3. Finding painpoints within our company that AI can make more efficient and implementing solutions.
If anyone feels the same way and wants to form a working group with me, give me a shout. Email is in my bio.
For the understanding part, Andrej Karpathy has a YouTube playlist that explains neural networks. I made a start on it today and found it quite accessible.
https://www.youtube.com/watch?v=VMj-3S1tku0&list=PLAqhIrjkxb...
The part I'm finding is kind of a shock to me is the impact of the centralization on what you can even think about doing. If your application falls under their random definition of "unsafe", then you can't do it. Not even manually, probably, because the infrastructure for that will go away. If your one off question or task doesn't meet their approval, it doesn't happen.
Basically not only do the owners of these things become the only really important people in the economy, but they also get a new kind of direct control over people's lives.
Because yeah it works fine for basic programming things but I believe you need to know wtf you’re doing when it comes to anything more complex, even something basic like some of our single endpoint services.
I suspect many large IT organizations are like this.
Writing is definitely on the wall for outsourcing and MVP-style work. GPT can create a landing page and a backend/frontend for a business _literally today_. You just have to ship it, but it won't be long until that isn't needed.
There will still be a lot of value in understanding how systems work and interact with each other, at least until ML is able to build and maintain entire systems.
Until that happens, there will still be a lot of value in being able to dive into codebases and refactor/optimize as needed, at least in the medium-term.
Once platform engineering is mostly automated and running AI-generated binaries is de-risked, then code quality doesn't really matter. Hell, _code_ won't even matter at that point.
To me, this sounds a lot like "at least until ML is able to reach level 5 self driving". We don't even know if this is possible yet without AGI (which we also don't know is possible). We can get close, but... that last 1% is a bitch, and it makes all the difference.
I missed this. Can someone show me what he is talking about?
I know it used a huge amount of energy / GPU cycles / time to train, but now that the weights are computed, what's involved in running it? I know the model is huge and can't be run on an ordinary developer's machine, but I believe requests to it can be batched, and so I don't really know what the amortized cost is. Right now, this is all hidden behind OpenAI and its credits; is it running at a loss right now? How sustainable is using GPT-4 and beyond, as a day-to-day part of professional life?
If the LLM has seen lots of instances of usage of an API, it can write code to target the API. It can generalize to some degree, but things go off track the further your requirements are away from the training data.
If your code is a lot of duct tape between well-documented, or at least well-named, APIs, that code can be automated. Which is great. That kind of code was always boring to write.
I'm less convinced that LLMs will be great at inventing new abstractions to map to a problem domain, and wiring up these new abstractions in a large codebase.
They'll need augmentation, fine-tuning, guidance, and it's not clear how well it'll all fit together, and where the limitations of the tech will show up as capability cliffs.
It's also a good time to really take our heads out of the sand and re-evaluate how we expect people to learn civil engineering if their only teacher is a minecraft world. You might get some people that are perfect in minecraft. The rest will be hopelessly stunted. Pretty soon it'll pivot to materials engineering to figure out how exactly a minecraft block adheres to a surface because we lost the original irl way to build a bridge.
Why is this extraodinary? What would be the advantage of going through all the effort of defining a new format just to create busywork for people trying to integrate with you?
It's not like there would be anything stopping Bard/Alpaca/etc. from reading the same format as OpenAI.
About twenty years ago, I had a professor explain to the class that Rational Rose would be replacing us all....yet here we still are.
In that endgame, anyone who can speak can command AI can do whatever they want it to do. Any kid with a louder mouth can outwish the wisest man on earth.
That means shortsighted impulsive criminals can use it to learn how to steal. Shady politicians can use it to astroturf entire campaigns. Everyone knows the tropes but it bears repeating as we all march dumbly towards what's coming.
It is far easier to destroy than it is to create. And humanity aside from China has not demonstrated any sort of sensible strategy to temper the tendency of destruction to outpace harmonious creation when it comes to AI. The more I see AI emerge and see people use it for exactly what people fear it shouldn't be used for, the more I feel China's centralized adoption of it, though maybe not "feel good", might be the DNA that survives in the natural selection of societies.
