Shhh don't blow our big secret. Software developers haven't actually been doing anything for decades! If the C suite finds out they just need to know their business and poof software we'll all need new jobs. (Heavy sarcasm)
Yea who needs to know if their code actually works anyways? We're looking at you aerospace, automotive, process control, and anyone with a customer SLA.
I can see it now "chatgpt, AWS says our spending is ten million in one week. What went wrong?" "As a chat bot I..."
Well you've certainly cherry -picked a vanishingly small amount of the code in the world. And in industries which generally have a budget for hiring professional programmers.
Generative AI cannot tell the difference between fact and fiction.
Just because it can generate code does not mean the code will work (error-free) or be safe (vulnerability-free) or not be under copyright by someone else.
Children who are 11 and 12 years old are physically capable of driving cars. Giving them keys and vehicles does not make them competent racecar drivers and the results could be quite bad.
Similarly, giving people a way to generate code does not make them programmers and, depending on the application, it might be very bad.
I agree, however, one difference between those other modalities is that with code, you have a compiler to tell you it’s wrong. This means:
1. The ai can self correct
2. The programmer can learn
As another commenter said, not everyone needs “good” code and not all code has to scale to hundreds of devs and millions of people to be extremely (millions in revenue) valuable
A compiler will only catch some errors and is language dependent, it will not catch logical errors or edge cases. There are ample examples in very limited contexts like SQL queries.
Certainly the LLMs have not been able to scale the size of an application they can write yet. I'm not sure the current techniques will be able to without significant human involvement by someone who can already program. Some other break through is likely needed for generative AI to reach the next plateau of capabilities.
>> not everyone needs “good” code and not all code has to scale to hundreds of devs and millions of people to be extremely (millions in revenue) valuable
The problem is that neither the AI nor the average person can tell good code from bad code. This is the classic problem of "the blind leading the blind".
Imagine generating code to build a business only to discover that the generated code is easy to hack into, putting the business at risk. Or that the generated code does not actually do what the requester asked for, causing subtle errors that are not detected until later.
Image making critical business decisions using inaccurate reports that were created using AI-generated code.
If you value your business would you trust an AI that cannot tell fact from fiction to write code your business will rely on? You would be beter off having the AI write your marketing plan than have it write your business applications.
This happens with humans too. Humans write code with bugs, technical debt, security flaws, etc.
I would trust an AI to generate code a lot faster than a junior engineer with 4 months experience. That’s the trade off, not AI vs a 10 year experience architect.
>> This happens with humans too. Humans write code with bugs, technical debt, security flaws, etc.
>> I would trust an AI to generate code a lot faster than a junior engineer with 4 months experience. That’s the trade off, not AI vs a 10 year experience architect.
That's not what the Nvidia CEO is selling. He is saying that the AI replaces programmers. Not junior engineers, not experienced architects, but programmers in general.
Jensen Huang said: "The programming barrier is incredibly low. We have closed the digital divide. Everyone is a programmer now - you just have to say something to the computer."
His statement is blatantly false. He is just trying to sell Nvidia hardware and tools. If he really believed what he said, he would be firing Nvidia programmers to reduce Nvidia's costs.
I don't read it that way at all. I read it like "The person who has never programmed can now write code"
It says nothing about replacing anyone, more about expanding the pie and lowering the barrier to entry. No where did he say those programmers "will be senior engineers and the senior engineers will be out of work" He didn't even imply it.
programmer and open source contributor here. Because you aren't the only one to tell this "alternative facts" in order to do marketing for your products ( i.e. the Microsoft guy told "we can literally double your programmer overnight, read "an half of your programmer in redundant" ), because I read here on HN that companies are firing people also because subjects like you do these kind of affirmations, because I also read big companies are reducing salaries also because apparently there are those "instant programmers" they can unfreeze when they need them, I have some consideration:
- what if you piss off a large part of the open source community and we stop to publish our code, like the screenwriters' strike in progress ? No more patches for six months, for example.
- What if all the public code is removed from on-line repositories: no new code is added anymore ? What will you use to training your AI stuff ? Your internal code? So we can finally have Nvidia drivers equivalent to the closed source :-) ? Time ago, I read about a "Endgame" initiative, I forgot the reason but then people started to remove code from Github.
- I was reading here a twit of a guy , open source contributor, who realised that copilot ( or other equivalent, I have to look up ) was copying his code 1:1, I don't like if that this AI stuff became a layer of indirection where the license disappear. Our software is open source but that don't authorise anyone to steal the work of people spending their time ( often they free time ) to contribute to community projects. Stuff like that is convincing me to change my level and modalities of contribution!
So, mr Nvidia chief, sir, you are spitting in the plate where AI is training, I don't think it's a good policy.
Unfortunately this seems to be the only option - either making open source licensing more restrictive or depriving these people from the code all together. It’s not the first time large corporations benefit from people’s good will and it won’t be the last.
simply not true. i've tried several times to get ChatGPT to write working emacs lisp. it rarely can write anything long, even after i correct it five or six times. anything to push the price down and the profits up, i guess.
Hard disagree with this. The current state of AI is impressive but it's not at the level of allowing everyone to program without prior knowledge. I suspect we're still a long ways away from that even though it seems close by all the progress made thus far. It'll be kind of like self driving cars being right around the corner. Maybe in the next decade we'll get there.
