Ben Aveling
Thomas Fuchsso if AI is so amazing, why don’t articles and books written with it have huge “proudly made with AI” banners and stickers on it
we all know why
Hylke Bons 🥜@thomasfuchs so glad the EU is making companies mark it as such.
Reformed Content Marketer@hbons @thomasfuchs It's a step forward, but it's not good enough. When I searched for information about that law, one of the first results was a marketing firm called Weventure talking about how they can "make sure your AI-powered content doesn't have to come with a warning label." I'm not linking to them because I don't want to boost their SEO. Basically, they're going to do just enough human review to get around the law.
If my past experiences in marketing are any indicator, people aren't going to check very closely for AI hallucinations before publishing articles that won't be marked as AI. Taking the time to do proper fact checking is obnoxious and not "efficient" enough.
@hbons @thomasfuchs still happy to see that the EU is trying to do something about unmarked AI content, though! Hopefully non-EU countries will follow suit, and people will keeping working on ways to deal with all the AI slop/hallucinations floating around the internet.
kkarhan@reformed_content_marketer @hbons @thomasfuchs #AIslop needs to be outlawed like #Asbestos & #Tetraethyllead!
kennethbspringer.au@thomasfuchs given the rate of "improvements" with AI, it is likely that the quality would improve in time to come. Be interesting to see if the content surpasses that on a human and said banners would proudly be displayed.
On the flip side, with media generation does have some marker added to the media content. While it doesn't say "proudly made with AI" but the intent to let you know it was AI generated was there.
These are done by visible Markers: Logos, icons, or "Imagined with AI" text placed directly on images, easily cropped or edited out.
Invisible Watermarks: Embedded digital signatures (e.g., Google’s SynthID) that remain detectable even after edits.
Metadata Tags: Technical data like C2PA or IPTC embedded in image files, describing how the image was created; these can be stripped easily.
This is happening now with Google (Gemini/Imagen): SynthID
OpenAI (DALL-E 3): Content credentials in metadata
Meta (Facebook/Instagram): Visible tags, invisible watermarks, C2PA metadata
Stable Diffusion (via Meta/Inria): Open-source invisible watermarking (Stable Signature)
Maybe not long to go until articles and books will follow suite? Your thoughts?
LR@kennethspringer @thomasfuchs the rate of AI improvements is subject to the law of diminishing returns. don't hold your breath for robo stephen king just yet.
@lritter — no Stephen King just yet, but perhaps a matter of a when?
Reminds me perhaps like early robotics, which were often pretty clunky, awkward, and even a bit funny—more prototypes than practical tools—AI has come a long way from those rough beginnings. These days, it's genuinely outstanding and extraordinary, but there's still a fair way to go.
I reckon the real push will come from where the big bucks are being invested and the potential to cash in on it. When there's serious money to be made, development speeds up and takes us even closer to real, game-changing AI.
@kennethspringer i give it another 50 years. money isn't what makes the difference here. that's what the big shots don't get. we're stuck again in a local optimum.
Polly Kraisus@kennethspringer @thomasfuchs
"Be interesting to see if the content surpasses that on a human"
How do you measure that AI-generated material "surpasses" it's meatspace counterpart? When it has all been trained on things created by humans? When does it go beyond jack-of-all-trades mimicry? Honest question.
And by using "imagined with AI", do you mean to say you see in AI tools and their associated LLMs the ability to imagine things just like humans would?
@stragu these are good questions and having metrics to validate the claim that "things are better" will be a bit of a challenge.
If AI learnings are based on things created by humans, and through all that big data, it can determine what materials makes a general success, including novelty. It's almost like AI can be predictive in what's the next step to increase the chances that it will be a winner.
On the flip side, if I may add my personal viewpoint/experience with AI is that AI seems to be amplifying abilities of those that use it.
Using the earlier comment reference to "no Stephen King yet". I truly believe it's a matter of when. BUT... Yes there is a but... imagine if Stephen King wielded AI to amplify his abilities? That would push Stephen's work to the next level and beyond... AI will need to continue to learn from great writers...
Flock of Nazguls@thomasfuchs It's had about 4 years to prove itself now and no earth shattering advances have come of it. So much for the singularity.
isol"We" is currently people who grew up before AI, and saw what humans can create.
If AI hangs on long enough, and gets pervasive enough, a generation will come along with a very different baseline, and they will think AI is amazing. 😢
Ian MalcolmAnd yet it's the younger generations that are the most vocal and widespread in rejecting it, so I don't think this is necessarily true or inevitable.
Human creativity isn't going away, no matter how much the slop merchants might insist that it is.
Noodlemaz@isol @thomasfuchs why would they think it amazing if it's always been there?
