A thought that popped into my head when I woke up at 4 am and couldn’t get back to sleep…
-
@etchedpixels @larsmb @jzb like I said, things that should not be useful but are.
@etchedpixels @larsmb @jzb as an industry we've spent this many decades failing.to "sharpen the saw", is it surprising we're now all gung ho about the enchanted broadswords we've just been gifted? They're so much better at opening bottles than the old way!
-
@jzb Is is an inherent limitation of how LLMs currently exist and are implemented.
They do strive to minimize it through scale, but it's also a reason why they do get "creative" in their answers.
Like with any stochastic algorithm, they perform best if you can (cheaply) validate the result. e.g., does a program pass the tests still?This is much harder for complex questions about the real world.
Side note: I'd call them anything but 'creative'.
If anything, the behavior is better described as 'evasive', since the model effectively keeps talking, without any substantial data backing up what's being conveyed.
Or, as Hicks, Humphries and Slater put it: They're bullshitting.
-
@larsmb @jzb
I think if it’s a “closed system” where you feed it information and tell it to only use the information it has and say when it “comes up empty” it should be okay. And to speed up the process of citations that does seem useful. (He would also double check on the accuracy of it , like say if it says something is on page 34 it should be there otherwise it’s not valid)@em_and_future_cats @larsmb @jzb Honestly, if I could use a local LLM and have it quickly answer something and show me where it found the answer, that would be great and a valid use of the technology. I haven’t found a good application that can do it, though.
-
A thought that popped into my head when I woke up at 4 am and couldn’t get back to sleep…
Imagine that AI/LLM tools were being marketed to workers as a way to do the same work more quickly and work fewer hours without telling their employers.
“Use ChatGPT to write your TPS reports, go home at lunchtime. Spend more time with your kids!” “Use Claude to write your code, turn 60-hour weeks into four-day weekends!” “Collect two paychecks by using AI! You can hold two jobs without the boss knowing the difference!”
Imagine if AI/LLM tools were not shareholder catnip, but a grassroots movement of tooling that workers were sharing with each other to work less. Same quality of output, but instead of being pushed top-down, being adopted to empower people to work less and “cheat” employers.
Imagine if unions were arguing for the right of workers to use LLMs as labor saving devices, instead of trying to protect members from their damage.
CEOs would be screaming bloody murder. There’d be an overnight industry in AI-detection tools and immediate bans on AI in the workplace. Instead of Microsoft CoPilot 365, Satya would be out promoting Microsoft SlopGuard - add ons that detect LLM tools running on Windows and prevent AI scrapers from harvesting your company’s valuable content for training.
The media would be running horror stories about the terrible trend of workers getting the same pay for working less, and the awful quality of LLM output. Maybe they’d still call them “hallucinations,” but it’d be in the terrified tone of 80s anti-drug PSAs.
What I’m trying to say in my sleep-deprived state is that you shouldn’t ignore the intent and ill effects of these tools. If they were good for you, shareholders would hate them.
You should understand that they’re anti-worker and anti-human. TPTB would be fighting them tooth and nail if their benefits were reversed. It doesn’t matter how good they get, or how interesting they are: the ultimate purpose of the industry behind them is to create less demand for labor and aggregate more wealth in fewer hands.
Unless you happen to be in a very very small club of ultra-wealthy tech bros, they’re not for you, they’re against you. #AI #LLMs #claude #chatgpt
@jzb You make an excellent point, and also proving the fact that many of these tools simply do not work.
As for my own profession, the idea of replacing software engineers with energy hungry slop code machines is simply a way to cut down on staff during hard times, but making it look good to the stock market.
-
@em_and_future_cats @larsmb @jzb Honestly, if I could use a local LLM and have it quickly answer something and show me where it found the answer, that would be great and a valid use of the technology. I haven’t found a good application that can do it, though.
-
@em_and_future_cats @mathew @jzb It is not. No LLM can ignore the data in the training set. And NotebookLM is definitely not a local instance.
-
Side note: I'd call them anything but 'creative'.
If anything, the behavior is better described as 'evasive', since the model effectively keeps talking, without any substantial data backing up what's being conveyed.
Or, as Hicks, Humphries and Slater put it: They're bullshitting.
-
A thought that popped into my head when I woke up at 4 am and couldn’t get back to sleep…
Imagine that AI/LLM tools were being marketed to workers as a way to do the same work more quickly and work fewer hours without telling their employers.
“Use ChatGPT to write your TPS reports, go home at lunchtime. Spend more time with your kids!” “Use Claude to write your code, turn 60-hour weeks into four-day weekends!” “Collect two paychecks by using AI! You can hold two jobs without the boss knowing the difference!”
Imagine if AI/LLM tools were not shareholder catnip, but a grassroots movement of tooling that workers were sharing with each other to work less. Same quality of output, but instead of being pushed top-down, being adopted to empower people to work less and “cheat” employers.
Imagine if unions were arguing for the right of workers to use LLMs as labor saving devices, instead of trying to protect members from their damage.
