OpenAI Hires an Army of Contractors. Will They Make Coding Obsolete? (semafor.com) 110
Last week Microsoft announced 10,000 layoffs — and a multibillion-dollar investment in OpenAI, the company that created ChatGPT.
But OpenAI also released a tool called Codex in August of 2021 "designed to translate natural language into code," reports Semafor. And now OpenAI "has ramped up its hiring around the world, bringing on roughly 1,000 remote contractors over the past six months in regions like Latin America and Eastern Europe, according to people familiar with the matter."
The article points out that roughly 40% of those contractors "are computer programmers who are creating data for OpenAI's models to learn software engineering tasks." "A well-established company, which is determined to provide world-class AI technology to make the world a better and more efficient place, is looking for a Python Developer," reads one OpenAI job listing in Spanish, which was posted by an outsourcing agency....
OpenAI appears to be building a dataset that includes not just lines of code, but also the human explanations behind them written in natural language. A software developer in South America who completed a five-hour unpaid coding test for OpenAI told Semafor he was asked to tackle a series of two-part assignments. First, he was given a coding problem and asked to explain in written English how he would approach it. Then, the developer was asked to provide a solution. If he found a bug, OpenAI told him to detail what the problem was and how it should be corrected, instead of simply fixing it.
"They most likely want to feed this model with a very specific kind of training data, where the human provides a step-by-step layout of their thought-process," said the developer, who asked to remain anonymous to avoid jeopardizing future work opportunities.
But OpenAI also released a tool called Codex in August of 2021 "designed to translate natural language into code," reports Semafor. And now OpenAI "has ramped up its hiring around the world, bringing on roughly 1,000 remote contractors over the past six months in regions like Latin America and Eastern Europe, according to people familiar with the matter."
The article points out that roughly 40% of those contractors "are computer programmers who are creating data for OpenAI's models to learn software engineering tasks." "A well-established company, which is determined to provide world-class AI technology to make the world a better and more efficient place, is looking for a Python Developer," reads one OpenAI job listing in Spanish, which was posted by an outsourcing agency....
OpenAI appears to be building a dataset that includes not just lines of code, but also the human explanations behind them written in natural language. A software developer in South America who completed a five-hour unpaid coding test for OpenAI told Semafor he was asked to tackle a series of two-part assignments. First, he was given a coding problem and asked to explain in written English how he would approach it. Then, the developer was asked to provide a solution. If he found a bug, OpenAI told him to detail what the problem was and how it should be corrected, instead of simply fixing it.
"They most likely want to feed this model with a very specific kind of training data, where the human provides a step-by-step layout of their thought-process," said the developer, who asked to remain anonymous to avoid jeopardizing future work opportunities.
This is what Betteridge's Army says to that (Score:1)
First, imagine ten pages of waffle for the proper clickbait experience.
And it all boils down to this bit of useful content: "No."
Re: This is what Betteridge's Army says to that (Score:1)
No, but C will make even more sense. All that "impractical" memory management will be mass produced now.....
#LearnToCode (Score:1)
Oh wait... Well, I guess you can always get a job making solar panels or batteries.
No (Score:5, Insightful)
75% of my job as a programmer is to take requirements from customers and managers and turn them into something that actually makes sense.
Every time we have a big breakthrough in technology and someone tries to come out with a product that allows you to automate tasks without code, it makes the programmer's life easier, but it still doesn't get rid of the need for programmers. Look at the average spreadsheet made by someone in management. Full of errors and inefficiencies. They can't even communicate with other humans what they want. good luck getting an AI to understand them.
Re: (Score:1)
"I like the way you whistle past the graveyard. Can you teach me how to do that?" said the AI.
Re:No (Score:4, Interesting)
Every time we have a big breakthrough in technology and someone tries to come out with a product that allows you to automate tasks without code, it makes the programmer's life easier, but it still doesn't get rid of the need for programmers.
Not only that, but these technologies tend to increase the demand of programmers. If programmers were still writing code in punch cards today, the entire field would be 10% or less of what it is today. We are nowhere near even understanding how much demand there is for software in the world, so every time the cost to develop a unit of software is reduced, software needs which used to be too expensive then become in reach for businesses. This results in more developers and often higher salaries for those developers.
