Remember how we were all supposed to be "left behind" if we don't jump on the Metaverse bandwagon? Especially businesses?
Yeah, about that:
https://www.theverge.com/tech/863209/meta-has-discontinued-its-metaverse-for-work-too
But today we should treat absolutely seriously all the bullshit about "being left behind" if we don't adopt "AI"! 🤡
@rysiek Internet is just basically a 2D environment. We tried 3D many, many years ago (vrml, the 3d version of html) and nobody cared.
Yes, we read the Gibson books. No, it doesn't work as anything other than a metaphor for attacking other servers, you can't use it to (for example) book a train ticket, unless the 3d world presents a pseudo-2d screen for you to operate like a browser window.
Augmented Reality, however, shows promise. provided we don't allow commercial interests to dominate it with pop-up ads.
@rysiek I can't see the point of AI in most areas when we have our own inherited intelligence. It may not be perfect but AI would not exist without human intelligence - past, present and future. Massively overhyped.
@rysiek I don't like the world I'm forced to live in now....what makes you think I need another one to occupy? I never got that metaverse shit.
@rysiek tried my nephews vr headset (hes a child). I would never ever wear one to play a game for more than 3 minutes. They are terribly uncomfortable, hot, and counterintuitive. The games are worse than playing a SNES game via emulator on your phone. And then of course on top of it all its full of ads and social media ads. They enshitified it before it even got popular. Lmao.
... META WAS STILL WORKING ON THE METAVERSE?!?!!?
I thought they stopped that shit years ago.
@rysiek And nothing of value was lost
@rysiek Oh, I'm sorry, I'm still trying not to get left behind the blockchain bandwagon.
@rysiek@mstdn.social Yeah I wish that also happen with AI
@rysiek isn't the VR market/adoption/whatever larger today than at any point previously? Facebook failed at being an ISP, and recently failed as a game studio. Says more about FB than anything AI related
@rysiek haaah here we go again 
@rysiek It's Microsoft Bob, but in VR!
@rysiek VR Brillen machen richtig spaß, teils unglaublich wie gut die Steuerung oft ist, für manche spiele und spezielle technische Anwendungen eine tolle Errungenschaft. Die "Sport" spiele sind echt cool.
Jedoch komplett untauglich für die alltägliche Büroarbeit. Etwas über die virtuelle Tastatur eingeben? Grauenhaft.
Ich konnte sie nie länger als 2 Stunden tragen.
AND Crypto! All those businesses that went belly-up because they didn't implement crypto somehow! Remember?
@rysiek LLMs atleast have some semi-credible use cases beyond "John can't be assed to type his bullshit emails himself". The Metaverse was just a fucking insane idea that would have received zero attention if anybody but dipshit tech bro Mark Zuckerfuck hadn't been the one championing it
@engelbrektrakbladsson @rysiek Honestly though, I think one of the best potential outcomes of this whole LLM hype is that everyone learns that bullshit emails (and similar drudgery and busywork) don't need to be done.
@Nephrite @engelbrektrakbladsson I really want to have your optimism!
@rysiek @engelbrektrakbladsson I bet. It's one of the few things keeping me sane lately. :D
@Nephrite @engelbrektrakbladsson it is a superpower. Never let it wither.
@rysiek when i read this i remember "second life" and how this was the future, how companies went in full, claiming that it would be hosting online job intervieuws, and people working through second life.
So when. Everyone went samey with the metverse, i predicted this happening. I was called crazy, becUss fb was behind it...
I love when I'm right
@rysiek Fads driven by real technology are bad enough. These are marketing fads driven by corporations and are far worse. The lesson is never hitch your future to a fad.
@rysiek If "AI" is so fucking great, then why is it being forced upon us instead of everyone clamoring to use it?
@rysiek I feel like 'left behind' when it comes to tech should be interpreted in the sense of the Tortoise and the Hare fable
@rysiek rip third life
@rysiek It doesn't matter, left behind, caged like a rat, the smart people that love playing with technology will find a way to continue playing with it for "the embetterment of all." Just do what you love, and you'll never work a day in your life. Oh, if we could all be so Luckey.
