Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: GPT4 Broke Me

news.ycombinator.com

Post reply on HN
Print view View original

131–140 of 183 posts

Re: Ask HN: GPT4 Broke Me

#131

Post by e12e »

brandall10 wrote: post #44
e12e wrote: post #23

This strikes me as a little absurd, but perhaps it is common for "developers" to misunderstand what value they actually add. AI won't change the value you add, any more than autocomplete or code generators and scaffolds do. The vast majority of programming is about facilitating, automating and improving business processes. The program artifact only has value as a tool to enable some business processes. A website faci…

"Now, with LLMs, a single system architect/senior developer might be able to do the work of a five person consultancy alone. You might not be able to charge five times as much - but perhaps three times for the same or fewer hours worked?" This is the point that underscores why the OP is stressing out, and it really is underselling the value GPT adds. Ignore a consultancy. Consider a team of 5 you might be on. Esp. fo…

> Boom, less engineers needed to produce novel products or features, more engineers on the market, depressed wages.

Possibly. But software is funny - increased software productivity tends increase demand for software. Perhaps the job market will be saturated, but perhaps the market will simply grow.

Re: Ask HN: GPT4 Broke Me

#132

Post by gizmo »

thaway_thaway34 wrote: post #112
gizmo wrote: post #82

Earlier quoted context omitted.

Today, chatGPT can do zero percent 0% of the actual programming I do. I use chatGPT to find a library function or to find command line options and other simple stuff. But actual programming? ChatGPT can only write completely trivial scaffolding code. I don’t understand how people get a significant productivity boost out of it when chatGPT can’t do design, can’t do algorithms, can’t do concurrency, can’t do architectu…

You’re probably not talking to it right way. You need to build up the story. It doesn’t look up information, it only maintains coherence. You need to bring up key points for your design, have a chat. Then start asking for code. Unless you have a starting code, then give that and chat on top of it. I am getting at least 5x boost in productivity. Just last night, I solved 7-8 bugs consecutively.

My prompts are pretty basic but also I have to spell everything out or the output will be wrong. I can poke and prod until I get something approximately correct, but I have to give step by step instructions for the simplest things like you would when coaching a slow student.

Re: Ask HN: GPT4 Broke Me

#133

Post by kaba0 »

brandall10 wrote: post #44
e12e wrote: post #23

This strikes me as a little absurd, but perhaps it is common for "developers" to misunderstand what value they actually add. AI won't change the value you add, any more than autocomplete or code generators and scaffolds do. The vast majority of programming is about facilitating, automating and improving business processes. The program artifact only has value as a tool to enable some business processes. A website faci…

"Now, with LLMs, a single system architect/senior developer might be able to do the work of a five person consultancy alone. You might not be able to charge five times as much - but perhaps three times for the same or fewer hours worked?" This is the point that underscores why the OP is stressing out, and it really is underselling the value GPT adds. Ignore a consultancy. Consider a team of 5 you might be on. Esp. fo…

I am honestly baffled — what kind of program are you working on that can be AI-programmed?

In my experience most CRUD apps are ultra-trivial for the most part in a language with sane ecosystem, you literally just glue together things. That’s one day either with ChatGPT or without - the real benefit are the ecosystem (as per Brooks), that’s not the bottleneck (besides the occasional “this doesn’t work together with that because..”, to which chatgpt is just as susceptible if not more)

Re: Ask HN: GPT4 Broke Me

#134

Post by thaway_thaway34 »

gizmo wrote: post #129
thaway_thaway34 wrote: post #108

Earlier quoted context omitted.

Yeah, but how do we protect and grow expertise when it can be easily copied in GPT-(n+1)

I think it’s more like truck drivers panicking in 1995 they will soon be obsolete after having seen the first self-driving demo. Truck drivers will get automated away eventually, but not today and not next year. Self-driving systems get better every year. It’s been 30 years. We’re getting close but we’re still not there yet. I expect the LLM revolution to be somewhat similar. The models will get better but they won’t…

You're right. I am definitely biased as a long time techie. There is one downside though in our own sector in comparison, we are ready to give our data the fastest and the environment we work in allows for it.

Re: Ask HN: GPT4 Broke Me

#135

Post by rationalist »

buddhistdude wrote: post #32
rationalist wrote: post #17

Earlier quoted context omitted.

Your statement can have multiple interpretations. I assume you mean this one: Your fear will be a self-fulfilling prophecy; to prevent that, stop being afraid? Of course that's like telling someone who is depressed, to "just be happy".

I think he means that thoughts have an actual effect on the unfolding of the physical world.

Which is a "self-fulfilling prophecy", also known by a new-age term as "manifestation".

