LLMs caused the layoffs
Let's face it, there is no need for tens of thousands of software engineering jobs if modern ChatGPT can code almost everything you can describe in human language. Sure, some roles are still needed to oversee the high-level picture and architect the solutions. But the point is, your ordinary day-to-day job of creating classes, functions, and whatnot is disappearing, much like the roles of human phone operators did. You might, of course, be lucky enough to find a place where your Director or VP is an empire builder and you serve just as an excuse for their organization to exist. Moreover, we are moving away from programming anyway. Soon, virtual assistants will be the front end of most online stores (e.g., Alexa) or social networks (e.g., Meta glasses). Aside from some low-level infrastructure roles managing databases, a programmer's job will be focused on fine-tuning language models by feeding them extra data. By the way, this is known as Software 2.0, described by Karpathy in 2017.
Op is correct. All companies are reducing current staff slowly by layoffs and not hiring back
Every time I see "moreover" in a sentence I think it was written by GPT
Oh no LLMs are replacing even Blind users
Nice try CEO/HR
OP must be working in sales.
Can't confirm that but can confirm that they are really stupid
It might not happen immediately, but, yes the long term trend in this direction. It is expected that GPT-5 will be able to deliver small projects end to end. Think about this, most tech jobs are working on glorified CRUD applications. Although, I think while LLMs reduce headcount, this also opens up opportunities for programmers to built something new. Now, since you don’t need many people to write code, you will see many new startups popping up, taking technology where it was not possible before. It means that newer startups will be prime to shakeup many traditional sectors. Economy is not in a steady state, new jobs will also be added, hard to make any solid predictions about the future. Overall, fun times ahead.
You still need human to verify every piece of that code. You have to realise LLM is just a stochastic parrot. At the moment it still fails to correctly predict the import paths in my React code. Even non-AI extensions have gotten this right like 5 years ago. The reason is ,it's just following a statistical pattern. I have checked codes in suggested by Copilot because I thought it was simple enough and didn't bother to double check. Each time, it was either the solution was too convoluted or wasn't solving the problems properly. Every single time I have to double check . It will replace many junior Dev and grunt work. But it's still far away from replacing an experienced Dev.
That’s exactly what I said. IT WILL NOT HAPPEN TOMORROW. Although, just imagine, gpt-4 is way advanced than anything we knew 5 years ago. It will keep on improving until it plateaus. Think about this, first smartphone, it had slow processors, meh camera and what not. At that point, you can argue that it is no where close to replacing an actual Digital sony camera, but look at the smartphone cameras now. 121 million digital cameras were sold in 2019, 9 million in 2021. Sure, it would not replace all humans, however, you would wrong to think that work 90% of sde do would not be replaceable, it would happen. It is a natural curve, quite possible we will be able to find something more meaningful to do.
I don't think any serious SWE would write this. Just like the ignorant BE folks who think AI will replace front end because it can generate HTML code. People should learn a domain properly before trying to have a strong opinion on such subject/domain. Unfortunately,the internet has turned everyone into experts in anything they barely know.
You sound like Paul Krugman: "By 2005 or so, it will become clear that the Internet's impact on the economy has been no greater than the fax machine's."
It will have impacts, it already has so many impacts. That's not close to what I'm saying I'm specifically talking about replacing software development. My team work on inference, so I have an idea how customers are using this and what it can do. It's going to replace many low level grunt work. I know that. I think that's one of the main problems. There's many places LLM would shine beyond imagination,but we're trying to shoo it in everywhere.
It’s a junior engineer’s biggest asset and worst enemy. You can supersize your productivity, learn faster, and at the same time the prowess of a LLM has the highest return in an experienced engineer’s hands. Asking a LLM is one thing, telling an LLM to do something is another. If you can convey intent accurately you can generate value. You can solve and code any leetcode problem with an LLM if you know how to convey intent. If you just copy and use what is spit out then you were easily replaceable to begin with.
Have you ever tried to use GenAI to make more than a simple script? More than 50 lines of code? It fails miserably! Even on simple tasks it takes more time to create the correct prompt than to write it yourself
Hi NASDAQ why don't you share with us if we should buy PUTs or CALLs on SP500?
That's why I find it hard to believe any SWE would think this way. Even for 50 lines, you need to validate. You need some knowledge to know it's doing what you asked it to do correctly.
I've been using an LLM for code recently, and it's so hit or miss. There's no way in hell it could replace an engineer right now.
Yawn. We’ve been hearing this for two years now. Meanwhile, not a single respectable company has replaced engineers with LLMs
No respectable company will admit that. Why do you think they are laying off people and not hiring back?
The idea is not full replacement of SWEs... It is all about making SWEs more efficient. My expectation/suspicion is that we are seeing the end of L3s and lower end of L4 style roles. We will have more L5/L6 level ICs (than managers), who has a broad view/understanding of the end system and drive towards it in a much smaller but experienced team. This does not mean the end of software engineering.... It is more akin to programming in Python relative to people writing code in assembly. More complex software systems still needs trained/experienced SWEs. Vanilla programming as such will be relatively different redundant. This might take several years to a decade to get there.