Since it is so quite on the board these days I thought I would inquire about future state views for a group that does populate the board here.
Any of the BA tech guys wondering if Devin is going to start steeling their playtime money?
Pos job is pretty rudimentary. Some actually are not even located in the US. Now a subscription service that you don't have to pay commision to for each client? Could be enticing.
Well OK, I tried making it kgirl related. Lol.
Will Devin as it is now take any nonentry level swe jobs? Not a chance in hell. Devin is a crude tech demo that basically uses Ai prompts itself. Could the tech in 2-3 years? Quite possibly. With any big non-open source codebases, it will take companies a long time to adopt and trust the Ai in general. So provided the tech is improved drastically from now, the reality of mid level and up FAANG engineers getting cut probably doesn't happen at least another five years out. That's my estimate and it might be too liberal, but I'm a skeptic by default.
One thing about Ai and llm learning, is that improvements are made incrementally. This is fine in text, image and even video generation as inconsistencies and small "mistakes" can bypass human eyes and perception. A mistake in a large scale software deployment can be incredibly costly. And if it's a mistake on design level, oh boy.
Should folks eyeing entry level jobs in next few years be worried? Yes. Ai can learn syntax and patterns and write basic crud apis faster and generally better than a human entry level swe. Because it's a pretty basic thing to learn and output. It follows strict protocols and generally accepted rules. Put Ai on something more complicated as debugging multithreaded legacy proprietary software that has dependencies on undocumented internal libraries... oh that will be fun.
Something to keep in mind is that the more proprietary or siloed-out a piece of tech is, the harder will be for Ai to fully "learn" it. Ai learns from context. When there's a lot of lack of context, it's a problem.
Sorry for the wall of text. And maybe it was completely off topic, maybe there's some escort named Devin who is amazing... then I apologize for this whole post. Lol.
Fit as a PO doesn't seem right -- some other AI yes.
Agree, no one needs to have bad dreams tonight about it but in 5 years? Well, maybe. And things move pretty fast here (how long ago was GTP3? Not years I think.) Then again by that point the Japanese dolls will likely have AIs running and either China or SK will have some bioengineered skins that make the doll look and feel real/natural. Maybe the 200/hour rates could return 
Oh, man, think about it. One robot k-girl could work around the clock, 24 - 7. The agencies would make a fortune. Bring it on!
However, the fears I have with AI are considerable. I think giving anyone a voice on the internet is generally a good thing.
Giving anyone an ability to create any content that is non-distinguishable from real life is scary. In around five years I expect the internet to become a wild wild west unlike it has ever been before. Add to the fact that generally people get their news and info off internet, believe curated material and are now complacent and can be easily misled.... I'll just say my "crusade" as you call it will probably be meaningless in that many years. The degree of misleading, misinformation and garbage content designed to mislead/troll/scam, will be ridiculous.
Anyone will be able to create thousands of fake reviews, all different in style and content, simulating real mongers. Anyone will be able to create unique, completely fake Ai generated pictures videos and so forth.
This is why already there are huge ethics debates. Tbh feels almost like 1984 (the book). There will likely be a ministry of truth established to tell people what can or cannot be trusted. That or Ai tech will be safeguarded somehow.
Anyhow. I have a lot to say about Ai but feels like it's very out of place on this board. I'm sorry for off topic. It's something that hits home.
perhaps TER needs to team up with, or borrow from, WorldCoin and get a TER Orb?
Certainly agree that AI introduces lots of benefits (I will be interested in seeing just how good Devin can get and what the pricing and how service is delivered as there are a number of things I would do if I just didn't hate the tedious nature, for me at least, of coding shit.) as well as some significant risks. But I'm not nearly as worried as some about it. But your point about all the fake content potential in all areas of our lives is of concern.
Edit -- and are you really dissing the original StarTrek where several of the episodes clearly domonstrated the future would have very sexy and desirable androids?
-- Modified on 3/14/2024 12:28:38 PM
If you need some small specific task done but don't want to code it, gpt is already good at it. Usually these tasks are done in a scripting language like python and using well known libraries. All you need is having succinct enough prompt. I think most people don't realize this.
Or maybe they realize it but also understand that if you can describe algorithm of a task well, you actually have done more than 80% of the task. Ie why pseudocode solutions on interviews are almost always accepted
In other words, I don't think Devin is a use case for you specifically. Devin is an attempt at creating an AI software engineer with everyday responsibilities. In your case, I don't think you want to maintain a codebase, manage pull requests, find and fix bugs, use issue tracking systems, etc. But I might be wrong.
Yeah, I know about using one of the GPTs for getting some simple bits done -- and have at times used that approach. And sometimes it's good for a different "set of eyes" on the structure of an algorithm you've started with.
And yes, my interests would actual lead to all those headaches you mention -- even the simple shit I do I end up running into the change management aspect -- and largely ignore it so find my self either reinventing wheels, light "reverse engineering" of old code I write to figure out just what I need to do to update something or just starting from scrach again. In other words, my design patterns really suck from the maintenance perspective I.e., not using functions where I should, not making things as modular as they should be - not actually planning the shit out beyond the I'm at X and going to Q and then taking a random path to the destination. LOL
Of course it's a bit sad that I'm sitting here discussing this rather than K-girls but the board seems largely dead -- and people are even asking about non-K providers here.
It's pretty interesting that you need change management etc for what I assume is very granular purposes.
Honestly for scripts I do for basic tasks - including hobby tasks such as this hobby, I am a pretty bad pattern designer too. I don't really bother with making it elegant or concise - as long as it works, it works. If I need to do something fast, I'm not going to experiment with performance or take the most efficient algorithm possible because input constraints are usually quite lenient in these tasks as input is very finite.
Yes and no. The recent pain point was wanting to implement some changes in training and testing some ML models I'm playing with. The AIs suggest some specific things. But when trying to implement them turns out my enviroment is not the same as assumed by the AI (differing versions of python and scikit packages) so upgrade. Well that is not so straightforward as say updating MS Office where old version goes away and everything is updated to new version. So do I manage multiple versions of python and packages? Do is do that with virtual enviroments (do I need to learn all the little tricks and nuances of either approach?) After some more digging around I have (I think) resolved the problems and stayed on the current version of python. But still will need to figure out the plan to upgrade as some day what I have is not going to keep working. The upgrade/change management functionality in pip and PYPI only gets me so far (but really happy it is there!)
I would think something like Devin would do a better job with that, know what to do and just do it (once it is actually able to work as intended). But, that's mostly moot at this point as seems that this change/maintenance aspect of SW work and SW AIs may well be a big weakness even if Devin turned out to be a really good coder for highly complex tasks.