Re: Ask HN: GPT4 Broke Me

#136

Post by thaway_thaway34 »

wopwops wrote: post #28

Draw salary as long as you can, save as much as possible and buy land outright somewhere cheap. No mortgage. I saw where it was all headed back in the early 2000s. I was in my mid 30s and feeling the grind back then. Rent? Fuck that. Your life is on the line and this is for all the marbles. You find a way to go rent free. Maybe do like the bums are doing all over the place, live in a van, etc. Nobody seems to care. S…

That's what I am thinking of doing if things go for worse. I am planning to do in 4-5 years time. I already have enough to buy a small house in the country side back in my home country. I could be building my own projects for fun.

I think you nailed it. My current fear arises from the fact, I am at one of the highest HCOL cities in the world in comparison to wages. This makes the fear more imminent, as I basically would have to move if my salary goes significantly lower.

That wouldn't be the case where I am originally from, in Scandinavia.

Re: Ask HN: GPT4 Broke Me

#137

Post by brandall10 »

kaba0 wrote: post #133
brandall10 wrote: post #44

Earlier quoted context omitted.

"Now, with LLMs, a single system architect/senior developer might be able to do the work of a five person consultancy alone. You might not be able to charge five times as much - but perhaps three times for the same or fewer hours worked?" This is the point that underscores why the OP is stressing out, and it really is underselling the value GPT adds. Ignore a consultancy. Consider a team of 5 you might be on. Esp. fo…

I am honestly baffled — what kind of program are you working on that can be AI-programmed? In my experience most CRUD apps are ultra-trivial for the most part in a language with sane ecosystem, you literally just glue together things. That’s one day either with ChatGPT or without - the real benefit are the ecosystem (as per Brooks), that’s not the bottleneck (besides the occasional “this doesn’t work together with th…

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 portion of existing opensource work beyond just SO posts and the like. It seems too many folks saw it produce code w/ some bug or hallucination. The correct response to this is not "welp, my job is safe". It's to feed that error back in have it correct what is broken. If it builds the wrong thing, explain what is wrong, provide multi-shot examples.

You could conceivably create an agent that takes the generated output w/ tests, runs them locally, then feeds the error back in until there is fully working output. Right now the context window limit would be an issue compared to using the web interface. Perhaps OpenAI or Github will provide that capability as a feature.

Re: Ask HN: GPT4 Broke Me

#138

Post by ljm »

brandall10 wrote: post #44
e12e wrote: post #23

This strikes me as a little absurd, but perhaps it is common for "developers" to misunderstand what value they actually add. AI won't change the value you add, any more than autocomplete or code generators and scaffolds do. The vast majority of programming is about facilitating, automating and improving business processes. The program artifact only has value as a tool to enable some business processes. A website faci…

"Now, with LLMs, a single system architect/senior developer might be able to do the work of a five person consultancy alone. You might not be able to charge five times as much - but perhaps three times for the same or fewer hours worked?" This is the point that underscores why the OP is stressing out, and it really is underselling the value GPT adds. Ignore a consultancy. Consider a team of 5 you might be on. Esp. fo…

With that amount of experience I imagine GPT is explaining stuff you already know and is thus saving you time.

For something completely novel, or as someone with little experience, it’s going to be harder to get that kind of result in terms of understanding the requirements, writing the prompt, and comprehending the output.

Re: Ask HN: GPT4 Broke Me

#139

Post by tomlue »

iJohnDoe wrote: post #122
tomlue wrote: post #99

Earlier quoted context omitted.

One very common task we have is writing and maintaining ETL scripts. This was never a particularly difficult task, just time consuming and normalized enough that I could outsource it. It's not that chatgpt can do the whole task, it's just that I can do it myself much faster now. Paying somebody to set up a new source that is consistent with our protocol is just too expensive relative to the 1-2 hours it takes me to d…

I don’t know much about the ETL industry. Are these ETL jobs for all sort of data? Accounting data? Log data? Getting the data from point A to point B? You mentioned UI. Is the UI so the customer can now move the data around or to visualize the data.?

I would not get overly focused on this use-case. It is more difficult to think of use-cases that won't get disrupted than those that will.

Re: Ask HN: GPT4 Broke Me

#140

Post by thaway_thaway34 »

tarmol wrote: post #65

I am a bit concerned too, the current state of tech is nowhere near to replace a senior engineer, but where it will be 5 years from now is a scary thought. Yesterday I tried to get GPT4 to generate me a Dovecot Sieve filter to move .ru email to Junk and mark it seen, it failed 3 times and hallucinated some syntax that looked plausible, but actually is not supported by Dovecot. My hope was that I would not have to spe…

I do use Google in combination with GPT4 and give tips to it, to help. But it cut down my Google searches by a huge factor.

About in 5 years, imagine all of us, using Chatgpt in that time...along with all the enterprise integrations of it directly into workflow of companies through JIRA etc. (if not chatgpt, closed models could be trained in that time)

AI could even learn corporate politics from all the Slack, Jira etc. conversations. If this hype survives 5 years, and we keep putting data into it, it will only get more powerful from here on.

Post reply on HN