this post was submitted on 26 Oct 2024
604 points (99.2% liked)
Apple
17597 readers
74 users here now
Welcome
to the largest Apple community on Lemmy. This is the place where we talk about everything Apple, from iOS to the exciting upcoming Apple Vision Pro. Feel free to join the discussion!
Rules:
- No NSFW Content
- No Hate Speech or Personal Attacks
- No Ads / Spamming
Self promotion is only allowed in the pinned monthly thread
Communities of Interest:
Apple Hardware
Apple TV
Apple Watch
iPad
iPhone
Mac
Vintage Apple
Apple Software
iOS
iPadOS
macOS
tvOS
watchOS
Shortcuts
Xcode
Community banner courtesy of u/Antsomnia.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
People are treating AI like crypto, and on some level I don't blame them because a lot of hype-bros moved from crypto to AI. You can blame the silicon valley hype machine + Wall Street rewarding and punishing companies for going all in or not doing enough, respectively, for the Lemmy anti-new-tech tenor.
That and lemmy seema full of angsty asshats and curmudgeons that love to dogpile things. They feel like they have to counter balance the hype. Sure, that's fair.
But with AI there is something there.
I use all sorts of AI on a daily basis. I'd venture to say most everyone reading this uses it without even knowing.
I set up my server to transcribe and diarize my my favorite podcasts that I've been listening to for 20 years. Whisper transcribes, pyannote diarieizes, gpt4o uses context clues to find and replace "speaker01" with "Leo", and the. It saves those transcripts so that I can easily switch them. It's a fun a hobby thing but this type of thing is hugely useful and applicable to large companies and individuals alike.
I use kagi's assistant (which basically lets you access all the big models) on a daily basis for searching stuff, drafting boilerplate for emails, recipes, etc.
I have a local llm with ragw that I use for more personal stuff like, I had it do the BS work for my performance plan using notes I'd taken from the year. I've had it help me reword my resume.
I have it parse huge policy memos into things I actually might give a shit about.
I've used it to run though a bunch of semi-structured data on documents and pull relevant data. It's not necessarily precise but it's accurate enough for my use case.
There is a tool we use that uses CV to do sentiment analysis of users (as they use websites/apps) so we can improve our ux / cx. There's some ml tooling that also can tell if someone's getting frustrated. By the way, they're moving their mouse if they're thrashing it or what not.
There's also a couple use cases that I think we're looking at at work to help eliminate bias so things like parsing through a bunch of resumes. There's always a human bias when you're doing that and there's evidence that shows llms can do that with less bias than a human and maybe it'll lead to better results or selections.
So I guess all that to say is I find myself using AI or ml llms on a pretty frequent basis and I see a lot of value in what they can provide. I don't think it's going to take people's jobs. I don't think it's going to solve world hunger. I don't think it's going to do much of what the hypros say. I don't think we're anywhere near AGI, but I do think that there is something there and I think it's going to change the way we interact with our technology moving forward and I think it's a great thing.
So here's the path that you're envisioning:
Someone wants to send you a communication of some sort. They draft a series of bullet points or short version.
They have an LLM elaborate it into a long-form email or report.
They send the long-from to you.
You receive it and have an LLM summarize the long-form into a short-form.
You read the short form.
Do you realize how stupid this whole process is? The LLM in step (2) cannot create new useful information from nothing. It is simply elaborating on the bullet points or short version of whatever was fed to it. It's extrapolating and elaborating, and it is doing so in a lossy manner. Then in step (4), you go through ANOTHER lossy process. The LLM in step (4) is summarizing things, and it might be removing some of the original real information the human created in step (1), rather than the useless fluff the LLM in step (2) added.
WHY NOT JUST HAVE THE PERSON DIRECTLY SEND YOU THE BULLET POINTS FROM STEP (1)???!!
This is idiocy. Pure and simply idiocy. We send start with a series of bullet points, and we end with a series of bullet points, and it's translated through two separate lossy translation matrices. And we pointlessly burn huge amounts of electricity in the process.
