|
For a while I believed that everything would be an AI generated mash, as time has gone on I think those fears are overblown. Certainly 2023 and 2024 data will have quite a bit of AI trash swimming in it but I think that more people who use AI to generate things are getting more cautious, double checking stuff and being more selective of what is being output. Also, when it comes to images I think it's a bit more of a concern right now but as tools become better, allowing more control, the images will at least start to become more artistic and more intentional rather than just bland visually appealing images worth of a corporate slideshow. I also view it as a form of human-reinforced selective learning if they do learn on previous data as (hopefully) the data will be the stuff that's chosen to be better, more correct, and higher quality. The way I see it, generative AI is going to bridge some gaps of accessibility in terms of skills and the effort required to use them. Skilled people will simply make better stuff. Though, yes, many jobs will probably be lost unfortunately. I'm not too fond of that.
|
# ¿ Jan 6, 2024 21:27 |
|
|
# ¿ May 12, 2024 17:00 |
|
biznatchio posted:The problem isn't people who are involved in legitimate attempts at communicating with AI generated work, since they'll do the self-curation you mentioned. It's the people who are and will continue to be flooding the internet with non-curated garbage (e.g. this) to game systems intended for humans. Someone dumping woims articles on the internet doesn't give a drat about quality, they have other motivations; and 1) there's no end or limit to what those motivations might be, and 2) machines can spit out garbage far faster than we can effectively filter it out, especially if its garbage that superficially appears legitimate. To be fair though, the net has been full of algorithmically generated swill for decades now be it junk web pages for SEO or spam or low effort nonsense. Maybe the new crud will be harder to differentiate but I'm not sure it's too much different overall. Obviously this might not apply to images or music though.
|
# ¿ Jan 7, 2024 02:16 |
|
KwegiboHB posted:You're the fourth person in not even a single page to say this. I'll ask you too, what are you basing this idea off of? I'm trying to see why the U.S. Bureau of Labor Statistics https://www.bls.gov/ooh/arts-and-design/home.htm is suggesting job GAINS instead or if there's a better source of information I should be looking at. I'll be honest, when I think of job losses, I don't mean art directors and floral arrangers. I mean telemarketers, office admin, paralegals, warehouse workers, draftspeople and a whole swath of semi professional jobs that are incentivized to be replaced. poo poo is moving at a breakneck pace and I think denying that there will be job losses is naive. Sure, some will be created and maybe it'll ease certain ideas around who does what but there will be an adjustment period and I think it'll be painful.
|
# ¿ Jan 7, 2024 18:30 |
|
I will say though that AI poisoning of photos is a concern of mine. For example, my home town: AI image with the prompt "Edmonton, Alberta, Canada": This is the closest image I could get in terms of looking like my city.... It looks nothing like it. It's more a blend of edmonton and calgary, most of the generated photos are. If people are using it to generate images of places or things, it's going to be very wrong and this is going to be filtered back through. The good thing is that it's super-easy to watermark an image with both metadata and in the image to identify it as AI generated. Not sure if Midjourney is actually doing that.
|
# ¿ Jan 9, 2024 04:24 |
|
mobby_6kl posted:I finally had a reason to try ChatGPT for code. Told it to replace a function that printed sensor values over serial with one that saves it in a database using the SimplePgSQL library and it came up with this: So, the thing is that the further you get off the beaten path, like you surmised already, it may not have complete information or it may get 'confused' with a similar library. I tried the same thing with the library and got similarly poor (albeit different) results. However, when I put the .h in or the sample code from the library (I tried both) I get what appear to be very good results. When dealing with relatively niche cases like libraries for comparatively lesser known stuff like ESP SQL libraries you can dump some sample stuff that you know works or should work and it can put it together from context. Interestingly enough, Claude and GPT4 did know what the library was but not exactly how to use it. Claude is a very good tool for this because its context window is so much larger, even the free Sonnet version is very handy for this. What's interesting as well is that when I have bizarre issues that befuddle me, I can give it code, tell it what I'm experiencing and it can usually tell me what the problem is. Especially if it's a programmatic issue, it'll know and tell me 99% of the time. It's an incredibly powerful tool.
|
# ¿ Mar 10, 2024 23:17 |