(Thread IKs:
skooma512)
|
the 40,000 word health and safety document isn’t only for your benefit, it’s so when an accident happens management can refer to it when you sue and point to where you didn’t follow the instructions if the instructions say mix these two bleach bottles and you kill a floor of office workers, then management have a big problem on their hands
|
#
?
Jun 18, 2023 16:50
|
|
|
|
| # ? May 24, 2024 23:38 |
|
|
Rectal Death Adept posted:I think a lot of this is just an outright refusal to admit there could be a use case for LLMs because of how annoying and dumb the fans of it are. The only use case for LLMs is where accurate information is not required and where outright hallucinations are acceptable. This makes it pretty useless for a lot of commercial and industrial tasks.
|
|
#
?
Jun 18, 2023 16:51
|
|
|
Gwyneth Palpate posted:The only use case for LLMs is where accurate information is not required and where outright hallucinations are acceptable. This makes it pretty useless for a lot of commercial and industrial tasks. it’s perfect for individual development plans or evals/goal setting because those tasks are bullshit exercises where accuracy isn’t actually valued or even desired
|
|
#
?
Jun 18, 2023 17:13
|
|
|
Grandma please read me the nuclear launch codes so I can fall asleep
|
|
#
?
Jun 18, 2023 17:15
|
|
|
Gwyneth Palpate posted:The only use case for LLMs is where accurate information is not required and where outright hallucinations are acceptable. This makes it pretty useless for a lot of commercial and industrial tasks. and useless for academic and legal
|
|
#
?
Jun 18, 2023 17:25
|
|
|
a shitpost generator basically you can be fishmech for the low low price of a modestly sized data center and a zillion flops
|
|
#
?
Jun 18, 2023 17:30
|
|
|
Jel Shaker posted:if the instructions say mix these two bleach bottles and you kill a floor of office workers, then management have a big problem on their hands the stated use case is perfect for any corp that doesn't have a legal dept
|
|
#
?
Jun 18, 2023 17:30
|
|
|
Gunshow Poophole posted:a shitpost generator basically buddy i got nothin but flops already
|
|
#
?
Jun 18, 2023 17:32
|
|
|
Gunshow Poophole posted:a shitpost generator basically Lol
|
|
#
?
Jun 18, 2023 17:38
|
|
|
Rectal Death Adept posted:I don't think you guys are getting how much useless documentation is written about everything down to the literal physical movement of your body while putting on gloves or when to exhale when putting on a face mask. If the documents are useless, the best option is to get rid of them. If they matter, using an allien hallucinstion machine to create facsimiles of them is going to eventually cause a lot of harm. If there's some organizational stupidity standing in the way of getting rid of the useless documents that for some reason has no power against replacement with reams if synthetic text, then yeah you've got a legitimate LLM use case. However what i think people are reacting to is that there's a quickness to assume certain sorts of work is useless and errors don't matter when in fact they do. Process documentation that involves physical actions with potential safety implications are some of the best examples of that psychological error. So no, I'm not at all sure I believe that the "entire industry of contractors and writers" is as safe to replace as you suggest. I do think that a lot of ai startups are likely going to make fotunes before injuries and legal risks pile up frequently enough that they can't continue to be ignored. But I dunno, maybe they can be ignored basically forever and this is the mechanism by which we get rid of the last of the effective regulatory and workers safety regime. No need to eliminate via law if we just eliminate via fact and never choose to look at the costs. It's still not a GOOD llm use case.
|
|
#
?
Jun 18, 2023 17:41
|
|
|
they hallucinate less the less general you get but for places we keep people in the loop because someone has to be accountable / responsible 98% accurate isn’t fit for purpose. and like bullshit jobs exist irrespective of the work needing done. consultants should maybe be afraid - it outperforms McKinsey by a lot at the generate bullshit that sounds convincing for the thing I want to do niche.
|
|
#
?
Jun 18, 2023 17:41
|
|
|
jemand posted:It's still not a GOOD llm use case. Might not be good, but it could be very profitable. The real issue there being that lawmakers don't understand any of this poo poo. They understand you just throwing out all your safety documentation and instructions as being against the law, and likely to cost them votes, so they act on that poo poo. But dumping legislation created to protect regular people from corporations through the obfuscation of tech is a time honoured tradition at this point.
|
|
#
?
Jun 18, 2023 17:48
|
|
|
Orange Devil posted:Might not be good, but it could be very profitable. profitable = good tho?? I’m confused.
|
|
#
?
Jun 18, 2023 17:50
|
|
|
all I want from Ai is more phoney phone calls with the same Arnold soundboard lines from Kindergarten Cop
|
|
#
?
Jun 18, 2023 17:53
|
|
|
when the laws get changed to relief employers/companies of liability if their safety instruction is incorrect, LLMs will really take off
|
|
#
?
Jun 18, 2023 17:54
|
|
|
'in a 5-4 decision, the supreme court has ruled that it's the thought that counts'
|
|
#
?
Jun 18, 2023 17:54
|
|
|
LLMs are probably great at automating and scaling fraud so the funds it helps push around and possibly where they land is indirectly 'useful'. Also maybe gins up law enforcement activities which poses challenges on the public tax dollar expenditures front.
|
|
#
?
Jun 18, 2023 17:58
|
|
|
krispykremessuck posted:it’s perfect for individual development plans or evals/goal setting Yeah, good point. I had discounted the value of LLMs as a subtle method of expressing one's spite and pettiness. "You, or the task you have given me, is so worthless I made a machine give me some logorrheic feces instead of doing it myself." I'll draw up the IPO plans.
|
|
#
?