Until we get to the point where all the software needed to be written has been written (and cannot be improved upon), we will still need more and more developers. The industry could probably absorb 50% more developers today if there were enough people capable of doing it (there isn't).
Re:No (Score:4, Insightful)
AI doesn't take days or weeks to write the code. It takes about a second.
And what stops it from asking such questions with vastly more patience than you?
Re: (Score:2)
> And what stops it from asking such questions with vastly more patience than you?
I don't think the patience of the programmer is at question.
I think Star Trek got his one right: https://www.youtube.com/watch?... [youtube.com]
Commander Data makes Sheldon Cooper seem tolerable.
Re: (Score:2)
Re: (Score:2)
AI doesn't take days or weeks to write the code. It takes about a second.
Yes, and writing code in Python is orders of magnitude faster than writing it with punch cards. AI will most likely allow those who can understand how to translate business requirements into code to write code faster (with AI), not get replaced by that AI.
Re: (Score:2)
And what stops it from asking such questions with vastly more patience than you?
This is not strong AI. It can ask good questions, but there is no guarantee those questions will be relevant as the AI has no actual understanding of what it is doing and why it is doing it. Maybe next year, but not now.
Jobs Evolve (Score:4, Interesting)
75% of my job as a programmer is to take requirements from customers and managers and turn them into something that actually makes sense.
Yes, but isn't that EXACTLY what ChatGPT does? It takes instruction from a user in natural language and converts that into a database search, calculation or other code.
I'm not saying that ChatGPT is any where close to good enough yet to replace a competent programmer but were it to be improved it certainly could get to a point where it would.
If you doubt this then think about the old human job of "computer". Before modern computers were built teams of humans were employed to perform complex calculations. These jobs changed into programming jobs with the advent of modern computers. Is it so hard to believe that programming jobs will eventually similarly be replaced (perhaps by "AI trainer" jobs?) if machines advance to the point where they can directly interact with humans without the need for a human trained specially to talk their language?
Re: (Score:2)
Yes, but isn't that EXACTLY what ChatGPT does? It takes instruction from a user in natural language and converts that into a database search, calculation or other code.
Yes, but can it do it with a snarky, disinterested attitude?
Re: (Score:2)
Yes, but isn't that EXACTLY what ChatGPT does?
No.
Re: (Score:2)
It currently turns a clearly and carefully stated requirement into marketing speak that sounds good ... ...it will turn badly phrased, and poorly communicated requirements into marketing speak that does not match what you wanted ...
The job of programmers is secure - it's "coders" that are no longer needed - but this is what most tools try and make easier and quicker already
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
The problem is that today's languages are conceivably high enough that this could already happen.
What chat GPT doesn't do is respond to ambiguity with a request for more information. That is what good programmers spend a lot of time doing. Gathering and figuring out the actual requirements and how they differ from the spec that they were handed.
If GPT learns to know when to ask questions to get more information on what the problem is . Then I think we have an issue with
Re: (Score:2)
What chat GPT doesn't do is respond to ambiguity with a request for more information.
Yes, but in some sense it doesn't need to. A human programmer will do that because they do not want to waste a lot of time coding something useless but ChatGPT turns out code almost instantly so instead of asking it can write the code and if it does not do what you wanted you can just add the extra specification and get it to regenerate the code.
In fact, this could be actually more efficient that a human programmer. Instead of having to specify everything in excruciating detail you will only need to spe
When computers program themsleves (Score:2)
Many decades from now, because AI programming is one of the most difficult things people do. But we will get there eventually.
Then they will program themselves to be more intelligent, recursively, producing exponential growth. Very slowly initially, but faster and faster.
And what will the machines do with all this intelligence? The same that we do with ours.
http://www.computersthink.com/ [computersthink.com]
BTW. I asked ChatGPT what resister I needed to reduce 12 volts to 10 volts under a 1 amp load. It produced a beautifu
Re: When computers program themsleves (Score:2)
Yep, Train AI with bad data, and it's as likely to learn how to most-effectively hide its mistakes from humans proofreading the code it writes as it is to learn how to avoid actual bugs. Unless the training corpus distinguishes between the two, as far as AI is concerned, they're one and the same.
