Also, in one of their points they argue that, in the case of non-programmers, the more domain expertise one brings to the table, the higher the chances of a successful outcome. Well sure, but what's the quality of the output? Someone with little to no coding ability cannot meaningully judge a piece of source code. It might not matter when we're talking about one-off scripts and such, but for anything more serious, somebody really needs to be in the driver's seat, 'cause this ain't a Waymo yet.
Not exactly overthinking, more like misinterpreting. Claude is not a person. LLMs don't think, want, feel, or have any kind of inner experience whatsoever. They just generate words (or tokens, to be more precise) based on a complex statistical algorithm. The algorithms tries to generate along the same patterns it saw in its training data. This means that the responses you saw are nothing but echoes of the impatience of other people, as captured in diverse forms of text (for example reddit conversations).
Your line of thinking rests on the assumption that "AI enables machines to understand English", but this is false. AI, or more precisely LLMs, are a class of algorithms that take a string as input, and return a new string generated through statistical black magic. They don't enable the machine to do anything in particular. You can't ask the LLM to run some arbitrary computation directly on the CPU, for example. If that were possible, your argument would hold some water, at least to some extent.
Small startup. I don't use a laptop. I prefer work/home separation, so I use a workstation, and I'm in the office 5 days a week. I have three screens attached to my PC. A main 27" (slightly curved) screen in the middle, horizontally, and two 24" screens turned vertically, on each side. I also use a tiling WM for managing desktops and windows. I have arrived at this setup, that I really enjoy using, because I have a lot of screen real-estate, but I don't really need to rotate my head much, just glance a little to one side or another. Tiling WMs also allow you to create muscle memory for where each one of your tools is, and you can get there instantly, rather than having to cycle through everything all the time.
It's a bit of an over-generalization, though. Tech stack matters more and more, the larger a project gets, and the more people join it. Some languages and frameworks lend themselves better at long-term maintainability, as the code base gets larger. Other put so many foot-guns at your disposal, that you need a lot of discipline to keep things manageable. And, of course, once you start working together with others, the choice of language and framework becomes much more consequential, as it will limit who can join the project, how quickly they can become productive once they join, etc.
There are a few layers to this problem. They're not entirely wrong about developments in AI-assisted coding – it is progressing quickly. But for now, LLMs doing "any software engineer job" is still far off. At least in the way people like your brother mean it: that some non-technical person can just tell an AI "build <this or that> for me," and the AI handles it end-to-end.
Until we reach that stage (if ever), technical people will still be needed in the loop – to check that the AI's work is correct, integrates properly with the rest of the system, and meets the requirements. And this leads to the main role of any engineer: figuring out the requirements. In other words, determining what exactly needs to be built.
People, in general, aren’t great at describing things in a precise, unambiguous way – and non-technical people even less so. I don't see this part of the equation being automated anytime soon.
As long as you enjoy figuring things out, solving technical problems, architecting software systems, and understanding software at a deep level, I think there will be work for you for the foreseeable future – no matter how "AI-assisted" it becomes.
I thought I was depressed, but recently realised that I’m actually feeling off because of long COVID symptoms. The brain fog is pretty much constant, and I’m definitely more forgetful and confused than I used to be. There are some cardiac symptoms as well, but those interfere less with my day to day. I’d love to find a way out of this state, but it seems that my doctor doesn’t keep up with the latest research, and doesn’t seem to think that long COVID symptoms are a thing…
Weird post, could be misleading to others. You really don't need "multiple terabytes of SSD storage" to run Llama 3.1 70B. All you need is about 40GB for the model weights. Plus the tooling around it, which is assumed you've already installed.
I don't see absolutely any incentive for a company to spend more on tracking technologies (whatever their nature may be) in order to lower their profit. So guess who gets the proverbial shaft in this equation?
I guess I missed the news about the license change. We still use it in production in my startup, but now I'm considering migrating to something else. After a quick google search, there seem to be a bunch of alternatives, but this one from Microsoft stands out:
"[...] async should be the hot cake compared to synchronous programming. But that doesn't work out in practice for some reason. If that really were the case, 80% of web apps would have been powered by Node instead of PHP today, right?"
You're not taking into account people's reluctance to change. PHP has been around for much longer than Node has. This fact is practically enough to explain the phenomenon you're pointing to.