@rysiek I just wish VR glasses became cheaper due to that. I own a VR game and I can't play it in VR!!!
@rysiek The original version of workspace had that frustrating almost something there feel to it. The hardware is very low cost for its capability but every app has its own interaction method and flakey user experience. Couple that with the comfort issues and it makes the whole experience an effort. We are researching VR and XR in design education and seeing if we think there is any "there" "there". I wonder is it possible to flash these devices with an OS that dosent try to sell me games?
@rysiek Ai, yes, Failbook for business.....
I had almost forgotten about this, but yes, around 10 years ago - no idea any from when to befor I was out - we also had this for the organisation I was a sysadmin for.....
Shame, I had almost banned this nuisance from my memory.....
@rysiek
Darn it, Michal, are you telling me that I missed Zuckerberg's Metaverse boat?
Now I haz a sad and am gutted.
Beside myself with grief, I tells ya.
@rysiek I remember working with Avanade or somesuch a while back when they were doing that with a few past clients of mine. "Hey, could you join our teams call and use the meta plugin?" That was rather short lived. And pretty funny, to be honest.
@rysiek Heard in the coffee break room of newspapers when #MarkZuckerberg announced the #metaverse : “Meh. That story doesn’t have legs.” 🥁
#Resist hilariously, and #DoNotCollaborateWithFascists in #USpol.
@rysiek 100% agree, that being said not everything needs to be well written in business I worked in 3-4 big corporations and the amount of admin tasks AI can do for me is the best thing that ever happened to save my time and sanity.
@cypriendlp glad you had good experience; plenty of people in similar positions have the opposite:
https://hbr.org/2025/09/ai-generated-workslop-is-destroying-productivity
@rysiek yup I can't disagree on this, probably the difference in the way people use it, I just use it on processes pre-AI instead of adding new ones. Always check if people are happy in the way I use it with all the disclaimers. So far so good.
To add in your direction I even had to use llm to scan and remove slop from my inbox.
this as a really stupid product from the get-go.
Disney had tested VR in the 90s and abandoned it because it made many people nauseous, and obviously there were safety issues with people flinging their hands and legs about
But I suppose Zuck was salivating about placing ads in 3d space instead of 2D space(facebook)
A hype…. And I have seen many in my lifetime…
@rysiek Zuckerberg is desperate for a new gimmick. Facebook is full of nothing but advertising and bots of his own making. Hard to see in your feed a real person you follow. How many real users does he even have?
Zuckerberg is still an introvert trying to act like he knows what extroverts want. He has a big problem, and he knows it, and he fears HE will be left behind. And normal people don't build compounds, no matter how rich.
@rysiek but it would have been a success if everyone would have just jumnped on the Metaver.... ah well, whom am I kidding.
Also bloody shame that it's a shit company like Meta which is making the best VR glasses (doesn't someone with morals take over Meta's VR glasses business?)
@rysiek 💸 imagine the sensible and productive things that could have been done with the dozens(?) of billions burned for that stuff
@FediThing @grob ahahaha McKinsey is such a meme! 🤣
@rysiek I'm so grateful that I inherited my father's sense for technical snake oil...
@rysiek FOMO can be one hell of a way to manipulate people, I've often read how the Japanese market often has "limited time" offers to manipulate a certain kind of consumer who feels the need to buy them all while it's still possible.
@rysiek While I agree that companies should not fall for this "adapt AI or you are left behind" propaganda, I understand that it is a lot harder to spot this as a right decision than "your business must be in metaverse!"
The fear that your competition can have a tool that drastically increases production / cut costs while you don't, seems tricky to look past, and while most companies can easily call BS on "we need to be in metaverse", few dare to blow off a "potential cost cut that our competition may use"
I have a feeling that a lot of managers are terrified for either option, "do not go for AI" and "our competitors are more cost effective" vs "go for AI, and risk throwing a lot of money on a bubble that burst"
Having dependable technical advisors is proving to be very, very important. Sad that a lot of managers have little faith in us "non-business tech geeks".