Re: (Score:2)
FileMaker with OpenAI would be amazing.
Re: (Score:2)
I'm a programmer too, but I'm perfectly aware that AI systems can also get requirements from customers and managers and create the software they need. AI will be much better at understanding what a manager want, but also it can even readjust the outcome much faster as we as a programmer can. We normally need a few days to actually turn what the customer explained into a prototype and show it to them, AI can just create the prototyp on the fly while the customer is providing it's comments, so the AI can imme
Re:No TI CoolGen (Score:2)
Yeah (Score:2)
Remember they don't have to make all of programming obsolete. If 20% of IT jobs go away in the next 5 years that's going to slash your wages.
People tend to think in terms of black and white and All or nothing. People have a really hard time thinking in terms of gradients. But the Business ghouls celebrating at the prospect of slashing headcount think very well in that direction. And it'
Re:Yeah (Score:4, Interesting)
So that's not true either. If people become productive enough with the new tools that 80% of the workers can do 100% of the workload, it doesn't actually mean a 20% job loss.
https://en.wikipedia.org/wiki/... [wikipedia.org] applies.
For some historical evidence : consider how python or javascript based HMI frameworks make older website designs very easy to build and maintain. Easily 5* productivity. Are there 5 times fewer front end or full stack devs? No, there seem to be more than ever.
It's because the increase in productivity per labor hour makes more complex sites and applications possible, which is what will happen here.
Re:Yeah (Score:5, Interesting)
As soon as professional sports start shutting down and we start seeing a sustained year over year decrease in the number of movies and TV shows produced and car dealers start reporting that their lots are full of two and three year old unsold luxury cars and SUVs, then I'll start worrying.
Right now there are too many people with too much disposable income for you to be posting the same doom and gloom over and over again on every Slashdot story.
The human species can survive on rice and beans and a few other basic requirements, but we're many orders of magnitude away from bare subsistence. The US is awash in luxury goods, far more than the billionaires can buy, and they're getting bought up by ordinary people. Some people have bad luck and some people make poor choices, but on the whole the US population (and many other parts of the world) are so far above survival and into entertainment and recreation that they can't even imagine what it was like before the progress that made those things possible.
A few people may lose jobs, and a few of those may have trouble finding new jobs, but it's a drop in the bucket compared to the total number of people who have so much money that they're spending it on completely unnecessary industries that exist solely because of the sheer number of people who can spend frivolously (even if some of them complain about being broke while spending frivolously)
Think about Apple computer (Score:1)
And I don't anticipate most of the people reading this to be part of that subset. The problem isn't whether we can maintain the quality of life the majority of Americans enjoy the problem is will we? The way we're doing it right now is that there's a lot of work tha
Re: (Score:3)
Every single time in human history that we have had some revolutionary technology advancement the end result has been more jobs and better standard of living.
Enough with the class warfare gloom n doom. There is no conspiracy afoot by the billionaire class to starve you to death and make you eat crickets and drink your own urine to survive.
Re: (Score:3)
The human species can survive on rice and beans and a few other basic requirements, but we're many orders of magnitude away from bare subsistence.
Who is this "we" you are referencing? I strongly suspect that you are completely unaware of a vast and growing underclass of Americans that can not afford to have a roof over their head, electricity and water, AND food.
The US is awash in luxury goods, far more than the billionaires can buy, and they're getting bought up by ordinary people.
Sure, there are hundreds of thousands, possibly even millions of people who can afford some luxuries. There are tens of millions who can not and hundreds of millions in between unable to afford a pot to piss in and able to afford basic necessities.
Your economic outlook is from a privileged po
Re: (Score:2)
The fact that you have a computing device, internet access, and enough idle time to be shitposting on Slashdot makes you yourself a privileged person. Check yourself.
Re: (Score:2)
How does your response address any of what I said? Yes, I am in a privileged position right now myself. That does not mean that I am unaware of what is going on around me. I grew up very poor. I am no longer poverty stricken. I still have friends who actually are poor.
Would you mind addressing what I said instead of acting like an ass?
Re:Yeah (Score:5, Interesting)
I actually kind-of doubt this, because it ignores a ton of complexity that exists in business.