CEOs would be screaming bloody murder. There’d be an overnight industry in AI-detection tools and immediate bans on AI in the workplace. Instead of Microsoft CoPilot 365, Satya would be out promoting Microsoft SlopGuard - add ons that detect LLM tools running on Windows and prevent AI scrapers from harvesting your company’s valuable content for training.
The media would be running horror stories about the terrible trend of workers getting the same pay for working less, and the awful quality of LLM output. Maybe they’d still call them “hallucinations,” but it’d be in the terrified tone of 80s anti-drug PSAs.
What I’m trying to say in my sleep-deprived state is that you shouldn’t ignore the intent and ill effects of these tools. If they were good for you, shareholders would hate them.
You should understand that they’re anti-worker and anti-human. TPTB would be fighting them tooth and nail if their benefits were reversed. It doesn’t matter how good they get, or how interesting they are: the ultimate purpose of the industry behind them is to create less demand for labor and aggregate more wealth in fewer hands.
Unless you happen to be in a very very small club of ultra-wealthy tech bros, they’re not for you, they’re against you. #AI #LLMs #claude #chatgpt
@jzb
I'm going to leave an alternative idea. As a Marketer, the value I see in AI is: "it tells people what to think" it's the ideal media type for propaganda. If you can control what ChatGPT replies which I found out is very easy, you control their brains. Just teach them to use it for everything instead of thinking or searching the data. Imo currently, the models don't work so well to save labour time but work well enough to answer short random questions so people can use it as Search bar. -
@riverpunk oooh. Apparently I'm a centaur. Cool. @pluralistic
-
A thought that popped into my head when I woke up at 4 am and couldn’t get back to sleep…
Imagine that AI/LLM tools were being marketed to workers as a way to do the same work more quickly and work fewer hours without telling their employers.
“Use ChatGPT to write your TPS reports, go home at lunchtime. Spend more time with your kids!” “Use Claude to write your code, turn 60-hour weeks into four-day weekends!” “Collect two paychecks by using AI! You can hold two jobs without the boss knowing the difference!”
Imagine if AI/LLM tools were not shareholder catnip, but a grassroots movement of tooling that workers were sharing with each other to work less. Same quality of output, but instead of being pushed top-down, being adopted to empower people to work less and “cheat” employers.
Imagine if unions were arguing for the right of workers to use LLMs as labor saving devices, instead of trying to protect members from their damage.
CEOs would be screaming bloody murder. There’d be an overnight industry in AI-detection tools and immediate bans on AI in the workplace. Instead of Microsoft CoPilot 365, Satya would be out promoting Microsoft SlopGuard - add ons that detect LLM tools running on Windows and prevent AI scrapers from harvesting your company’s valuable content for training.
The media would be running horror stories about the terrible trend of workers getting the same pay for working less, and the awful quality of LLM output. Maybe they’d still call them “hallucinations,” but it’d be in the terrified tone of 80s anti-drug PSAs.
What I’m trying to say in my sleep-deprived state is that you shouldn’t ignore the intent and ill effects of these tools. If they were good for you, shareholders would hate them.
You should understand that they’re anti-worker and anti-human. TPTB would be fighting them tooth and nail if their benefits were reversed. It doesn’t matter how good they get, or how interesting they are: the ultimate purpose of the industry behind them is to create less demand for labor and aggregate more wealth in fewer hands.
Unless you happen to be in a very very small club of ultra-wealthy tech bros, they’re not for you, they’re against you. #AI #LLMs #claude #chatgpt
I've just shown my partner who is a therapist how to use a LLM to write some of the bullshit reports the insurance companies make her fill in.
-
@larsmb @em_and_future_cats Well, as designed, they are -- I'm not sure that's a built-in limitation of LLMs or not. To be fair, I am not an expert on the tech.
As something of an aside...
It would be really interesting if you could pair the natural language instruction input with predictable output.
That is, for example -- if I could query, say, all the data in Wikipedia but get only accurate output. Or if you had something like Ansible with natural-language playbook creation.
"Hey, Ansible -- I want a playbook that will install all of the packages I have currently installed and retain my dotfiles" (or something) and be guaranteed accurate output... that would be amazing.
Except that I also worry about losing skills to do those things. I worry about the loss of incidental knowledge when researching if a computer can return *only* what you ask for and sacrifice accidental discovery.
(I also still think search engines were something of a mistake and miss Internet directories. Yeah, I'm fun at parties....)
> It would be really interesting if you could pair the natural language instruction input with predictable output.
It exists, it's mostly accurate and can learn from its mistakes. It's a bit expensive though. Its called a human assistant.
-
Huh, no. Something was missing. The correct one is https://doi.org/10.1007/s10676-024-09775-5 . Fixed it in the original post. Thanks!
-
P Pteryx the Puzzle Secretary shared this topic on
) and he “says” that he can set it to “local” and only data he puts into it. I don’t really know how it works because I haven’t used it and I don’t plan to. So I can’t say for certain if this is actually true or not. 