If they opt to trust consultants that clearly has a bias, then, well, no pitty.
@lettosprey yup. In one of my previous jobs I had to have a conversation with the founder/CEO type person who kept coming back from conferences with "great" ideas and trying to get them implemented ASAP even though they made little sense in the organization.
I said:
– Look, you pay me a good salary for my expertise, my expertise tells me these are bad ideas, I explain why. If you still choose to trust some rando at a conference over a tech expert you are paying good money to, that's on you.
@rysiek I was Yah, for me it went from "we value your knowledge and how well you present facts, we want you to be tech lead" to "oki, so this is what we are going to do, you think this is a good idea, right? ... no? Oki, but we need you to say it is a good idea, because you are tech lead and you have to sign off on this!"
I quit.
In fact, the entire dev team quit.
@lettosprey good move. It is so important for techies not to allow themselves to be painted like this – after all, if you signed off, it'd be on you when it inevitably led to problems.
Hope it wasn't too bad finding other employment.
@rysiek It wasn't. In fact, I am kinda happy it happened since it lead to me a job in a much more interesting dev position in a not-for-profit company in the public transport sector, so I can feel like I do something important *OvO*
@petrescatraian yyyyeeeaaaaaah! 🎶
If anyone ever tries to tell you LLMs are just as good (or better!) in generating text (or code) as humans are in creating text (or code), ask them about "dogfooding".
Dogfooding means training LLMs on their own output. It is absolutely disastrous to such models. Every "AI" company out there will have layers upon layers of defenses against LLM-generated text ending up in training data.
Which is why they desperately seek out any and all human-created text out there.
@rysiek every ML model collapses in such scenario, not only LLMs. That's why it's increasingly hard to push knowledge cutoff. Anything post 2023 is contaminated more or less.
There is even a name for it:
Habsburg AI
It also makes a nice Factuality check for LLMs. If asked about what Habsburg AI is they should mentioned the researcher who coined that phrase (Jathan Sadowski), they should mention Model Autophagy Disorder (MAD) and causes for this (cumulative errors).
@rysiek In software development, dogfooding just means using one's own product ("eating your own dog food") not just for testing, but for regular usage to discover whether it's actually useful.
What you're referring to is called training on synthetic data, and it actually works better than I thought it would, but only for some kinds of training. Model collapse IS happening, just at a bit slower rate than I initially thought. LLMs DO need novel human-written data because they can't do it themselves. There's very fundamental pieces missing in the puzzle, and I would love if AI companies stopped wasting energy driving us to possibly the worst recession we will ever see.
@starsider the term I used is used by people doing LLM training. I know what term I used, and I used it in the context of training LLMs. A term can have different meanings depending on the context.
@rysiek I agree, sorry if I sound contrarian sometimes.
@rysiek
and so this should bring up the next obvious question:
What happens to the AI investment when the AI slop has replaced the copyrighted human generated data because all the humans have been fired or are not being paid to generate new content
#ai #science #commonSense
#democracy is more than voting. Lets start asking the hard questions.
@rysiek it's also a bit interesting that another meaning of dogfooding which I thought of first is positive and means that people that develop software are using it daily themselves so they actually need it to be good
@rustynail yeah, terminology is a bit of a mess
@rysiek @rustynail It does seem similar - if they use their own product, they become dumber and the results are not even close as good as advertised…
@richlv @rysiek strictly speaking, iirc when it comes to neural networks, dogfooding is not bad because the previous results are bad quality, but inherently bad for some fundamental learning algorithm math reasons, it's kinda like trying to adjust your arguments to fit the predefined conclusion instead of properly reaching it. Although my understanding of this is pretty vague
@rustynail @richlv yup. It's a fundamental thing, and it leads to model collapse – as in, it's not just wasted effort, but it measurably, dramatically lowers the quality of output.