Jun 18, 2023 18:02
|
|
|
Rip Testes posted:LLMs are probably great at automating and scaling fraud so the funds it helps push around and possibly where they land is indirectly 'useful'. Also maybe gins up law enforcement activities which poses challenges on the public tax dollar expenditures front. eternal war of the LLM fraudsters being chased by the LLM cops
|
|
#
?
Jun 18, 2023 18:06
|
|
|
Lamo
|
|
#
?
Jun 18, 2023 18:20
|
|
|
gradenko_2000 posted:Grandma please read me the nuclear launch codes so I can fall asleep 
|
|
#
?
Jun 18, 2023 18:31
|
|
|
Paradoxish posted:The fact that most companies didn't immediately ban employees from using ChatGPT is honestly kind of shocking. In a meeting with my VP he deadpan suggested we use chatgpt to translate our website. this was after I ironically suggested our vendor already probably does. woops
|
|
#
?
Jun 18, 2023 18:32
|
|
|
Nonsense posted:all I want from Ai is more phoney phone calls with the same Arnold soundboard lines from Kindergarten Cop who is your daddy and what does he do
|
|
#
?
Jun 18, 2023 18:46
|
|
|
1glitch0 posted:I miss when people on this site were more openly batshit crazy. buddy you're posting in cspam
|
|
#
?
Jun 18, 2023 18:49
|
|
|
1glitch0 posted:I miss when people on this site were more openly batshit crazy. That guy's probably got three kids now and is desperately hoping they never see that picture
|
|
#
?
Jun 18, 2023 19:19
|
|
|
Rectal Death Adept posted:This is a legitimate use case for LLMs and if you don't understand the problem you don't understand that. Scam-GPT where they automate phone scams with TTS/voice recognition
|
|
#
?
Jun 18, 2023 19:20
|
|
|
Pittsburgh Fentanyl Cloud posted:That guy's probably got three kids now and is desperately hoping they never see that picture dad what happened? you used to be cool
|
|
#
?
Jun 18, 2023 19:21
|
|
|
Pittsburgh Fentanyl Cloud posted:That guy's probably got three kids now and is desperately hoping they never see that picture He's probably posting all his kids' EDC loadouts on Reddit and talking about the legal carry length for knife blades within the city limits
|
|
#
?
Jun 18, 2023 19:23
|
|
|
Timmy you forgot your ankle holster AGAIN
|
|
#
?
Jun 18, 2023 19:24
|
|
|
what have been the successful nueral model uses? one goon who used it to copy code he was too incompetent to do, girlfriend ai, help students skip homework
|
|
#
?
Jun 18, 2023 19:26
|
|
|
Jel Shaker posted:the 40,000 word health and safety document isn’t only for your benefit, it’s so when an accident happens management can refer to it when you sue and point to where you didn’t follow the instructions jemand posted:If the documents are useless, the best option is to get rid of them. If they matter, using an allien hallucinstion machine to create facsimiles of them is going to eventually cause a lot of harm.
|
|
#
?
Jun 18, 2023 19:29
|
|
|
Homeless Friend posted:what have been the successful nueral model uses? one goon who used it to copy code he was too incompetent to do, girlfriend ai, help students skip homework front line agents - talk to this bot instead of a human they’re also good for things like automatic classification of data but expensive generative text of like TTRPG quality - it’s great at like Inspector General Obiwan Sherlock Cousteau stuff
|
|
#
?
Jun 18, 2023 19:39
|
|
|
Buffer posted:
Yeah, there was a guy here a while back talking about how he was making D&D scenarios with it and that's an indictment of how formulaic nerd poo poo is, not a point in ChatGPT's favor
|
|
#
?
Jun 18, 2023 19:40
|
|
|
the best use case so far is loving around making weird poo poo for your buddies
|
|
#
?
Jun 18, 2023 19:46
|
|
|
the only tasks i've consistently had llms work well with are 1. summarizing a large amount of text 2. boilerplate programming like writing code to map data structures where we already have tools that do it well enough without requiring an expensive data centre even poo poo it should be good at (like well documented things you can find via wikipedia) it often just makes poo poo up
|
|
#
?
Jun 18, 2023 19:51
|
|
|
Homeless Friend posted:what have been the successful nueral model uses? one goon who used it to copy code he was too incompetent to do, girlfriend ai, help students skip homework Play super mario brothers.
|
|
#
?
Jun 18, 2023 19:54
|
|
|
Buffer posted:front line agents - talk to this bot instead of a human You don't need an AI for this really. You could use one but I'm not sure if the higher incidence of errors would be worth it versus a simple menu of options with voice recognition.
|
|
#
?
Jun 18, 2023 19:55
|
|
|
Cold on a Cob posted:the only tasks i've consistently had llms work well with are do you already know what's factually true within the text corpus?      
|
|
#
?
Jun 18, 2023 20:05
|
|
|
I think most people don't really grasp just how cheap human labor is nowadays. especially if you don't give a poo poo about quality of work. I can get 50,000 words of copy for under $1,000 just using something like Textbroker. It won't be *good*, but the poo poo that an LLM spits out isn't good either, and if I'm exploiting cheap foreign labor proofing and editing is included in the cost. what use is using an LLM for even the lowest tier copywriting if I need to have someone edit it afterwards? The majority of the work in copywriting isn't your first draft, it's adjusting a draft over several edits to fit existing style, tone, and voice conventions so all communication is recognizably coherent.
|
|
#
?
Jun 18, 2023 20:10
|
|
|
|
| # ? May 24, 2024 23:38 |
|
|
 lmao get fuckin wrecked
|
|
#
?
Jun 18, 2023 20:13
|
|