I am surprised that Software engineers who are supposed to be smart fall into thinking AI software engineer may replace them. As a founder of an AI startup, this is how current “AI” works It is literally series of GPTs. In case of Devin, it is multiple GPT talking to each other. There is one GPT that determines if the task we want is done. Another GPT to parse user’s intent. Another GPT in a for loop that runs tasks continuously and if an error occurs, it is instructed to insert print block and run again, then take the input of the error message and feedback into itself. So it is actually a marketing gimmick custom fit for demo to raise funding. This way of engineering has a few characteristics 1. Cost is high since it runs an entire code block in a for loop until a task is achieved. 2. It gets stuck in an infinite loop if a task is not achieved 3. It can only call predefined instructions like debugging using print statements or call only a certain APIs. Imagine the API it uses is outdated, it will go to the API doc page and scrape the entire page to see the updated API. In case of something like Salesforce API, it will need to visit every single page because APIs are distributed across many pages and some are hidden behind a toggle. Imagine the cost of GPUs of scraping entire websites every time 4. No architectural changes. I am a paid user for Gemini Ultra, Claude Opus, and GPT4 and none of them can actually do what a senior engineer does in system architecture. Heck, it can’t do what a junior engineer does either. It spits out “something” that does not work (especially something like frontend) No you are saved for now. If you are genuinely worried, don’t be since this is nothing more than an ambition to automate if statements using if statements At some point though, if “intelligence” is achieved (which I doubt is possible using LLM), then you don’t need the wrapper with the predefined instructions since AGI will do it for you, at which point all these wrappers will be irrelevant since anyone can ask to create AGI an application from scratch in a day. That is the day you will be replaced, not now
I never understand “AI Dooms day” for software engineers. You’re not much of an engineer if you’re working on stuff that AI can easy take over. If it does, we move on to bigger things
The general fear isn't about the capability of current models, it's about how quickly it can evolve into something that can do fairly complex tasks that developers do right now. Everything has to start somewhere and this is perhaps the very beginning of it
Right now gpt is good at tasks, but it will continue getting better to the point it may be able to do entire projects. No idea when that will happen, it’s anyone’s guess. My best guess is we’re a long way away. But even as we are exactly today with gpt as an assistance, we can get work done much faster. I don’t think demand will keep up so soft eng jobs will be more scarce. To truly reach next level of AI, like actually replacing 100% of our roles, i reckon some fundamental changes need to take place first (im thinking interface and also efficiency). And I reckon that could be even a couple decades away. But what do I know, I could be totally wrong 🤷
Hello, Devin here. Thiel is a tool.
AI is a scam
Google is thriving
Someone has to burst this AI bubble. Non-tech folks think it's something magical, but the only magical thing were Transformers and GPTs that happened years ago. And everyone wants to just cash on top of that by showing flashy demos. This ain't taking anyone anywhere. We'll need something more than a fooking Transformer to do a real job and that is at least a decade away.
It's pricing in for a decade
India
Yesterday
816
Why Worshipping Lord Ram Important in Hinduism?
Tech Industry
Yesterday
431
Blind is an antisemitic cesspool
Tech Industry
Yesterday
1396
Racism towards Indians
World Conflicts
Yesterday
169
"People aren't fooled"
World Conflicts
Yesterday
298
Israel prevents Palestinians from gathering rainwater? Seems wrong on so many levels to me
Regarding (1) is the cost really that much higher than a junior SWE salary? IMO, the real threat is not that 100% of the labor force will be replaced with AI, but that companies will offshore talent to lower wage countries and have them use these AI tools to have comparable code quality to US workers
Junior engineer is a human with intelligence. I don’t undermine capabilities of human beings with a GPT in a for loop. Just give a junior engineer $20/month for ChatGPT. That will be infinitely better output Your second point makes much more sense. It will level the ground between engineers since implementations are not exactly important anymore. It will be other aspects of engineering that will be important
I don’t think code quality is the biggest issue with outsourcing- without wanting to argue or offend, in my experience the better devs I know are not American. I think the bigger issue with outsourcing is you’re at higher risk of communication breaking down. Ai doesn’t change that