Albi 
🇵🇱 
@thomasfuchs I don't think you realised it yet, but they are. I saw a book in a regular old super market with a ribbon "Made with ChatGPT" in a bottom corner. I saw Ice cream "designed by AI" with ai-made art. I saw so much AI merchandise it's incredible how tone-deaf regular people are. Just 2 days ago I took down AI-made posters for a event in a club with real fucking musicians. It still had the Gemini star on it.
ROTOPE~1 ⭐️@Albi is that the official ice cream flavour of Battlefield Earth?
@rotopenguin IDK, i don't want to know how shit tastes based on how it smells.
bachimusprime@Albi @rotopenguin it tastes very sugary uwu
@Albi @rotopenguin it's very yummy!! and yes, i've eaten my own shit :3
billy joe bowers 🗽I had downloaded an app that I saw on Mastodon, a security reporting app. It wasn't bad. When I looked at their home page it said it was completely made with AI.
I deleted it.
bigiain@billyjoebowers Did that happen to be an app named after a jewellers assistive vision device? (Ugh, I just wen't to look, and found an Agents.md file (that didn't say not to use AI)) I guess I'll need to delete that too. @thomasfuchs
[Edit: typoed assistive ]
Keith Hoodlet 
@thomasfuchs it’s like knowing you’re looking at a forgery versus the real thing. The way we perceive its value and beauty changes on that one small detail.
Bo Tally@thomasfuchs I just put together a really cool picture with AI. The same day that AI canceled all of my meds at CVS
datatofu@thomasfuchs I am going to assume this is an extension of this:
https://hachyderm.io/@thomasfuchs/116727389440175494
And respond as if it is just a singular cohesive thought spread over multiple threads. Is that cool?
So the premise that the set of objects called Developers and the set of objects called Engineers are equal sets is false. Pretend the timeline does not exist. Which bootcamp teaches web development _and_ system engineering before handing a completion certificate? That said, your frustration is valid.
Thomas Fuchs (@[email protected])
And even for the most optimal of use cases, code generation, which gives an LLM every possible advantage (huge amounts of self-similar training data with millions of iterations of the same problems and solutions, strictly deterministic symbol sequences in highly simplified languages), LLMs are starting to approach local maxima; they’re even regressing when context windows get too large (so more capable hardware doesn’t help). I still strongly believe that they’re useless for 99% of tasks they’re marketed for and for the rest (like coding) there will be locally run solutions that are affordable. ¯\_(ツ)_/¯
Maddie 🎉 1️⃣7️⃣ 
@thomasfuchs to be fair, when word processing became mainstream, there weren't books that proudly said "made with a macintosh" on them, except books written on a macintosh weren't immensely worse because they were written with a keyboard
edit: this post isn't meant to be taken too seriously, i'm very much against AI generated books
edit: this post isn't meant to be taken too seriously, i'm very much against AI generated books
Unionized Netizen@mjdxp @thomasfuchs point taken but also “Sent from my IPhone”
@mjdxp generating isn’t writing
tusharhero@mjdxp @thomasfuchs this is a qualitative difference. Let's not pretend LLM generated books and Books written in a word processor are the sane thing.
Riley S. Faelan@mjdxp Typewriters also had keyboards.
氷@thomasfuchs I'm truly terrified that AI will take over the world in the future.
JL Johnson 
@thomasfuchs if I made a thing with AI, I wouldn’t be proud of not being able to do it myself.
Rene Wooller@thomasfuchs the largest LLM has something like a trillion parameters. A little more than the number of synapses of a mouse. So I guess you can think of an LLM as an extremely learned mouse that is trained to appease
CohenTheBlue@renetron @thomasfuchs A mouse will truly learn *without 3rd party input* to respond to situations, like if the cheese in the middle of the dungeon that their mouse friend ate kills the friend, they'll avoid it.
LLM is not intelligence, it is a statistical generator with a huge amount of exceptions (not like synapses). I don't think the mouse and an LLM equate in the least.
@cohentheblue @thomasfuchs cohen what do you mean by “huge number of exceptions (not like synapses)”?
@renetron @thomasfuchs 1/2 As I understand it, an exception (parameter) is a rule describing how to act in case a certain situation comes up.
I was referring to the use of the word synapse by you, @renetron
"... trillion parameters. A little more than the number of synapses of a mouse."
A synapse is something both much more specific than an exception (parameter), flexible but also more general. I don't know enough to say if using "synapse" in this context is correct.
@renetron @thomasfuchs 2/2
What I was trying to get at is that LLM parameters are not an indication of intelligence / sound learning decisions.
