tahewett1d ago
BeyondMachines 
You gave a stranger with no soul and no skin in the game the keys to everything you own.
May 31 at 6:58amWeb
Show threadRay McCarthy1d ago

@beyondmachines1
And maybe 40% of the time it's garbage, but plausible. Without the experts you fired, you don't know when it's wrong.

So even if it was free, it would only really be a toy.

Show threadnjsg1d ago
@raymaccarthy @beyondmachines1 It's really expected - they're just adhering to the hype that Generative Autocomplete can do all things humans do. If they were at least adopting it as a *tool* and not joining that specific hype, they'd be able to see that problem right away, or at least avoid it...
Show threadVive Levant1d ago

@raymaccarthy @beyondmachines1 s/it would only really be a toy/you would only really be a tool

Edit : I misread the original so here is what I intended originally for the sentence to look like.

#thereifixedyourtoot

Show threadRay McCarthy1d ago
@Vive_Levant @beyondmachines1
I've used real, working computer tools professionally for decades and written them.
Also computer toys. I know the difference.
Show threadVive Levant1d ago
@raymaccarthy @beyondmachines1 you can check the two possibilities of the third meaning to get my point : https://www.merriam-webster.com/dictionary/tool
Definition of TOOL

Definition of 'tool' by Merriam-Webster

Show threadVive Levant1d ago
@raymaccarthy @beyondmachines1 oups, my bad, I overlooked your sentence and was certain to have read « YOU would only really be a toy. »
In this case my substitution worked, but you were right, with IT, it didn’t. Apologies.
Show threadCykonot1d ago

@beyondmachines1 yeah the security aspect has been bugging me this whole time. Like, do you REALLY want to surrender your logic to a competing tech company? Who gets to just steal / train on your hole business model? Who you incidentally just handed all your data to?

Okay mr executive. Mr risk taker. Mr get bailed out whenever you make a mistake

Show threadEmma: ƒ(☕️, 🐈‍⬛️, 🚀️) ⇒ ℂ1d ago
@beyondmachines1 the CEOs are the ones without souls or skin in the game. They always get bailed out.
Show threadVerena Rupp1d ago
@beyondmachines1 Good luck re-hiring those people - or hiring any people at all. They now know exactly what they are worth to you.
Show threadcanleaf08✅ 加拿大葉子 🏳️‍🌈⚧️🇪🇺🍁1d ago
@VerenaRupp @beyondmachines1 They won't hire people back at the rate they used to work with. They froze hiring long ago...
Show threadMatt Palmer20h ago
@VerenaRupp the addiction to food, clothing, and shelter is very strong.
Show threadKepeken1d ago
@beyondmachines1 Why are so many of these anti-ai posts so obviously written by an LLM?
Show threadpgcd1d ago

@kepeken spacing paragraphs and using short sentences is not an indicator of LLMs. It's what we used to call "effective writing" back when I was young.
Which is why it's used by LLMs, of course, since they're trained on stuff that survived and is available (so, likely, the most effective).
Also, it's been the standard communication style on LinkedIn for several years as well so any corpo would probably try to emulate it.

@beyondmachines1

Show thread⁡jaseg 🔜 GPN241d ago
@pgcd @kepeken to me that also read as LLM slop, and not just because of the short sentences.
Show threadpgcd1d ago

@jaseg I know, and I agree that it _looks_ like AI slop. But that doesn't mean it _is_ AI slop, since the whole point of LLMs is to make stuff that looks like a human could've written it.
Not all cans of Campbell's soup are a Warhol painting, I mean (although some are).

@kepeken

Show threaddatenwolf1d ago

@pgcd @jaseg @kepeken

<benefit of="doubt">
Maybe it's deliberate, to make it easier to ingest and digest by people who's brain already marinated in LLM slop so much, that they're incapable of comprehending any other style <benefit> (/s)

Show threadpgcd1d ago

@datenwolf Unironically, that could be a reason (not saying it is).
If you're talking to LinkedInsane people, it might be easier to use their way of communicating, than attempting to break through otherwise.
Cthulhu knows how poorly my posts have fared on linkedin, even when I wrote important stuff like "today I'm bored as fuck"

@jaseg @kepeken

Show threadKepeken23h ago

@pgcd

That is a great way to describe the motivation someone would have to get AI to rewrite or write a post for them.