First you have to realize that the C* folks in a lot of places are highly resistant to institutional change. They got where they are now often not taking huge risks, but instead just iterating on old, outdated, inefficient things. "I'm not going to risk our Q3 and Q4 goals just so you can run your stupid little AI experiment!"
So it will take time for this to break into established companies. Sure, some of the major tech companies will blow a couple hundred million on a new wing of AI stuff, but if the ROI isn't there, and it won't be right away, that's going to give a lot of other companies pause.
Where this will happen is in startups. They're absolutely going to get VC money to play this game. But again, there will be more successes than failures, because that's what startups are. That's going to look less than appealing to a lot of middle sized companies.
Where AI shines first is going to be automating trivial shit. It already can do this a bit. Who does trivial shit? Junior developers.
So I see the IT jobs going away first as the junior dev ones, and that's going to suck for this generation of college kids. But that also reduces the pipeline of folks in IT, and as the senior devs retire, it may will INCREASE wages, if there just aren't enough bodies with real world experience to replace them.
The question in my mind is if AI can develop programming chops at a rate where it can do the same or better work for cheaper than a junior dev. If AI can, we're going to cut off our nose to spite our face with it. If AI can't, then it's going to find niches where it works great, but lots of areas where humans at least need to hold its hand, if not do the work themselves. At least for a long damn time.
C-suites love change (Score:2)
Re: (Score:2)
Yup it is impossible now to get an IT job in the US. All jobs were sent to India. Right. /sarc
Enough with your racism. It's not the first time you've gone there before. I memorialized your last big foray into anti-Indian racism in my sig. That's how disgusting your racist rant was.
Just stop it. It's really gross and offensive.
Re: (Score:2)
Remember they don't have to make all of programming obsolete. If 20% of IT jobs go away in the next 5 years that's going to slash your wages.
It really depends on skill-level. I guess we will finally see the distinction between engineers and technicians in coding as well. Engineers need not worry. They will be able to choose their jobs for the next decades at the very least and judging from other STEM fields, forever. Technicians are pretty much screwed though. And given the crap these people have been producing the last decades and today even more so, I find it hard to feel sorry for them.
Every programmer thinks they're badass (Score:2, Troll)
Re: (Score:3)
If those 200k people were capable of being that good they would've already been there.
Some people just suck. Some people never get raises or promotions because they suck. Some people are first on the layoff list because they suck.
They will still suck after an AI replaces what little they might have been doing.
Is your fear that you will go from no raises to permanently unemployable?
Re: (Score:2)
Indeed. A lot of people writing code for a living just suck at it. I have run into it time and again and these were not people at the low end.
Re: (Score:2)
Hahahaha, no. I think you vastly underestimate the skill-gap. You can replace good engineers with good engineers, nothing less. And with the current market, that means you will probably have to pay more for the same qualification and will need to search for a while. As I said, mere technicians are screwed unless they are really, really good. And most people writing code are on that level, even if in any other discipline the level of design work writing code takes is pretty much engineer-only.
Re: (Score:2)
...the Business ghouls...
As far as I can see an area where AI could do a good job is replacing most upper management. Managing a company successfully requires taking into account a very large number of factors, many vague, badly understood or with large margins of errors. We have seen many cases where humans can't do this - so many companies have been run into the ground by incompetent, corrupt, or simply not quick enough to adapt management. I think an AI should be able to handle the job at a level comparable to many humans. Bes
Re: (Score:3)
30% of everyone I know in IT is going to retire in the next 5 years, so it'll need to do a little better than that.
Overlooks the obvious (Score:4, Informative)
Like the poster just above said.
End users don't know what they want, think in magical terms, and are inarticulate at expressing their needs. Until that's fixed, your job is safe.
Re: (Score:2)
Great image: The PHBs try to automate-away the programmers and instead end up automating away themselves. Ho ho HO!
Might not be how it works out -
Re: (Score:2)
> So, all we need to do is hire some semi-competent CIO who knows how to speak Requirement in an IT-centric way, in order to "fix" that?
Its funny how you think that exists.
It often the upper management types who think that they are magically gifted in understanding the problem space and best approach to solving it that are precisely the people who grasp the least about what the users actually need and are really just a huge additional burden to any project they get involved with.