I know of one person who pays for GPT, and I'm guessing they use it to astroturf demand for their own business's products, since that's what they were doing by hand when they were younger.
OpenAI's plugins are equally temporary. Right now they will be generating actions through APIs, but GPT4 is probably already capable of performing the same actions on your browser. All it needs is a "control my browser" plugin that allows it to make that reservation on expedia, without expedia having any control in it. It will inevitably eat the world again
What chat app? Is this gpt-4? I haven't seen anything executing the code that is generated. So is the above quote a hypothetical or what?
As of right now, even if ChatGPT were to generate 99% accurate responses, it's quite a chore to communicate with it in full sentences. I don't want to have to explain my business in full painstaking detail and then upload tax documents to a system that can then output an answer in book form back to me.
Then I predict we'll get more business analysts than programmers, since managements will still need people to translate their needs to AI.
Why would analysts be harder to replace than devs?
The question is - how will competition influence the job market? if everyone has AI, everyone has the same powers. So how do you differentiate yourself? You put more humans in the loop, like "human plugins". You need humans to extract the most from AI.
We still need actual experts to vet the code LLMs produce and to choose the optimal solutions. This is what senior devs have done so far with junior and mid level devs always. There are people who can write code, but someone needs to review and approve what they have done.
Obviously LLMs will also eat into that space, but before we come up with AGI LLMs alone won't be able to completely replace humans in software.
Key thing for adoption is to make models smaller and more context specific (to make them smaller), we've seen how LLaMA was downsized to run on commodity PCs, we've seen how Stable Diffusion can run on mobile phones. Even when we have to use larger models remotely, cost and ownership matters.
> Accepting the term "intelligence amplification" does not imply any attempt to increase native human intelligence. The term "intelligence amplification" seems applicable to our goal of augmenting the human intellect in that the entity to be produced will exhibit more of what can be called intelligence than an unaided human could; we will have amplified the intelligence of the human by organizing his intellectual capabilities into higher levels of synergistic structuring.
Now that the computers can talk and think and program themselves, and we can expect them to become exponentially better at it (to some limit, presumed greater-than-human), there is approximately only one problem left: how to select from the options the machines can generate for us.
It's still an open-ended challenge, it's just a new and different challenge from the ones faced by all previous generations. And again, just to repeat for emphasis: this is the only intellectual challenge left. All others are subsumed by it (because the machines can (soon) think better than we can.)
> OpenAI made the extraordinary and IMO under-discussed decision to use an open API specification format, where every API provider hosts a text file on their website saying how to use their API. This means even this plugin ecosystem isn’t a walled garden that only the first mover controls. I don’t fully understand why they went this way, but I’m grateful they did.
So accelerant, definitely. Beyond that, I'm on the sceptical side but accept there's quite a chance that's the wrong way to bet.
It's a good point and some have already got this to work:
https://twitter.com/vaibhavk97/status/1639281937545150465
Given that there's no technical obstacles to drop-in compatibility here, I wonder if we'll soon start seeing exclusivity requirements and such.
It's one thing to ask GPT to write a high level script to trim 5s off a video using ffmpeg. It's another thing to ask GPT to make ffmpeg, or even to make a specific modification to ffmpeg.
It's hard to say how good GPT will be at real-world programming since we currently can't try it out. Maybe it can scale to the task, or maybe it can't, but i wouldn't say that programming is "finished".
Doesn't this show that we can now use this technology to generate and execute code for modest problems that have already been solved, while we can spend more time on even more complex problems?
Once we can send LLMs to meetings with each other, we can move down to 15 hours of purely joyful work :-D
That the draft happened to work on the video clip is more luck than something you want to bet your engineering life on.
You still need to go through an verify every character this statistical package spits out - it is not magic - it is just a probabilistic machine.
Or the end of the beginning (of software development)...
What do I need them for if I can get equivalent code written for me on-demand?