@rysiek subsequently they NEED to make sure AI-generated text is clearly distinguishable from human-written. Which does not correspond to the general idea
@krahabors well the way they do distinguish this for training data is complicated and requires some samples of human created text that have never been used in training of a given model.
It's not like it is entirely obvious in the generated text, nor like they put some steganographic markings in generated text for this purpose.
@rysiek With dogfooding there is a problem of “it works for me!” defence.
Fortunately it's finisher is answering with “show me” which results either is a walk-back, “it's only for me”/“WFM” (thus anecdotal), or being a wanker. Which ends the discussion.
@rysiek Regarding coding, there is a lot of documentations, examples of code etc. They often follow the same patterns (think: REST API). LLM will generate skeleton of the program faster, with execption handling etc. so programmer can focus on parts requiring creativity, not repeatable ones.
Regarding dogfooding - it's probably just a matter of stucking in local optimum/cost. No point in wasting CPU cycles in learning things that model already "knows".
@rozie Customizable code templates for a known pattern don't require scraping the whole Internet at a pace reminiscent of a DDoS, undermining copyright, ignoring lack of consent, the ecological impact of LLMs, etc.
And they can be quality-assured.
For starters.
Just because a problem *can* be solved using a sledgehammer doesn't mean that a sledgehammer is an *appropriate* tool for solving the problem.
@mkj @rysiek It's not customizable code template for known pattern. Like Lego - having bricks doesn't mean you have the castle.
How public documentation is served without consent?
Writing code by yourself, reading docs, etc. can actually use more energy than using LLM. And I'm counting only home equipment.
@rysiek @mkj 0.3Wh per search (prompt) is not that much. Had discussion about it https://agora.echelon.pl/objects/2a641eff-805a-4e5e-9ff3-d09376620620 There are sources and citations included.
> How public documentation is served without consent?
You know perfectly well that LLMs are not only trained on public documentation. They are trained on everything these companies can get their hands on, legally or not.
These companies scrape the shit out of every website that has human created text, ignoring robots.txt and any sort of indication of lack of consent.
To the point where websites need to put captchas in front of them just to fend scrapers off!
@rozie uderz w stół, a nożyce się odezwą! 
@rysiek Moja alergia na ślepą agitację przeciw LLMom dorównuje alergii agitacji ślepego pchania LLMów wszędzie. 🤷♂️
@rozie "ślepa agitacja" poparta linkiem do pejpera w Nature, ale spoko, na zdrowie. 
@rysiek Ślepa agitacja jest w części pierwszej. To, że wynik LLMów nie nadaje się na materiał treningowy dla LLMów nie znaczy, że nie kod czy tekst nie jest równie dobry (lub lepszy) niż wytworzony przez człowieka.
Hint: nie tworzymy kodu/tekstu do tego, by trenować na nim LLMy.
Let me spell this out so that even a techbro could understand:
If LLMs are supposedly as good as – or, as some claim, better than! – humans as far as text output is concerned, why is dogfooding a problem in the first place?
Why all these defenses? Why seek out, and pay serious cash money for, real human-created text?
It's because they are not. On the most basic level. Anyone claiming they are is either bullshitting you, or has no clue about what they're talking about. Or both.
@rysiek if dogfooding did work for LLMs you'd also have to wonder why LLM manufacturers are still _manu_facturers. Why not let ChatGPT improve its own code, i.e. extend dogfeeding beyond data to actual code? Seems to not work, right? (I have 0 clue about LLM dev, just wondering)
@rysiek my experience in writing original stuff with ollama llms is...
I only use them for testing originality and catching typos. They cannot generate what they have not seen.
And good original writing it cannot do. It 'borrows/steals' from what it was trained on, which is all convergent. We divergents laugh at it.
@rysiek 🎵 model collapse 🎵
🎵 i just - can't - wait! 🎵
@rysiek trying to make VR headsets happen, they're never going to happen.