Show threadpgcd23h ago
@kepeken unfortunately I think you're right. Having an LLM (re-) write something for me is something I find as appealing as having somebody drink my beer for me, though, so my opinions on the motivation of somebody who does that, are necessarily flawed.
Show threadKepeken23h ago
@pgcd some people are looking for self-expression (truth as you understand it, questions, unique views, ideas that someone else won't understand unless they are open to you), others for a successful post in terms of attention (AI, dog-piling, takes they know are false but will get approval from the crowd, false accusations, etc.)
Show threadDźwiedziu1d ago

@jaseg @kepeken
Folx, don't accuse other people of slop if you're basing on vibes only.

As @pgcd mentioned “AI” extrusions are designed to mimic human works.

And yes, you can fondle a prompt to do this, yet I would argue that there is less chance for an “AI”-critical person to do this.

Show threadpgcd1d ago

@dzwiedziu well said.

Also, TBH, this is one case where I agree with the message enough that I'd be sharing it even if it had been written by Elmo Skum.

@jaseg @kepeken

Show threadRockgecko16h ago
@dzwiedziu @jaseg @kepeken @pgcd The vibes in question:
Show threadRockgecko16h ago
@dzwiedziu @jaseg @kepeken @pgcd tho I have now found Pangram finds other "LinkedInsane" type posts AI, so can't be sure
Show threadKepeken15h ago
@rockgecko_dev @dzwiedziu @jaseg @pgcd good luck finding a control group sample of human-written "linked in" posts...
Show threadHazelnoot14h ago
@rockgecko_dev @dzwiedziu @jaseg @kepeken @pgcd you know those "AI detector" tools are a scam, right?
Show threadRockgecko14h ago
@hazelnoot @dzwiedziu @jaseg @kepeken @pgcd yes https://aus.social/@rockgecko_dev/116672378083534238
Rockgecko (@[email protected])

@[email protected] @[email protected] @[email protected] @[email protected] tho I have now found Pangram finds other "LinkedInsane" type posts AI, so can't be sure

Aus.Social
Show threadDźwiedziu13h ago

@hazelnoot
Worse, they're also “AI” \s

@rockgecko_dev @jaseg @kepeken @pgcd

Show thread💙🩷💜Brett🐡🍉🐧14h ago
CC: @[email protected] @[email protected] @[email protected] @[email protected] @[email protected]

An AI detector tool recently decided that Mary Shelley's Frankenstein was written by AI

Show threadRockgecko13h ago
@brettm @dzwiedziu @jaseg @kepeken @pgcd debunked here. All except "ZeroGPT" (including Pangram) found it human https://lindac.substack.com/i/192171264/4-pangram-fully-human-written
No, Mary Shelley's Frankenstein doesn't really test positive for AI

We need to talk about how AI detection really works

Hello, Writer!
Show thread💙🩷💜Brett🐡🍉🐧12h ago
CC: @[email protected] @[email protected] @[email protected] @[email protected] @[email protected]

No, that is not "debunked." All it shows that when the second person did it those AI agent found it to be written by human. That bears no relationship to what was found when the first person did it. Maybe the first person made the whole thing up. Maybe someone at the AI company saw the post and added some exception to stop their LLM looking useless. Maybe the LLM ingeted more data in the meantime and got better at predicting. I don't know. And you don't know either. Since neither of us know, nothing got debunked.


Show threadLicho22h ago

@jaseg these days even people adopt the "AI style" because they read so much of it. It's a brain disease

@pgcd @kepeken

Show threadSiderea, Sibylla Bostoniensis21h ago

@jaseg If that reads to you as AI slop, at this point you should probably seriously entertain the possibility that you actually don't know what AI slop is like and are bad at telling what is and isn't AI slop.

@pgcd @kepeken

Show thread⁡jaseg 🔜 GPN2421h ago
@siderea lol
Show threadFierce Luddite14h ago
@jaseg @[email protected] this, is always the greatest response to a troll. (This person is a therapist. 😬)
Show threadMikalai16h ago
@jaseg @pgcd @kepeken
Slop has no sunbtance. This does.
Show threadWesDym1d ago
@kepeken What gives you that impression? Please don't tell me that you're one of those people who confuses competent literacy with fakery.
Show threadKepeken1d ago

@wesdym It's because it's structured like this:

No X. No Y. Just Z.

And also
You didn't X. You YYYY.

Then there's the fact that "the AI stops when it runs out of tokens" is not a fitting expression. It shows a sentence written with an understanding of the context and English, but not the mechanics behind the topic.

None of these prove that a human did not write it, so I said "why are so many" instead of "why is this" directly. (Since, if you have 10 80% chances you have a lot of examples.)

Show threadWesDym1d ago

@kepeken Um, okay, sure, whatever.

It's a nice day to go for a walk. I suggest that.