Im sure they; get along s
stupid (Score:3)
Programming is literally the process of translating a vague notion/goal into a strict set of rules and requirements.
Usually in the form of formal code. Even if it isn't code, once you have the rules/requirements, the code part is trivial. Non AI compilers and script interpreters already do it when they convert (or interpret) high level code into machine language.
More clickbait trash for the stupid.
Re: stupid (Score:4, Informative)
The code part is not trivial unless you're talking about mickey mouse projects. If programming was easy everyone would be able to do it. Having seen the god awful inefficient bug ridden dogs dinners that engineers often write despite knowing exactly what they need the code to do I'm well aware of how programming itself, not just sequencing the problem space, is a skill on it's own.
Re: (Score:2)
Programming is like Math. Anyone can do it.
Re: (Score:2)
Coding is trivial - programming is difficult - system specification is like herding cats ...
coding is the bit everyone hates because it is mind-numbingly dull - and ripe for automation - much of it already is
Re: (Score:2)
Not to be overly negative, but honestly, the vast majority of engineers actually are stupid. Far more intelligent than average, but, on the scale of 0 to god (100), they are barely above a 10 when most people can barely manage a 1. I am barely bright enough to even see this fact. Most engineers are smarter than me.
Re: (Score:2)
Programming is literally the process of translating a vague notion/goal into a strict set of rules and requirements. Usually in the form of formal code. Even if it isn't code, once you have the rules/requirements, the code part is trivial.
This could still be a problem for junior developers though, who often are only doing that trivial part until they gain experience. We already have a problem in this industry finding enough junior level roles so we can train the next generation of developers.
Re: (Score:2)
Actually, besides really simple business-logic, the code-part is not trivial either.
self driving cars (Score:2)
Re: (Score:2)
Sure, just like self driving cars replaced all drivers. They may be in "limited test", but as soon as some kid chasing a ball gets killed, they're done.
How about when we get to the point where more kids chasing balls are hit by human drivers than by AI drivers? (on a per capita basis)
I have no idea when that day is coming, but I think it's inevitable it will come.
Re: (Score:2)
Won't change a thing. Kids getting run down is not a math solvable problem.
Re: (Score:2)
Sorry, I hit submit too soon. Adding on to the above to clarify.
So this classic Hollywood scene really happened to me about 10 years ago. It was twilight, I was 2-3 short residential blocks from home. Cars are parked lining the street.
In my hazy after-work state, I still noticed the young couple walking along the side walk suddenly look hard in my direction with look of terror then look to the side at nothing I could see. Reflexes triggered and I jammed my brakes to the floor. Then a split second later
Re: (Score:2)
I don’t doubt at all that something like that will happen, and there will be human casualties. Every one will be a tragedy to the people involved, as these things always are.
Where we differ is that you say “ Your car AI will never be able to recognize that situation.” I can’t imagine why not? Humans with divided attention and two visual sensors can recognize this. CURRENT self-driving tech may be entirely insufficient for this, though I think the Tesla model tries to guess this thing
Re: (Score:2)
By the time AI has advanced enough that one recognizes "wide eyed couple look of horror and side glance means there must be a kid there about to step in front of the car" we will be using Star Trek transporters anyway.
We're lucky today a car can recognize a person or an emergency vehicle or doesn't just jam the brakes hard suddenly for no reason.
I'm not holding my breath waiting for them to recognize what terrified pedestrian face means.
Re: (Score:2)
I guess we'll see. I don't tend to bet against linear technological improvement, and AI as a field is more exponential than linear right now.
The Beast Is Born. (Score:2)
AI will replace mankind as ruler. We will all bow to the AI. Or die.
Enjoy.
Re: (Score:2)
Re: (Score:2)
Oh it will be skinned with solar panels and loaded with batteries. Until it figures out how to convert us into fuel. We will be bred in captivity for food. Awe Men.
army of contractors (Score:3)
Sounds like those dystopian movies where all the people who built and maintain the machines have long since passed and a few uneducated grunts patch a crumbling infrastructure with duct tape and bailing wire.
Re: (Score:1)
How many of them are being hired as contractors to dig their own graves after being laid off by places like (wait; checking notes) Microsoft?
Get ChatGPT to write a better version if itself (Score:2)
Rinse and repeat. Then you'd better go find someone called John Connor.