This is fucking stupid. If no one is actually going to read the long-form communications, the long-form communications SHOULDN'T EXIST.
Also neither side necessarily knows the others filter chain. Generational loss could grow exponentially. Not only loss but addition by fabrication. Each side trading back and forth indeterminate deletions/additions. It's worse than traditional generational loss. It's generational noise which can resemble signal too.
So if I receive a long form then how do I know if the substantial text is worth reading for the nuance from an actual human being. I can't tell that apart from generated filler. If a human wrote the long form then maybe they've elaborated some nuance that deserved long form.
On the flip side of the same coin. If I receive a short form either generated by me or them. Then to what degree can I trust the indeterminate noisy summary. I just have to trust that the LLM picked out precisely the key points that the author wanted to convey. And trust that nuance was not lost, skewed, or fabricated.
It would be inevitable that two sides end up in a shooting war. Proverbial or otherwise. Because two communiques were playing a fancy game of telephone. Information that was lost or fabricated resulted in an incident but neither side knows which shot first because nobody realized the miscommunication started happening several generations ago.
Yep, pretty much every single “good” use case of AI I’ve seen is basically a band aid solution to enshitification.
You know what’s a good solution to that? Removing the profit motive.
That's not what I am envisioning at all. That would be absurd.
Ironically, an gpt4o understood my post better than you :P
" Overall, your perspective appreciates the real-world applications and benefits of AI while maintaining a critical eye on the surrounding hype and skepticism. You see AI as a transformative tool that, when used appropriately, can enhance both individual and organizational capabilities."
if you believe that ai summary, i have a bridge that i'd like to sell to you.
As the author of the post it summarized, I agree with the summary.
Now, tell me more about this bridge.
do look up the "forer effect" and then read that ai summary again.
Haha, yea I'm familiar with it(always heard it called the Barnum effect though it sounds like they are the same thing), but this isn't a fortune cookie-esque, meyers-briggs response.
In this case it actually summarized my post(I guess you could make the case that my post is an opinion that's shared by many people--so forer-y in that sense), and to my other point, it didn't misunderstand and tell me I was envisioning LLMs sending emails back and forth to each other.
Either way, there is this general tenor of negativity on Lemmy about AI (usually conflated to mean just LLMs). I think it's a little misplaced. People are lumping the tech I'm with the hype bros- Altman, Musk, etc. the tech is transformative and there are plenty of valuable uses for it. It can solve real problems now. It doesn't need to be AGI to do that. It doesn't need to be perfect to do that.
I read this comment chain and no? They are giving you actual criticism about the fundamental behaviour of the technology.
The person basically explained the broken telephone game and how “summarizing” will always have data loss by definition, and you just responded with:
Just because you couldn’t notice the data loss doesn’t mean the principle isn’t true.
Your basically saying translating something from English to Spanish and then back to English again is flawless cause it worked for some words for you.
I'm not saying any thing you guys are saying that I'm saying. Wtf is happening. I never said anything about data loss. I never said I wanted people using LLMs to email each other. So this comment chain is a bunch of internet commenters making weird cherry picked, straw man arguments and misrepresenting or miscomprehending what I'm saying.
Legitimately, the llm grok'd the gist of my comment while you all are arguing against your own strawmen argument.
Here are two cases from your original comment that would have data loss. I get you didn’t use the phrase “data loss” but that doesn’t mean your examples didn’t have that flaw.
Sorry if you view all this as lemmy being “anti ai”. For me, I’m a big fan of ML and what things like image recognition can do. I’m not a fan of LLMs becoming so overhyped that it basically gave the other ML use cases a bad name.
The problem is basically this: if you're a knowledge worker, then yes, your ass is at risk.
If your job is to summarize policy documents and write corpo-speak documents and then sit in meetings for hours to talk about what you've been doing, and you're using the AI to do it, then your employer doesn't really need you. They could just use the AI to do that and save the money they're paying you.
Right now they probably won't be replacing anyone other than the bottom of the ladder support types, but 5 years? 10? 15?
If your job is typing on a keyboard and then talking to someone else about all the typing you've done, you're directly at risk, eventually.