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

Command line functions around OpenAI

kadekillary.work

Post reply on HN
Print view View original

151–160 of 316 posts

Re: Command line functions around OpenAI

#151

Post by angarg12 »

roenxi wrote: post #10

Being efficient at writing code can barely make someone a 5x engineer. A team of people who write code effectively is easy to do. Hyperproductivity in software is all about deciding what problems to tackle. Richard Hipp isn't worried about chipotle restaurant orders in golang, he's worried about how to store data reliably. That isn't a coding puzzle that ChatGTP is likely to help with. Either ChatGTP can do the whole…

I discussed my experience using ChatGPT in HN before and I was accused of being skeptical. But it isn't true at all, I think LLMs are a wonderful tool that will change software development. However I'm not worried about losing my job.

As you say, I find the act of writing code is hardly the bottleneck to my productivity. One of the best case scenarios, LLMs become the ultimate high level language. In that case, Software Engineer roles would switch from coding to more of a mix of old school Software Architect, PM and entrepreneur.

By the way, I find interesting to run the argument of "coding is not the bottleneck" in reverse. What if you design an organization so that coding is the bottleneck? How transformative would LLMs be?

Re: Command line functions around OpenAI

#152

Post by mjr00 »

zeroonetwothree wrote: post #123
roenxi wrote: post #10

Being efficient at writing code can barely make someone a 5x engineer. A team of people who write code effectively is easy to do. Hyperproductivity in software is all about deciding what problems to tackle. Richard Hipp isn't worried about chipotle restaurant orders in golang, he's worried about how to store data reliably. That isn't a coding puzzle that ChatGTP is likely to help with. Either ChatGTP can do the whole…

> A team of people who write code effectively is easy to do. Based on my experience, you’d be surprised how hard it is sometimes. I have a team of 10 engineers that failed to ship a basic CRUD UI after several months (and it wasn’t even close to done). Eventually it got taken over by one individual eng who finished it in two weeks.

I've seen this happen several times before, and it's never been because people weren't typing fast enough: it's always bad or nonexistent management, unclear accountabilities, or lack of technical direction or oversight (one failed CRUD project was written purely functionally by an engineer in Scala + Scalaz, absolute insanity). None of these are solvable by LLMs.

Re: Command line functions around OpenAI

#153

Post by agentultra »

roenxi wrote: post #10

Being efficient at writing code can barely make someone a 5x engineer. A team of people who write code effectively is easy to do. Hyperproductivity in software is all about deciding what problems to tackle. Richard Hipp isn't worried about chipotle restaurant orders in golang, he's worried about how to store data reliably. That isn't a coding puzzle that ChatGTP is likely to help with. Either ChatGTP can do the whole…

Can't agree more.

The problem isn't generating code. It's articulating a functional specification with enough precision and clarity that you can solve for the unknown, which is the program that satisfies the specification.

I agree with the conclusion that tools like ChatGPT won't be replacing programmers any time soon. And it won't be helping us to write programs until it can properly reason about the information it is trained on and prompted with. The problem with natural language is that it is often too imprecise. When you get to concurrency or optimizing data layout an LLM isn't going to understand anything about the problem space and the training data probably won't contain enough examples to produce something remotely plausible let alone good or elegant.

Writing, thinking, and asking the right questions will still be important for quite some time. We're not in the age of having a magical genie in our computers that can do all the work for us... yet.

Although using it as a procedural generation tool for fake data is nice... still not as nice as writing a property test with generators and shrink... but as a little time saver for quick prototyping, neat.

Re: Command line functions around OpenAI

#154

Post by AlwaysRock »

uncletaco wrote: post #35

At what point did we stop making fun of that 10x engineer thread and started taking it seriously? Who forgot that 10x was a silly little thing that made zero sense in the real world?

The article is named after a 50 Cent album. I don't think its taking 10x very seriously.