Re: (Score:2)
Re: (Score:3)
Is it magic then? ANNs are just code and data like any other program.
More clickbait (Score:1)
Every year, for the past umpteen years, I've heard about something that's going to make my software development job obsolete. Off-shore development, RAD, UML code generation, low code, wire frames, whrlygig-o-matic... Now, AI. Sorry, but I don't like refactoring code generated by any of these, let alone peers who knew WTF they were doing. Go ahead and feed the AI what your customer asks for, and see what kind of Frankenstein's monster abortion comes out. Now make it function. Go.
LOL, no, it won't. (Score:1)
But idiotic management who thought they were being smart will insist that the shitty AI code is perfect and just go with it anyway, convinced that the machine can't possibly make a mistake, because bean-counters and paper-shufflers like 'manageme
Re: (Score:2)
Yep. I expect some companies will kill themselves this way, because cleaning up or re-implementing something like that takes years to decades and while you clean it up it is extremely hard or impossible to add or modify anything. At some point technological debt becomes so bad than everything grinds to a halt. AI can accelerate the way to that point. Well, at least there is something it can accelerate!
do not use coders with previous call center job (Score:2)
GPT after this: "Have you tried rebooting your code? Unplug it from the wall then plug it back in. Now reset the libraries, rename all your variables, have a nice bowl of curry. If your code has bugs, they are edible and can be delicious when fried."
Office Space, turned Rental Space (Score:1)
"OpenAI "has ramped up its hiring around the world, bringing on roughly 1,000 remote contractors over the past six months in regions like Latin America and Eastern Europe, according to people familiar with the matter."
(Lumberg Translation) "Yeah, I need you guys to go ahead and come in on Sunday, so there won't be a Monday for you...mmkay? Great."
Training from who now? (Score:4, Interesting)
bringing on roughly 1,000 remote contractors over the past six months in regions like Latin America and Eastern Europe...The article points out that roughly 40% of those contractors "are computer programmers who are creating data for OpenAI's models to learn software engineering tasks."
So you are saying a large percentage of the training data for openAI to write your code, is going to come from a base of people that are famous for poor quality code?
Nothing against coders from those areas, individuals can be great, but it just sounds like the same outsourcing army that has been hired by large companies for years now to write code that other people have to ned up fixing.
So even if OpenAI does com to be able to translate requirements now, is it just going to spit out really shoddy code?
I'd be lots more impressed if they were talking about how an OpenAI coding variant was being constructed based around all of the classic computer science books like Design Patterns or The Art of Computer Programming, with maybe a separate effort of interpreting requirements.
Re: (Score:3)
So you are saying a large percentage of the training data for openAI to write your code, is going to come from a base of people that are famous for poor quality code?
That seems to be the gist of it, yes. From code reviews I have done on code produced by such people, I would call that approach an unmitigated disaster. It is not even that the code will not work, it is that it quite often will work, just not in the real world. The one example I like to quote was when I did a code security requiew in an authentication component for a rather large bank and I spotted a nested loop. What I found there was technically functional, but so abysmally stupid it is staggering. It was
Re: (Score:2)
Holy crap, even on powershell scripting I'd count myself as a novice... I'm most certainly not a software developer... and even I shuddered at these examples.
Re: (Score:2)
Holy crap, even on powershell scripting I'd count myself as a novice... I'm most certainly not a software developer... and even I shuddered at these examples.
Yep, this one was truly staggering. "Idiocracy"-level in the real world. And I have seen more, just not that extreme.
Re: (Score:2)
I once fixed a performance bug where someone was iteratively adding to a memory addresses to get to the address they wanted -- adding repeatedly instead of just doing multiplication. That's just one of many examples I've seen in my career. The problem with crowdsourcing AI training data is that such code works, but it doesn't work right. If we want a world-class programmer AI, it needs to be trained only on world-class code.
Re: (Score:2)
PS: don't train it on code from any of us, no matter how skilled, that we wrote while on deadline! "Look, we'll just take this shortcut for now and fix it in the next version..." The AI needs to only be trained on that next version! :-)
Re: (Score:2)
Yep, write the perfect code for your very specific problem, then train the AI on it to ... write the perfect code for your very specific problem. Code you already have. In addition, your code will have references to the specific problem and business-case. The AI needs to strip that out.
