[aussieguy1234]

Remember, with GPT-4 around its alot easier to work with a language or technology you only have minimal exposure to. It's like having a Principal Engineer around 24/7 that you can ask questions to unblock you. So dive right in. You can also use it to practice the interview before the actual interview.

[madeofpalk]

I believe only not-so-strong developers actually believe this, after actual experience with this tech. That's just like saying that "Stackoverflow is like having a Principal Engineer around 24/7" which could be true to a degree, but it still doesn't make up for not knowing stuff.

[aussieguy1234]

This tells me that you haven't used GPT-4 (Thats not the free ChatGPT), or have not learned how to prompt properly. You can't ask Stack Overflow follow up questions. If the code from stack overflow does not work in the way you expect it to or the answer is not 100% accurate, you can't ask Stack Overflow to correct it.

Also, its better in the hands of Senior Engineers who can evaluate answers for correctness, because just like Google/Stack overflow, it can be wrong.

[wizzwizz4]

On the contrary: the system contains a canned comment describing just how to ask a follow-up question!

> If you have a new question, please ask it by clicking the Ask Question button. Include a link to this question if it helps provide context.

And if the answers to an existing question aren't what you expect:

> You can also add a bounty to draw more attention to this question once you have enough reputation.

[verbify]

My experience with GPT-4 is that it's been amazing for python and mediocre at DAX. It frequently hallucinates functions, gives code that can't run, or does not fit the prompt.

It is still helpful, but there's a big gap between how helpful it is, depending on the tech.

[dukeyukey]

It's more like having an overeager intern around 24/7. Good at spitting back documentation at me or finding something from Stack Overflow, but really struggles with much else.

Ise use ChatGPT and Copilot all day, they're incredible useful! But let's not overstate their power.

[aussieguy1234]

I don't use copilot so can't vouch for its accuracy but I have heard it produces code with security holes a large percentage of the time. I do use GPT-4. ChatGPT (GPT 3.5) is not that great. GPT-4 blows it out of the water. Let's take SQL as an example.

As a software engineer/founder, I've used it to write all kinds of advanced queries for Tunnelmole that get all kinds of metrics which previously would have require a Data team. You can write a prompt like "here is my table structure, give me a query to get my abandoned cart rate".

If I ever hire engineers for my project, prompting is a skill I'll be looking for.

[NicoJuicy]

Gpt-4 isn't an engineer. It's a stack overflow summarizer.

If you would do new things, it won't be able to help you.

[aussieguy1234]

The idea that LLMs like GPT-4 are glorified stochastic parrots who are merely predicting the next word has been thoroughly debunked, almost no one in the field believes that anymore. Whats been proven so far is they actually understand your questions and their responses. Its also proven that they have some level of reasoning capabilities.

I highly recommend using it. Think of it as a Google replacement. Every engineer out there uses Google. Instead of spending hours on Google, you can spend a minute or two doing something new with the help of an LLM. They can be wrong, which is why they are better used in the hands of Senior Engineers.

[NicoJuicy]

Nah, their text classification is better then their text generation.

Just let it do something novel instead of something typical stack overflow.

Eg. Nest, a c# library for elastic search, is something it has consistent troubles with because it's not with "and" or "or", but should, must and filter.

But yeah, you won't get into those cases as a junior or doing typical stuff it has been trained on :)

[sirsinsalot]

If your principal is like GPT-4, I'd consider this a red flag.

Your comment is an advanced version of "fake it till you make it". That's another red flag when I'm hiring disgustingly highly paid consultants.

[aussieguy1234]

If the person writing the questions is a Senior Engineer already and knows how to evaluate the answers properly for correctness, you probably would not be able to tell, even if they've only had minimal exposure to let's say, Ruby or Python as opposed to JavaScript/TypeScript. Or writing advanced SQL queries when they don't do this very often.

[sirsinsalot]

You're assuming the red flag relates to technical correctness. It doesn't. It's a red flag about mindset and diligence.

Using genAI is fine, using it to bolster a lack of underlying knowledge as I read it, is a red flag.

[aussieguy1234]

Most engineers will come across something they haven't used before in most roles. Perhaps some legacy system in some dying language, for example. Previously, they might have spent hours on Google. Now, GPT-4 can unblock them in seconds.

It cant replace the mindset of a human, but what I'm basically saying here is with GPT-4 and good prompting skills, you can be alot more brave when it comes to new unfamiliar tech. That's an advantage in a fast changing tech landscape.

[wizzwizz4]

GPT-4 doesn't work for legacy systems in dying languages. It only "works" for things that are well-documented on the internet, or described in what books were included in the dataset.

I can't think of a situation where you'd go from spending hours on Google to being unblocked in seconds. If you're spending hours on Google, then your first dozen search queries aren't turning up the information – in which case, GPT-4 wouldn't turn up the information immediately, either! (It would say something, but it's unlikely to be based on a true story.)

[te_chris]

I can’t work out if your responses are being written by GPT or not.