(Thread IKs:
skooma512)
|
most people at work anymore just see a task to be disposed of, with a ‘good enough’ result being, well, good enough. they’re paid too little, have too much to do, and yeah who cares enter something like chatgpt, which excels at producing a ‘good enough’ result that lets a task be marked as complete. until someone else has to deal with it. who will probably treat the problem similarly
|
#
?
Jun 3, 2023 18:01
|
|
|
|
| # ? Jun 7, 2024 19:08 |
|
|
I feel like the idea it spits out answers that have a large chance of being wrong but looking reasonable should be the death knell. Like imagine a calculator that was only right 80% of the time but tended to give answers that at least had the right shape, it'd be worthless because you'd have to thoroughly check every output it gave either by hand or using some other calculator which would defeat the whole purpose.
|
|
#
?
Jun 3, 2023 18:02
|
|
|
Gorson posted:I'm on the board of the local animal shelter, construction is underway on a new building to replace the current crumbling one. The city just pulled a $500k pledge and left us scrambling. I have a feeling their plan (lol there is no plan) is to just eventually outsource this to companies like Pet Smart. So what would the city rather do with their money? They've got 1.5m budgeted for a new shooting range for the police, complete with a snack bar. I've highlighted the plan, op.
|
|
#
?
Jun 3, 2023 18:02
|
|
|
JAY ZERO SUM GAME posted:most people at work anymore just see a task to be disposed of, with a ‘good enough’ result being, well, good enough. they’re paid too little, have too much to do, and yeah who cares I'd like to see Chatgpt replace those petsmart workers then
|
|
#
?
Jun 3, 2023 18:03
|
|
|
JAY ZERO SUM GAME posted:most people at work anymore just see a task to be disposed of, with a ‘good enough’ result being, well, good enough. they’re paid too little, have too much to do, and yeah who cares Yeah sure but these are all nonjobs in the first place, automating positions that don't need to exist in the first place isn't really an economic win for the wealthy, that'd just be rich people taking jobs away from their nepo hires.
|
|
#
?
Jun 3, 2023 18:03
|
|
|
everyone has correctly identified work is something you are forced to do for the benefit of others so you should spend as little time and effort as possible doing it
|
|
#
?
Jun 3, 2023 18:05
|
|
|
blatman posted:i know the spirit of your post already acknowledged this but its worth reiterating that the major disadvantage is that wrong answers look a heck of a lot like right answers Yeah, totally. IMO it works best in contexts where you can quickly test and verify the correctness of the answer. And even then that's not always possible, e.g. the code it gave you may work, but might have a security vulnerability that is not immediately obvious! But if you're knowledgeable and experienced enough to recognize that, you probably didn't need ChatGPT to begin with. triple sulk posted:it struggles with context in code Version 4 was actually way better about this when it first came out (not perfect, but definitely better than 3.5), but quality has somehow degraded noticeably since then. I think there was a tweet a few pages back about that.
|
|
#
?
Jun 3, 2023 18:05
|
|
|
is an instant, wrong answer better than a slow, accurate one? yes.
|
|
#
?
Jun 3, 2023 18:06
|
|
|
RealityWarCriminal posted:everyone has correctly identified work is something you are forced to do for the benefit of others so you should spend as little time and effort as possible doing it
|
|
#
?
Jun 3, 2023 18:08
|
|
|
Slow News Day posted:Yeah, totally. IMO it works best in contexts where you can quickly test and verify the correctness of the answer. And even then that's not always possible, e.g. the code it gave you may work, but might have a security vulnerability that is not immediately obvious! But if you're knowledgeable and experienced enough to recognize that, you probably didn't need ChatGPT to begin with. This seems to be a long climb down from the initial post tbh.
|
|
#
?
Jun 3, 2023 18:09
|
|
|
Hubbert posted:the full book review article is an INCREDIBLE read Yes it definitely moved me to check out these books at some point. A general question regarding private equity's impacts: my understanding is that what these PE firms are doing is not fundamentally different from normal private sector capitalist behavior, correct? PE owned nursing homes badly mistreat residents to make more money, but that isn't new. The major difference is these PE takeovers typically involve saddling the purchased company with large debts from bank loans used to fund the takeover itself or payouts, on top of the new owners looting everything. It's these new debts that drive the worst of the cost cutting and abuses.
|
|
#
?
Jun 3, 2023 18:13
|
|
|
Slow News Day posted:ChatGPT has three big advantages IMO: These are not true. 1. Specifically with the first, even if you didn't get locked out from GPT 4 after 25 messages or whatever, it clearly doesn't not remember very many responses and will begin to drift to incoherent, disconnected nonsense in any long exchange, where conversely a person learns more, connects more, refers to more over a long conversation. It has rapidly diminishing returns. 2. Books exist, JSTOR exists and because a person has a working memory and can intuit, see above, any decently literate person is more useful in turning access to a body of knowledge into something worthwhile.
|
|
#
?
Jun 3, 2023 18:15
|
|
|
"All models are wrong, but some are useful".
|
|
#
?