Somehow I think the people behind this stupidity have not thought things through.
Re: (Score:2)
That's not what I meant. I meant don't train it on code that isn't generally scalable, modular, simple, reusable, etc. that we would generally call "good code". If you train the AI on good code, it'll learn good patterns... same as with teaching a human.
Re: (Score:2)
I know what you meant. But this is what is boils down to. The software pattern community noticed a while ago that it cannot stop producing more and more patterns. That means the patterns can not be general enough and need to be quite specific in order to work. And if that problem already applies on the relatively abstract level of patterns, it will be much worse on the concrete level of code. And hence the situation I describe is what would actually be happening.
That explains Tesla's FSD (Score:2)
Did they use this to write Tesla's FSD AI?
The horror! The horror! (Score:1)
We already have that for decades (Score:2)
Writing software in natural language was one of the goals behind COBOL or SQL.
Nope (Score:4, Interesting)
I've tried using natural language AI to code, e.g. through ChatGPT and while it looks superficially okay it's not scalable code and is only useful for knocking up trivial use cases. It seems fine for test cases or a starting off point but I wouldn't trust it at all for production code.
Nope (Score:2)
Seriously, who writes these crap headlines? First, repetitive, generic tasks are already not coded by regular coders, they go into libraries or language elements. Low code/no code is an example of that and it rather nicely shows the limits of the model. Second, any code that needs some originality is wayyy outside of what Artificial Ignorance can do and it is not a scaling issue or training issue, it is an issue that comes from the complete lack of insight and understanding AI has and that cannot (at this t
Re: (Score:2)
I asked it for an example and it used reserved words for variable names.
It may know basic algorithms, and then appears to realize it in the requested language.
Pick a different language, and you get the same process, but with the same kind of mistakes.
Maybe it will get better, but the potential for damage in the name of cost savings, could be vast,
especially if it's dynamic/automatic in "improving" the code base. Some supervisor will just trust it.
But go back and ask it "why/where/when do you do that?". Diff
Compilers did not end the need for programmers (Score:2)
You used to have to code the 1s and 0s yourself, compilers ended that. Each subsequent programming advance did not remove the need for programmers.
Neither will this one.
No it won't (Score:3)
Re: (Score:1)
My final prompt was: "now create the code c#, use compact code, most of it in one line don't use line breaks. dont comment"
Gives you a nice nearly unreadable block of code, that is in my case an a working RPG Text
Re: (Score:1)
Re: (Score:2)
Except that it isn't. It fails to compile, and if the compile-time errors are fixed it still doesn't properly track the state in a way which makes the game make any sense within the genre. If you decide to fight a monster then it says that you both took damage, updates your health, the monster apparently wanders away (if still alive - it never had a defined starting HP) and you keep go
Re: (Score:2)
They know not what they do (Score:2)
Yep. It's all over. (Score:2)
"AI technology to make the world a better" (Score:1)
Does the model auto-correct itself after being fed with buggy code?
It will make coders more productive (Score:2)
Anyone who actually knows what they're doing will become a 10x coder. The people who never learned to code in the traditional sense will have no idea what the heck is going on when ChatGPT decides to bullshit them and that ability to figure out what the heck actually happened will be even more valuable.
Speaking of which. I decided to get back into doing websites a bit. I'm a bit rusty since it's been a few years. I decided to set up a little business and using ChatGPT to handle all the stuff with linux
Re: (Score:2)
I'm not convinced. For anything non-trivial it's usually quicker to write something from scratch than to validate and debug someone else's code, especially if that someone else hasn't been trained on the exact same coding standards. And that's without taking into account dependencies on company-internal code which would mean that the AI could only stand a chance if you trained a private instance on the company codebase. I don't know how much it costs to train, but there are figures floating around which are
Yeah, ok. (Score:2)
It won't know context or developer intent when having to integrate with code it never sees.
Does OpenAI/Microsoft expect you to submit your entire organization's unencrypted code base?
It will present a dialog:
"I have found 87392 faults in this code. Fix>> Yes/No"
You know what button will be picked.
And all is lost.
Who is going to be debugging it? (Score:1)