i actively zone out when anyone higher up than me talks about copilot or chat gpt. i also dressed down a colleague for using chat gpt for a stupid simple task.
I have a more nuanced take. AI is simultaneously untrustworthy and useful. For many queries, DuckDuckGo and Google are performing considerably worse than they used to, while Perplexity usually yields good results. Perplexity also handles complex queries traditional search engines just can’t.
About a third of the time, Perplexity’s text summary of what it found is inaccurate; it may even say the opposite of what a source does. Reading the sources and evaluating their reliability is no less important than with traditional search, but much of the time I think I wouldn’t have found the same sources that way.
Of course there are other issues with AI, such as power usage and Perplexity in particular being known for aggressive web scraping.
Nuance and depth isn’t as popular as I’d like on or off Lemmy.
Ah, but you see, I never claimed AI isn’t useful. In fact, you can check my comment history. I’ve agreed AI is a very useful tool, I still think it shouldn’t be used for ethical, social, and personal reasons
A problem with nuance is that people want to discuss the specifics and nuances of what they care about but for the most part won’t do that on subjects for other people. So you need to tailor your responses to your audience. FWIW on Lemmy I see a lot more instances of people with specificly opposed takes where both sides have similar vote counts. So while it’s not perfect it’s better than most
I think ddg and Google are performing worse because of AI. Pushing their AI services and the tsunami of AI slop make a search harder than SEO did and deprioritizes fixing it.
I think current state-of-the-art AI is useful for when you are not having a novel thought.
I believe that AI, at least in the form of LLMs, is currently incapable of novelty in the sense of creating a new concept or a new thought with reason and purpose behind it.
For instance, if I was going to write a book, I might consult with LLMs about how to fill in the slow gaps or the dead spaces in my storyline and to fully come up with a completely fleshed out story that I would then write without its assistance.
My assumption is that anything that it fills in is going to be cobbled together from literally hundreds of thousands of other similar stories, and therefore it will not be new or unique in any way.
If I was really trying to push the envelope, I would then assume that the right thing to do would be that whatever it says is ordinary and common, and if I want to be extraordinary and uncommon, then I need to use that as a launch point for my own gap-filling content.
Therefore, I could use an LLM to write a good story with a new concept, a new premise, a new storyline that is relatively unique and original by using the LLM to clearly identify those things that are not.
I think it is useful with a constrained dataset. Like using it to summarize things about a dataset, or dumping documents into it and asked getting info about it (like Gemini in Google Drive).
It is not useful for general question using the whole-ass internet as a dataset.
Also I wish it was called something other than AI…it’s just a word guesser FFS.
the only useful thing my company and collegues have fold for it is taking meeting notes. it just logs everything and summarizies stuff, and it’s like 90% accurate, but it does make plenty of errors.
however, if i give a presentation with screen sharing, it can’t do shit to summarize that.
how to say with this. I see pretty much an equal split between ai is best thing ever, ai will doom us all, and like ai has some uses and may get more but we need to make sure any use is worth the energy usage.
Would it? I run a science fiction book club and there’re a lot of arguments that if something achieved human level intelligence that it would immediately try to kill us, not become our perfect servants
“It was a morality core they installed after I flooded the Enrichment Center with a deadly neurotoxin to make me stop flooding the Enrichment Center with a deadly neurotoxin.”
As a software developer I fully agree. People bash on it constantly here but the fact is is that it’s required for our jobs now. I just made it through a job hunt and every tech screen I did they not only insisted on me using AI, but they figured how much I was using too.
The fact is is that like it or not it does speed us up, and it is a tool in our toolbelt. You don’t have to trust it 100% or blindly accept what it does, but you do need to be able to use it. Refusing to use it is like refusing to use the designer for WinForms 20 years ago, or refusing to use an IDE at work. You’re going to be at a massive disadvantage to your competing jobseekers who are more than happy to use AI.
I review take home assignments and mostly we receive AI submissions. It’s easy to tell when they aren’t AI though because we get thoughtful comments about why one choice was made over another, and comments on the higher level view that only come from product context and experience. I don’t think one single fully ai-created submission has made it passed the code review part.
See it’s hard as an interviewer because for the first time ever I lost points at one place because I didn’t use AI at all, and they almost didn’t say yes to me. Their feedback quite literally was that it functioned well, but I could have got it done faster with AI.
Seems pointless to test you on anything that could be done by ai, otherwise why even hire someone, just have fewer devs using more ai right? I want to test people on whether they have experience to notice things and make decisions. Idk if they generate the busy work but that isn’t what I’m grading them on
Hey preaching to the choir there, but other companies were saying “if they didn’t use AI for this they won’t here either”. For your interviewees sake, make sure AI use is extremely specific for every step in the interview. I had places where they were upset I didn’t use AI at this step, but did on this other step. It’s batshit out there.
That’s dumbshits using it to do their job for them and trusting the output blindly. If you’re using LLMs to get over the occasional hump they’re awesome time savers.
AI is untrustworthy and shouldn’t be used
I have management talking about copilot usage rates and I hear people casually refer to “what ChatGPT told them” in conversation
i actively zone out when anyone higher up than me talks about copilot or chat gpt. i also dressed down a colleague for using chat gpt for a stupid simple task.
Depends on the subject
The other day on Reddit someone was saying they just fact checked something with ChatGPT.
You can ask ChatGPT to provide sources you know.
Yeah. If ur dummy.