Jun 3, 2023 18:16
|
|
|
Slow News Day posted:ChatGPT has three big advantages IMO: No it can't lol jesus christ are you seriously getting fooled by the transparent fakery? I thought everybody knew that there's a detector for humans indicating a wrong answer and it just rerolls the original prompt with a new RNG for the spicy autocomplete. It's why you tell it it's wrong and you get the same wrong answer worded differently. every time prompt the black box it's "new session who dis?"
|
|
#
?
Jun 3, 2023 18:17
|
|
|
Harik posted:No it can't lol jesus christ are you seriously getting fooled by the transparent fakery? I thought everybody knew that there's a detector for humans indicating a wrong answer and it just rerolls the original prompt with a new RNG for the spicy autocomplete. It seems like it would be a lot more useful on a intranet where it can fully access data and be specifically trained on it as well as users not having to constantly restart their prompts. You may have to flush it eventually but it seems like it could have more utility. It is why I wonder if the PRC may simply greatly limit consumer use of AI and keep it at the enterprise/state level.
|
|
#
?
Jun 3, 2023 18:21
|
|
|
Harik posted:No it can't lol jesus christ are you seriously getting fooled by the transparent fakery? I thought everybody knew that there's a detector for humans indicating a wrong answer and it just rerolls the original prompt with a new RNG for the spicy autocomplete. Exactly. As Ardennes was saying, it can't perform the basic techniques of history we teach to undergraduates because, forget dialectical materialism, it does not have a working memory lending it enough coherence to use the empirical method. Every prompt exists in some degree of isolation, that only grows at the conversation continues, so it is not able to test theories, use evidence, construct arguments that build to something. Imagine if someone wrote you a paper by googling line by line as they went on and had no memory of the previous paragraph or page.
|
|
#
?
Jun 3, 2023 18:21
|
|
|
I want to hear more about the adventures of the worlds first ChatGPT lawyer, how is that going?
|
|
#
?
Jun 3, 2023 18:24
|
|
|
This would be a funny troll if SND wasn't legitimately this stupid.
|
|
#
?
Jun 3, 2023 18:25
|
|
|
I tried feeding CGPT some C code and said “make this MISRA compliant.” It came back with using memcpy in a way that wasn’t MISRA compliant. I told it that the way it used memcpy wasn’t MISRA complaint. CGPT said “sorry, here’s a corrected version that’s MISRA compliant.” It was the exact same code as the original. 
|
|
#
?
Jun 3, 2023 18:27
|
|
|
What if 80% correct but sufficiently plausible looking is good enough for most jobs?
|
|
#
?
Jun 3, 2023 18:30
|
|
|
Slow News Day posted:ChatGPT has three big advantages IMO: The human factor can be negative in a lot of interactions. If I need mental healthcare I need a human that understands complex social interactions and emotions. This is one of those areas where it would be dumb to suggest automation can take over. However, If I need to have the bill for the mental healthcare explained to me multiple times I don't need the receptionist punishing me because they don't like repeating themselves and they think I'm stupid for not understanding medical codes or insurance billing. Maybe I look like their ex or they hate my shirt or they are having a bad day. I think everyone can relate to me saying there is a lot of extra bullshit when you are dealing with other people. How many times has anyone reading this needed a simple task completed or answer to a question that turned into something ridiculously difficult? If people really did just perfectly do their jobs at the peak of human accuracy and ability without bias or complaint then the LLM critics might have a point but pretending the LLM is so inaccurate to be useless and Humans work the exact same way but more accurate is just a dumb way to reinforce preconceived notions.
|
|
#
?
Jun 3, 2023 18:31
|
|
|
Slow News Day posted:Yeah, totally. IMO it works best in contexts where you can quickly test and verify the correctness of the answer. And even then that's not always possible, e.g. the code it gave you may work, but might have a security vulnerability that is not immediately obvious! But if you're knowledgeable and experienced enough to recognize that, you probably didn't need ChatGPT to begin with. can't really say for sure to be honest, but i think it's not holding the context as long or as well, maybe because they have to dedicate resources to the 8k gpt-4 apis that said, i have access to said 8k gpt-4 apis and amidst testing them out for a personal project and realized it's way too loving expensive to be worth it
|
|
#
?
Jun 3, 2023 18:32
|
|
|
i was also testing out langchain because it seemed like it had a good recursive solution to handling chunks of text too large for one query if i hadn't set a very low limit of  when testing it out, it would have charged me hundreds of dollars in minutes, because it capped at that limit instantlyit's a piece of poo poo library
|
|
#
?
Jun 3, 2023 18:33
|
|
|
Any of you laid off yet due to this bullshit?
|
|
#
?
Jun 3, 2023 18:34
|
|
|
relevant as just posted on hn https://humanloop.com/blog/openai-plans quote:OpenAI’s plans according to Sam Altman
|
|
#
?
Jun 3, 2023 18:34
|
|
|
they should make AI that's automated and intelligent because right now it takes a whole lot of humans to make it happen
|
|
#
?
Jun 3, 2023 18:35
|
|
|
triple sulk posted:can't really say for sure to be honest, but i think it's not holding the context as long or as well, maybe because they have to dedicate resources to the 8k gpt-4 apis 
|
|
#
?
