No offense but this sounds like it was written by GPT.
/s not /s
Just covering my bases. I really can't tell anymore.
Ask HN: GPT4 Broke Me
161–170 of 183 posts
Re: Ask HN: GPT4 Broke Me
#162Earlier quoted context omitted.
I don't want to be too descriptive, but it's not a CRUD app - it's a multi-proc app where each process acts as both a client and server (obviously in separate threads) connected w/ gRPC bindings and has an internal DSL that must be abided to. A communication framework of sorts governed by some very specific rules. Also, why is there some notion it can only do simple things? My understanding is it's trained on a large…
I don't think your last paragraph is conceivable actually. Consider what's happening when you give it a prompt. It receives some set of tokens through the prompt and then takes a sample from a very accurately weighted distribution of its very large training set to generate the most likely next set of tokens. When you feed it back in an error in the code it generated, all it's doing is generating a new sample from a n…
In regards to the last paragraph, I saw it happen again and again. Sometimes it wouldn't even be a directly obvious error, but an exception was thrown seemingly unrelated to the actual issue, a proximate cause. Of course that could be related to some SO post that is referenced, but who knows. There definitely is debate about emergent behavior in GPT-4 and it's not quite clear how some aspects work internally.
While it did create quite a few bugs and a couple random hallucinations... and sometimes it would update some piece of code without telling me other parts of the solution were updated (ie. choosing to change protobuf def without being asked to), it was able to fix everything even if I needed to give it a bit of help.
I don't think there is anything I'm doing that is truly novel in isolation, and it's not a lot of code (just a few hundred lines), but I doubt there's anything quite like it in totality out there.
Re: Ask HN: GPT4 Broke Me
#163An artist friend the other day posted a meme to the tune of "Artists will be displaced when clients are able to express their desires clearly. I'm not worried for my job."
Programmers are really artists that way. Our job is not to program computers. Our job is to figure out the client's needs. The implementation on the computer is a matter of craft. That is, it's a skill we've practiced and honed, but it's not the essence of our job. It's not the artistry, any more than holding a paintbrush is the artistry.
It's why I've long discouraged programmers from thinking of themselves as computer jockeys. A lot of developers pat themselves on the back for skills that aren't really that important -- jobs that computers can do better than we can. That was true even before AI, such as the interminable worry about "optimizations" that are better solved by compilers, libraries, and hardware.
I don't doubt that some artists will be replaced by Canva, with worse but acceptable results, and that some developers will be replaced by AI. But there's still a lot of room for us human beings to do what we're actually good at -- being human ourselves and knowing what other humans need. The more you think of yourself as somebody who writes code and are not a "people person", the more you should rightly worry for your job.
Re: Ask HN: GPT4 Broke Me
#164The more I use GPT, the less I'm worried. It is a tool, and a good one, but not a replacement for the thought required to design an app that will function, scale, and have good UX to result in a marketable product. So use it and enjoy its benefits while letting it help you perform even better. As far as everything else you've said... oof, you need a break. You seem focused on money and ego. Maybe it is time to simpli…
This is correct to a degree, but consider what doors GTP is opening. Its slow erosion of our responsibilities. If AI can do some of your work with minimal supervision then sooner or later managers will figure it out and reduce you job scope. Get an intern to do it. We are certainly not at level where you can talk to chatGTP and give it requirements to generate code, but who knows where we will be in 5,10 years. As a…
> Its slow erosion of our responsibilities. If AI can do some of your work with minimal supervision then sooner or later managers will figure it out and reduce you job scope. Get an intern to do it.
No.
Quite the reverse.
AI is a super intern.
Both super productive, and super clueless.
It's the interns (and possibly their managers) who are at risk.
AI is a liability in any area where you can't afford to slip up.
Re: Ask HN: GPT4 Broke Me
#165There are a lot of developers in denial here. I write code for scientific applications. I used to hire/contract out some work to other devs. In the last 6 months I haven't needed to hire anybody else, doing so would have been a waste of money. This is 100% because of chatgpt. Many devs are hired by other devs and these people are aware enough to know they don't need you anymore. Other developers have been spoiled by…
If you write code like a robot(read: junior developer level), you will be replaced by a robot. IMHO, lots outsourcing work is manual work, outsourced to save time. Some of it is to save costs. But the output from outsourcing is often horrible and reads like GPT code. And people will figure - hey, if I have to pay for this level of code, with not GPT?
Re: Ask HN: GPT4 Broke Me
#166Just remember that these LLMs like ChatGPT hallucinate.
We still need experienced engineers to see these mistakes and fix them.
If anything, using ChatGPT will help with productivity and reduce the workload that is pushing us at burnout levels.
On the job security front, even if the LLMs get better at coding, they still need technical people that can articulate what is needed.
You will be fine, just take a deep breath and enjoy the present.
Re: Ask HN: GPT4 Broke Me
#167Re: the mortgage, this will be a problem only if you cannot maintain your salary and the housing market collapses.
Otherwise, if you buy a house, and later you decide you don't want to (or cannot) pay for mortgage, you can just sell your house and buy a smaller one (or move back to renting).
Sure, the risk isn't zero, but if you keep renting and the housing bubble keeps going on you may also burn a lot of money in the end. Nothing is risk free, and you're in a better situation than most.
Re: Ask HN: GPT4 Broke Me
#168There are a lot of developers in denial here. I write code for scientific applications. I used to hire/contract out some work to other devs. In the last 6 months I haven't needed to hire anybody else, doing so would have been a waste of money. This is 100% because of chatgpt. Many devs are hired by other devs and these people are aware enough to know they don't need you anymore. Other developers have been spoiled by…
If you write code like a robot(read: junior developer level), you will be replaced by a robot. IMHO, lots outsourcing work is manual work, outsourced to save time. Some of it is to save costs. But the output from outsourcing is often horrible and reads like GPT code. And people will figure - hey, if I have to pay for this level of code, with not GPT?
So what will happen to junior developers and those developers whose code is mostly just plumbing?
Re: Ask HN: GPT4 Broke Me
#169Earlier quoted context omitted.
> It is a tool, and a good one Not even a good one imo, I ask it to cite its sources and it makes up the URLs 95% of the time.
That makes sense - it is an LLM, not a reference library. Part of using a tool well is understanding what it is good for and what it is not. If you are .looking for citable references... or even full factual accuracy, it is the wrong tool.
What is it good for?
Re: Ask HN: GPT4 Broke Me
#170Hard to take this post and the people in it with good faith. OP feels like a troll, and everyone posting how AI is replacing workers or is so amazing won’t post any of the code or talk too in detail about what it’s supposedly done.
Meanwhile, copilot x routinely spits out code with bugs, fumbles refactoring, misinterprets which variables are needed, and in general is a helpful but annoying autocomplete or template tool. Will it replace workers? Probably, but they weren’t really doing much to begin with then. You still need people who understand the business and domain, the systems in place, how they interact and connect, etc. I don’t believe a LLM is capable of that.
Also, people mentioning GPT10 or whatever might as well say it will be doing magic. There is no way to know how it will develop. How the models will be gatekept. How government policy will change or stifle it. And let’s not forget that progress isn’t exponential or linear forever, it’s generally logarithmic.