Re: Command line functions around OpenAI

#155

Post by kodah »

roenxi wrote: post #10

Being efficient at writing code can barely make someone a 5x engineer. A team of people who write code effectively is easy to do. Hyperproductivity in software is all about deciding what problems to tackle. Richard Hipp isn't worried about chipotle restaurant orders in golang, he's worried about how to store data reliably. That isn't a coding puzzle that ChatGTP is likely to help with. Either ChatGTP can do the whole…

I think reaching peak optimization as an engineer is a mix of:

- Knowing or having control of the tools you use

- Knowing what problems are most important to revenue and which are the greatest risk

- Knowing the language well enough to plan ahead without much labor

When a company I worked at mandated an IDE it took me a while to be productive beyond typing speed. When I discovered my company had support for generators, I was no longer spending forever on boilerplate REST code. When I know my test harnesses well it's easier for me to prove functionality in a smaller feedback loop.

Re: Command line functions around OpenAI

#156

Post by sebdufbeau »

esperent wrote: post #6

> First off, you’ll need an OpenAI API key. You can get this by going here and signing up. Note that you'll also need to be from a Western country, or at least have access to both a phone number from one and a VPN. And if you want a plus subscription, you'll also need a US credit card - not stated anywhere but I've been totally unable to sign up for plus with either Irish or Belgium cards, it just immediately flashes…

I signed up with no problem using a Canadian credit card

Re: Command line functions around OpenAI

#158

Post by yosito »

californical wrote: post #135
yosito wrote: post #113

A lot of people seem to be hung up on the title and totally missing how useful this tool is. People are right that the tasks that you're having this tool do aren't tasks that a 10x (or 1000x) developer would do. But what they're missing is that this is exactly why you can use this tool to multiply your productivity. If you don't have to spend your time writing silly scripts anymore or converting JSON to yaml or whate…

It possibly enables you to build the wrong thing, 10x faster. Do you think the company will let you just have fun learning for the other 90% of the time that you can free up? They’ll just expect more output. Being able to write code 10x faster will likely not free your time enough to become a 10x problem solver.

No, it frees you up to spend your energy on thinking about what the right thing to build is. If your head is buried in code all the time, that's when you can't stop and look at the bigger picture and ask if you're building the wrong thing or not.

Re: Command line functions around OpenAI

#159

Post by sebastianconcpt »

b7r6 wrote: post #95
roenxi wrote: post #10

Being efficient at writing code can barely make someone a 5x engineer. A team of people who write code effectively is easy to do. Hyperproductivity in software is all about deciding what problems to tackle. Richard Hipp isn't worried about chipotle restaurant orders in golang, he's worried about how to store data reliably. That isn't a coding puzzle that ChatGTP is likely to help with. Either ChatGTP can do the whole…

I'm aware that I'm in the minority here, but I find that investing heavily in being able to type fast and handle the tools well actually changes what code I write and not just how quickly. The faster I've gotten at moving code around, the more subconsciously willing I've been to try something with a modest probability of working out, and more importantly the easier it's gotten to throw away a draft that isn't working…

Yes, but being the faster foot shooter in the room doesn't cut it either.

Smartness isn't wisdom.

AIs are, at most, producers of synthetic smartness.

Re: Command line functions around OpenAI

#160

Post by synergy20 »

shagie wrote: post #116
synergy20 wrote: post #15

Earlier quoted context omitted.

change gpt-4 to gpt-3.5-turbo worked, gpt-4 api is only for beta testers and you need be on the waiting list to get approved. I tested the scripts and they work really well, esp like the image generation script. However unlike the web-browser the scripts does not keep the context, which is not great. is there a way for shell script to maintain context session somehow?

> is there a way for shell script to maintain context session somehow? Feed the output of the previous prompt back into itself. If you look at https://platform.openai.com/playground/p/default-chat?model=... you will see that the prompt for the next line is all of the previous generation conversation. This can consume tokens at an accelerating rate.

"Human: you seem to remember the context of the chat, how many context you keep, the most recent previous output, or a few conversations

AI: I generally remember the last few conversations and I'm also capable of keeping context in longer conversations."

I don't think there is a way to feed context back to the script efficiently, maybe need a local database or a queue of certain length to sustain the subject. To mimic the web-browser experience at terminal.

the real magic for chatgpt vs old chatbots is that it maintains context, how to keep that via API calls is something I do not fully understand yet. Need read the API reference I guess

Post reply on HN