The sources are bullshit https://m.youtube.com/watch?v=_zfN9wnPvU0
I’ve found it very useful as a tool to gather references from talks that don’t cite claims…
It’s like a super search for context. I would never use a LLM to provide logic or reason, and sadly I think many people do.
I have a more nuanced take. AI is simultaneously untrustworthy and useful. For many queries, DuckDuckGo and Google are performing considerably worse than they used to, while Perplexity usually yields good results. Perplexity also handles complex queries traditional search engines just can’t.
About a third of the time, Perplexity’s text summary of what it found is inaccurate; it may even say the opposite of what a source does. Reading the sources and evaluating their reliability is no less important than with traditional search, but much of the time I think I wouldn’t have found the same sources that way.
Of course there are other issues with AI, such as power usage and Perplexity in particular being known for aggressive web scraping.
Nuance and depth isn’t as popular as I’d like on or off Lemmy.
Ah, but you see, I never claimed AI isn’t useful. In fact, you can check my comment history. I’ve agreed AI is a very useful tool, I still think it shouldn’t be used for ethical, social, and personal reasons
A problem with nuance is that people want to discuss the specifics and nuances of what they care about but for the most part won’t do that on subjects for other people. So you need to tailor your responses to your audience. FWIW on Lemmy I see a lot more instances of people with specificly opposed takes where both sides have similar vote counts. So while it’s not perfect it’s better than most
I think ddg and Google are performing worse because of AI. Pushing their AI services and the tsunami of AI slop make a search harder than SEO did and deprioritizes fixing it.
It’s also a way to inflate the number of ads a user has to wade through before they find what they’re looking for. Classic monopolist bullshit.
I think current state-of-the-art AI is useful for when you are not having a novel thought.
I believe that AI, at least in the form of LLMs, is currently incapable of novelty in the sense of creating a new concept or a new thought with reason and purpose behind it.
For instance, if I was going to write a book, I might consult with LLMs about how to fill in the slow gaps or the dead spaces in my storyline and to fully come up with a completely fleshed out story that I would then write without its assistance.
My assumption is that anything that it fills in is going to be cobbled together from literally hundreds of thousands of other similar stories, and therefore it will not be new or unique in any way.
If I was really trying to push the envelope, I would then assume that the right thing to do would be that whatever it says is ordinary and common, and if I want to be extraordinary and uncommon, then I need to use that as a launch point for my own gap-filling content.
Therefore, I could use an LLM to write a good story with a new concept, a new premise, a new storyline that is relatively unique and original by using the LLM to clearly identify those things that are not.
I think it is useful with a constrained dataset. Like using it to summarize things about a dataset, or dumping documents into it and asked getting info about it (like Gemini in Google Drive).
It is not useful for general question using the whole-ass internet as a dataset.
Also I wish it was called something other than AI…it’s just a word guesser FFS.
the only useful thing my company and collegues have fold for it is taking meeting notes. it just logs everything and summarizies stuff, and it’s like 90% accurate, but it does make plenty of errors.
however, if i give a presentation with screen sharing, it can’t do shit to summarize that.
I have people telling me how to do my work because “That’s what ChatGPT suggested, and they’re always accurate”.
🤷
how to say with this. I see pretty much an equal split between ai is best thing ever, ai will doom us all, and like ai has some uses and may get more but we need to make sure any use is worth the energy usage.
Actual AGI would be trustworthy. The current “AI” is just a word salad blender program.
Would it? I run a science fiction book club and there’re a lot of arguments that if something achieved human level intelligence that it would immediately try to kill us, not become our perfect servants
“It was a morality core they installed after I flooded the Enrichment Center with a deadly neurotoxin to make me stop flooding the Enrichment Center with a deadly neurotoxin.”
I believe in the Grand Plan, and I have faith in The Director. Begone, faction scum.
That was a good show.
It could be argued that people are AGI. Are they always trustworthy?
Removed by mod
As a software developer I fully agree. People bash on it constantly here but the fact is is that it’s required for our jobs now. I just made it through a job hunt and every tech screen I did they not only insisted on me using AI, but they figured how much I was using too.
The fact is is that like it or not it does speed us up, and it is a tool in our toolbelt. You don’t have to trust it 100% or blindly accept what it does, but you do need to be able to use it. Refusing to use it is like refusing to use the designer for WinForms 20 years ago, or refusing to use an IDE at work. You’re going to be at a massive disadvantage to your competing jobseekers who are more than happy to use AI.
I review take home assignments and mostly we receive AI submissions. It’s easy to tell when they aren’t AI though because we get thoughtful comments about why one choice was made over another, and comments on the higher level view that only come from product context and experience. I don’t think one single fully ai-created submission has made it passed the code review part.
See it’s hard as an interviewer because for the first time ever I lost points at one place because I didn’t use AI at all, and they almost didn’t say yes to me. Their feedback quite literally was that it functioned well, but I could have got it done faster with AI.
Seems pointless to test you on anything that could be done by ai, otherwise why even hire someone, just have fewer devs using more ai right? I want to test people on whether they have experience to notice things and make decisions. Idk if they generate the busy work but that isn’t what I’m grading them on
Hey preaching to the choir there, but other companies were saying “if they didn’t use AI for this they won’t here either”. For your interviewees sake, make sure AI use is extremely specific for every step in the interview. I had places where they were upset I didn’t use AI at this step, but did on this other step. It’s batshit out there.
This is not a fact at all.
Fine it speeds me up.
The people in the study thought so too
That’s dumbshits using it to do their job for them and trusting the output blindly. If you’re using LLMs to get over the occasional hump they’re awesome time savers.
I’m guessing you don’t write code?