This is really interesting. Because if you actually think that Huang believes what he means then Nvidia stock should be worthless shortly. No really, if you think that he genuinely believes this then we're about 6 weeks away from Nvidia firing all their software engineers, hiring in a bunch of cheap idiots to talk to their magic AI, and as a result completely tank the company. If you believe what he says. What I think is that all the real running in the AI story has been done by incredibly well-qualified engineers with cumulative millennia of experience. Yes, the crypto crowd (an old nvidia favourite) are out on twitter screaming about how they can generate the mona lisa through these 10 weird hacks... but is that really the important story? Or is that a side show?
If anything I would say this trend actually makes engineers more valuable. We can do away with people who write code, and focus on where the real value is - people who can write the right code, in the right way, at the right time. Value software engineers, not coders.
Data science companies used to do this too. They'd advertise "who needs data scientists anymore?! They just don't deliver, use our automl" then you'd check thier hiring page and they'd have thirty open data science positions available. So much snake oil it's awful
27 comments
[ 3.8 ms ] story [ 52.0 ms ] threadI can see it now "chatgpt, AWS says our spending is ten million in one week. What went wrong?" "As a chat bot I..."
https://simonwillison.net/2023/May/27/lawyer-chatgpt/
Generative AI cannot tell the difference between fact and fiction.
Just because it can generate code does not mean the code will work (error-free) or be safe (vulnerability-free) or not be under copyright by someone else.
Children who are 11 and 12 years old are physically capable of driving cars. Giving them keys and vehicles does not make them competent racecar drivers and the results could be quite bad.
Similarly, giving people a way to generate code does not make them programmers and, depending on the application, it might be very bad.
1. The ai can self correct
2. The programmer can learn
As another commenter said, not everyone needs “good” code and not all code has to scale to hundreds of devs and millions of people to be extremely (millions in revenue) valuable
Certainly the LLMs have not been able to scale the size of an application they can write yet. I'm not sure the current techniques will be able to without significant human involvement by someone who can already program. Some other break through is likely needed for generative AI to reach the next plateau of capabilities.
The problem is that neither the AI nor the average person can tell good code from bad code. This is the classic problem of "the blind leading the blind".
Imagine generating code to build a business only to discover that the generated code is easy to hack into, putting the business at risk. Or that the generated code does not actually do what the requester asked for, causing subtle errors that are not detected until later.
Image making critical business decisions using inaccurate reports that were created using AI-generated code.
If you value your business would you trust an AI that cannot tell fact from fiction to write code your business will rely on? You would be beter off having the AI write your marketing plan than have it write your business applications.
I would trust an AI to generate code a lot faster than a junior engineer with 4 months experience. That’s the trade off, not AI vs a 10 year experience architect.
>> I would trust an AI to generate code a lot faster than a junior engineer with 4 months experience. That’s the trade off, not AI vs a 10 year experience architect.
That's not what the Nvidia CEO is selling. He is saying that the AI replaces programmers. Not junior engineers, not experienced architects, but programmers in general.
Jensen Huang said: "The programming barrier is incredibly low. We have closed the digital divide. Everyone is a programmer now - you just have to say something to the computer."
His statement is blatantly false. He is just trying to sell Nvidia hardware and tools. If he really believed what he said, he would be firing Nvidia programmers to reduce Nvidia's costs.
It says nothing about replacing anyone, more about expanding the pie and lowering the barrier to entry. No where did he say those programmers "will be senior engineers and the senior engineers will be out of work" He didn't even imply it.
programmer and open source contributor here. Because you aren't the only one to tell this "alternative facts" in order to do marketing for your products ( i.e. the Microsoft guy told "we can literally double your programmer overnight, read "an half of your programmer in redundant" ), because I read here on HN that companies are firing people also because subjects like you do these kind of affirmations, because I also read big companies are reducing salaries also because apparently there are those "instant programmers" they can unfreeze when they need them, I have some consideration:
- what if you piss off a large part of the open source community and we stop to publish our code, like the screenwriters' strike in progress ? No more patches for six months, for example.
- What if all the public code is removed from on-line repositories: no new code is added anymore ? What will you use to training your AI stuff ? Your internal code? So we can finally have Nvidia drivers equivalent to the closed source :-) ? Time ago, I read about a "Endgame" initiative, I forgot the reason but then people started to remove code from Github.
- I was reading here a twit of a guy , open source contributor, who realised that copilot ( or other equivalent, I have to look up ) was copying his code 1:1, I don't like if that this AI stuff became a layer of indirection where the license disappear. Our software is open source but that don't authorise anyone to steal the work of people spending their time ( often they free time ) to contribute to community projects. Stuff like that is convincing me to change my level and modalities of contribution!
So, mr Nvidia chief, sir, you are spitting in the plate where AI is training, I don't think it's a good policy.
If AI is more like calculators then we will see no effect (calculators didn't have an impact on accountant salaries).
If they are more like proper replacement they should have a significant negative impact in the near future.
If anything I would say this trend actually makes engineers more valuable. We can do away with people who write code, and focus on where the real value is - people who can write the right code, in the right way, at the right time. Value software engineers, not coders.