Not directly relevant, just trying to figure out synapses and intelligence:
This paper
https://pmc.ncbi.nlm.nih.gov/articles/PMC4685590/
states that equating the number of synapses with intelligence is debated. EDIT: clarified, added original publisher link to study
@cohentheblue @renetron generally you cannot compare brains with computers because they work fundamentally differently (the most fundamental one is that a brain doesn’t run software but is a physical object subject to, well, physics; and doesn’t “compute” things, there’s no conversion of symbols happening)
@cohentheblue @thomasfuchs re parameters and rules and exceptions - they are all different things. a real neuron has connections with other neurons that pass electrical signals. When the combined incoming signals crosses a threshold, that neuron itself is activated and starts firing signals out to other neurons. Some connections are easier for the signal to propagate (across the synapse), others are harder. In an artificial neuron, the ease with which a connection passes a signal is modeled in a ‘parameter’ . A rule would be learnt by a whole complex graph of neurons - ie given this input, produce this output. I’m not sure ‘exception’ belongs in any of that description …
Anyway my original comment is just an attempt to get some useful analogy way to think about the new technology. Analogies never really work, and open to other ones. something that has the properties of not being generalisable or adaptive but able to retain and scan huge amounts of data or deal with a very wide context and pick out patterns easily
@renetron @thomasfuchs "given this input, produce this output" is what I described as an exception. Why would this not fit? I don't mean the word in a programming language context, heavy with other technical implications like it's an object that needs further handling and stuff like that...
Basically a rule. Intelligent beings don't have just the rules, they can also create the rules on their own, by example of other situations and comparison, reflection etc, thus learning.
@renetron This kind of comparison is problematic, actually. Look at how much thinking jumping spiders can pack into their tiny cute brains, on the order of just hundreds of thousands neurons. Mammalian brains might be ridiculously inefficient and compensating by oversizing.
There's probably an evolutionary advantage to this inefficiency, but this might not apply to synthetic neural networks.
@riley @thomasfuchs Riley it’s true. ANNs are a gross simplification. Apparently Real neurons each have a unique signature in their spiking pattern which can identify them. I don’t think ANNs are anywhere near that level of information.
Thomas the weights in an ANN are intended to model the strength of a synaptic connection. So why not compare the model with the thing it’s modelling at least in some knowingly imprecise but potentially useful way
@renetron LLMs don’t work like brains, so no, you can’t
@thomasfuchs I've seen some "look, my AI" or "Claude wrote this!" It looks like it typically takes a true believer to happen upon that content, crypto or AI youtuber circles for example.
Blurp Faderpsen@thomasfuchs because llm's are also humble?
labbatt50Also, if ai is so good it should be able to support itself and not be a drain on taxpayers.
That's my opinion anyway.
Mousa Al Bateh | Cloud Security & Architect@thomasfuchs The fundamental problem is that our world was constructed based on the concepts of copyright and privacy.
AI developers are treating it us "What's mine is mine, what's yours is ours".
Jack's Countless Lack of Cool@thomasfuchs Surely next year's Hugo and Nebula awards, and by the end of the decade the Pulitzer and Mann Booker committees, will be rolling out the "written by nonhumans" categories.
Jenica Lake เหล็ถเลถ@thomasfuchs weights lifted with steroid support, twice the density, twice the rage, half ass your workouts but get real results!
@MamaLake Hey, there's nothing wrong with cis men getting gender-affirming hormone therapy. Their dicks might take offence at being thought of as incompetent, but that's between them and the dicks.
proedie@riley @MamaLake @thomasfuchs I can haz dick improvement? Why didn’t anyone tell me?! Wait... all that spam mail – it was real! 😱
@proedie Well, after the blazing success of the Reefer Madness lamentation of how virile cannabis would make Black men and the War On Drugs lamentation of how virile crack cocaine would make Black men, Reagan seized on a campaign of lamenting of how virile "steroids" would make Black men, and made most androgens into controlled substances in USA. He was not as successful of selling his War On Steroids overseas as Nixon had been selling his War On Drugs, but still, quite a number of influential countries bought it, and, well, here we now are: TRT artificially restricted, creating space for both charlatans selling bullshit and unscrupulous merchants selling androgen preparations with some, possibly ill-measured doses, of real ingredients, and possibly also some contaminants.
Netzblockierer@thomasfuchs cuz you can't Copyright AI slop in any decent juriatictions, since that necessitates Authorship!
Alexander Knochel@thomasfuchs Because it's not amazing for writing Books and Articles wholesale, specifically
@thomasfuchs Welctually, a weird niche of this genre exists, but it seems to be mostly Internet-based, at least for now.
But for a serious answer to your question: it's most likely mainly because genAI is primarily marketed as a replacement for humans, not as a high-quality source of knowledge. ChatGPT tried the latter in the beginning, but, well, they had to change course, because its falsity was too obvious.
@thomasfuchs This is a non-grifty counterexample. (The "AI" they used is not an LLM entity, but it is a large pattern-seeking neural network, and the research paper specifically discusses techniques for suppressing hallucinations.)