Show threadWesDym1d ago
@kepeken For that matter, I'm taking 30 days away from you, in the hopes that touching grass does you some good.
Show threadMycotropic1d ago

@kepeken @beyondmachines1

Drop in, stir some shit, ghost; I know that pattern as well and the response is; #WhackATroll good bye

Show threadFightScore1d ago
@beyondmachines1 To be fair, AI has its use, right? Maybe the business model just got to be better, maybe token count might need tweaking -- maybe put weight to something like output value, rather than input/output count.
To be sure, AI should not replace humans, rather augment us.
Show threadTor Iver Wilhelmsen1d ago
@fightscore @beyondmachines1 Real AI sure, but the stochastic parrot LLMs not so much. I used to call them "machine lying" but lying requires knowledge of true and false and LLMs do not have that.
Show threadfh01d ago
@toriver @fightscore @beyondmachines1 sounds like you never experienced an LLM finding the source of a production bug based on log files, code and given context, in less than 10min where you would have needed at least half a day
Show threadJacob Ibáñez 🔒1d ago
@fh0 @toriver @fightscore @beyondmachines1 and finding that bug is worth all the fucking misery that it brings? I'm so fed up with this argument, things don't exist in a void and magnitude matters. Please, use your brains and stop embarrassing yourselves before you use that goddamn argument again. Please.
Show threaddavecb 🇨🇦1d ago
@iakobsdesamos @fh0 @toriver @fightscore @beyondmachines1
LLMs understand language. Telling me
- what the computer said is contradictory, or
- what I said is ungrammatical
is genuinely helpful. The other stuff? Not so much.
Show threadFightScore3h ago
@iakobsdesamos @fh0 @toriver @beyondmachines1 AI is just a tool, with abilities as diverse as people. It is still in infancy, but there is hope it would be better, yes? At least there are improvements year on year.
But I would like some insight why all these hate towards it. It seems to me that the debate is not whether the thing is half-full or half-empty, but borders on whether or not the thing will somehow cause a cataclysm by itself. (Hoping it wont 🙏😁)
A deep breath might calm us down. 🙂‍↕️
Show threadKid Cavaquinho 🍉🌻1d ago

@fh0

The thing is that you will spend half a day working on something else anyway so what's the point?

@toriver @fightscore @beyondmachines1

Show threadfh01d ago
@ballalloi I don't get what you wanna say.
The point is, it is not simply black & white and the results aren't all useless. LLMs can be a helpful tool. (Doesn't imply that I think blindly using and trusting LLMs for everything is a good idea, nor that I don't see all the problems that it can cause.)
Show threadKid Cavaquinho 🍉🌻1d ago

@fh0

What I want to say is: if I'm working and I have half a day to solve a problem, and with LLMs it only takes 10 minutes, can I take that half day off?

Show threadfh01d ago
@ballalloi for that to happen we need a new system. :D Of course it only results in more stress, the expectation of doing more things in parallel and getting more things done at the same time.
If workers benefit when efficiency increases, is a totally different topic. My point was that it is not the case that all LLM results are useless and that of course it can be a helpful tool.
Show threadKid Cavaquinho 🍉🌻1d ago

@fh0

Yes, you are right in that, I took it off topic a bit :)

Show threadHolgintheair10h ago

@ballalloi

That’s how automation was said to be going to work back in the 70s.

@fh0

Show threadMaggie Maybe1d ago
@ballalloi @fh0 @toriver @fightscore @beyondmachines1 or lose a half day of pay because there’s nothing to do. Is AI going to spend that token $ in the community or will it be hoarded in off shore accounts benefiting no one but our wealthy overlords?
Show threadJargoggles1d ago
@fh0 @toriver @fightscore @beyondmachines1
This is shockingly similar to how gambling addicts talk.
Show threadKg. Madee Ⅱ.1d ago
@fh0 profile pic checks out
Show threadMycotropic1d ago

@[email protected] @toriver @fightscore @beyondmachines1

Sounds like you never experienced two day per week watering because of a changing environment and having that available water flowing into a data center drive up your cost for those two days as well.
Regardless; defend this shit all you want but I don't feel inclined to read your comments. #KTB

Show threadThe Lack Thereof 21h ago

@fh0
In my experience at work, that happens sometimes! But it's outnumbered easily 3:1 by cases where it goes down rabbit holes based on fundamentally wrong starting assumptions.

If the LLM bug-triage bot at work says one more time "oh yeah this kernel panic issue is clearly due to a problem in this testing shell script" I STG I will scream.
@toriver @fightscore @beyondmachines1

Show threadOctorine19h ago
@lackthereof @fh0 @toriver @fightscore @beyondmachines1 Yeah. Whenever I ask AI to review anything it finds all kinds of issues, 90% of which are completely made up.