Jun 3, 2023 18:37
|
|
|
Nothus posted:What if 80% correct but sufficiently plausible looking is good enough for most jobs? you do 80% of the job you are given, before handing it to the next guy. that guy does 80% of his job on the 80% that you did, and so on and so forth til the can gets too big to kick and crushes society.
|
|
#
?
Jun 3, 2023 18:39
|
|
|
triple sulk posted:relevant as just posted on hn I've also read that open-source LLM has made massive leaps over the past several months to the point where it's both significantly cheaper to set up and operate and also almost as good if not better than Open AI's stuff. Haven't looked into any of it but I hope it's true, because gently caress Sam Altman.
|
|
#
?
Jun 3, 2023 18:39
|
|
|
i had 50+ page documents i wanted to analyze and doing so cost anywhere from $1-2 per document that doesn't sound like a lot at a singular level, but when there's a scenario where hundreds or thousands of these could be analyzed, then lol, lmao
|
|
#
?
Jun 3, 2023 18:40
|
|
|
Slow News Day posted:I've also read that open-source LLM has made massive leaps over the past several months to the point where it's both significantly cheaper to set up and operate and also almost as good if not better than Open AI's stuff. Haven't looked into any of it but I hope it's true, because gently caress Sam Altman. supposedly the one called claude is pretty good but it's impossible to get access and i don't know if the pricing is really any better the open source ones might be fine but in the end you'll need several thousands of dollars of hardware to run any of the advanced models, never mind the electricity costs
|
|
#
?
Jun 3, 2023 18:41
|
|
|
an interesting article, quite old but still relevant https://lareviewofbooks.org/article/the-supermanagerial-reich/
|
|
#
?
Jun 3, 2023 18:42
|
|
|
i wouldn't be surprised if the total usage of electricity for all combined ai right now is in bitcoin territory, if not worse
|
|
#
?
Jun 3, 2023 18:42
|
|
|
Mr Hootington posted:Any of you laid off yet due to this bullshit? ill never be laid off due to this bullshit because im employed solely for my good looks and charming personality, traits the robut cannot replicate (yet)
|
|
#
?
Jun 3, 2023 18:43
|
|
|
there are over 200,000 people in the stable diffusion discord server alone hundreds of thousands of people running $2000-4000 graphics cards on 750 watts or whatever the gently caress they need these days
|
|
#
?
Jun 3, 2023 18:44
|
|
|
blatman posted:ill never be laid off due to this bullshit because im employed solely for my good looks and charming personality, traits the robut cannot replicate (yet) Lots of people are satisfied with the looks and personality the robot already replicates.
|
|
#
?
Jun 3, 2023 18:44
|
|
|
Triple sulk explaining why all computer nerds need to be *clears throat* downsized.
|
|
#
?
Jun 3, 2023 18:45
|
|
|
Going to watch small soldiers and know in my heart that if that were to happen the toys would be a smashing success before the toy's audience is killed in at home accidents.
|
|
#
?
Jun 3, 2023 18:46
|
|
|
Harik posted:You'll be completely shocked to learn that one of the most important functions of the IRS is to find out about things like this and turbo-poo poo on them with extreme prejudice. goddamn that middle part of the 20th century where everyone had been huffing leaded gas fumes for decades and the courts were churning out the case law that IRS now relies on was absolutely wild like how do you even catch this kind of stuff unless an IRS agent sat there and read both tax returns like a novel: https://en.wikipedia.org/wiki/Commissioner_v._Duberstein posted:Berman was president of Mohawk Metal Corporation. Duberstein was president of the Duberstein Iron & Metal Company. They would often talk on the phone and give each other names of potential customers. After receiving some particularly helpful information, Berman decided to give Duberstein a gift of a Cadillac. Although Duberstein said he did not need the car as he already had a Cadillac and an Oldsmobile, he eventually accepted it. Mohawk Metal Corporation later deducted the value of the car as a business expense, but Duberstein did not include the value of the Cadillac in his gross income when he filed his tax return, deeming it a gift. The Commissioner asserted a deficiency for the car's value against Duberstein. The Tax court affirmed. how many midcentury american nuclear families were stashing all their excess cash in accounts under the child's name, because the child having no wages paid a lower tax rate lmao https://www.nytimes.com/1984/02/23/business/irs-wins-gift-loan-tax-case.html posted:The decision, written by Chief Justice Warren E. Burger, supported the Internal Revenue Service, which has maintained since 1966 - in the face of several lower court rulings to the contrary - that the value of the use of interest-free money was a taxable gift. also that ice cold "appealing a victory to SCOTUS because you want to stamp out a conflicting opinion in another circuit" move
|
|
#
?
Jun 3, 2023 18:59
|
|
|
|
| # ? Jun 7, 2024 19:08 |
|
|
Twerk from Home posted:I want to hear more about the adventures of the worlds first ChatGPT lawyer, how is that going? which one, the scam company or the idiot who submitted a motion to a judge filled with hallucinated cases? the latter has a hearing to explain why they shouldn't be disbarred on the 8th.
|
|
#
?
Jun 3, 2023 18:59
|
|