Regression To The Meme22h ago
Gabriele Svelto
Given what it's gonna do to the world economy when the bubble will burst I think it's better if we drop all references to intelligence and just start calling it subprime computing
Mar 25 at 4:29pmWeb
Show threadNazo23h ago

@gabrielesvelto I think we need to call it "scam" because people just don't seem to be getting it through their heads that that's all it is. Pretending LLMs are or ever will be "AI" is absolutely a scam. Some CEOs and such are scamming themselves too I guess, but nonetheless, even if they believe it themselves, it doesn't change that a scam is exactly what it is. Anyone who knows even the basics of the mechanisms of how they work will understand they are not nor ever can have either the A or the I... But part of the scam is that most never bother to look into even the basics of it.

I feel like the best thing we can do right now is straight up just call it that: scam.

Show threadgotofritz23h ago

@nazokiyoubinbou @gabrielesvelto

Dude nobody care if they are "AI" or not. Do they work? Do they it help? That's what matters.

Show threadNazo23h ago

@gotofritz @gabrielesvelto Given that the whole scam is pushing them as "AI," yes, everyone but you cares. About 95% of the scam is in what they supposedly will do, not what they do right now. Except they won't. Ever. The fundamental mechanisms themselves simply can't.

And no, they don't work right. A huge part of it involves doing everything possible to ignore that such as mandating worker usage be certain amounts or etc.

Show threadgotofritz23h ago

@nazokiyoubinbou @gabrielesvelto

A lot of wild, unsubstantiated statements based on nothing but vibes, but whatever floats your boat 🤷

Show threadNazo23h ago

@gotofritz @gabrielesvelto Actually, I have a number of LLM models and utilize them for various tasks. My statements are based on an understanding of the actual mechanisms of how the technology works. There is a reason people call them things like "statistics machines" or even "overgrown spell checker." They have potential uses, but they can't exceed their most fundamental mechanism. The potential uses are small and limited and never should have been crammed into faking general "AI."

Meanwhile, as I recall, it is using LLMs to write code that is called "vibe coding" so... Yeah... I would pick a different word to use as an insult while supporting, you know, that...

Show threadgotofritz23h ago

@nazokiyoubinbou @gabrielesvelto

🤦‍♂️ 'vibes' was on purpose you silly goose

Show threadNtropic6h ago
@nazokiyoubinbou @gotofritz @gabrielesvelto Vibe Coding is when you don't care about the generated code. There areany ways to code with the aid of LLMs.
Show threadNoa21h ago
@nazokiyoubinbou @gabrielesvelto call it Annexed Internet, or soon to be Apartheid Internet
Show threadBornach8h ago

@nazokiyoubinbou @gabrielesvelto
To be fair, "artificial" is in the name

Cannot really claim buyers of artificial fur, artificial leather, artificial jewels, etc didn't know they would not be getting the real thing

Show threadGabriele Svelto7h ago
@nazokiyoubinbou it is. This is the biggest Ponzi scheme ever devised, with the complicity of the US administration, tech and hedge fund CEOs. They'll make it out of this dirty rich while trying to socialize the catastrophe that they will have caused.
Show threadyour auntifa liza 🇵🇷 🦛 🦦23h ago
@gabrielesvelto this is one of those moments that i cannot believe i hadn't thought of that. well done.
Show threadGabriele Svelto20h ago
@blogdiva thank you!
Show threadAlan23h ago
@gabrielesvelto I think it's sort of going on a tangent to put the blame on the tech when there is so much bad management at the core of the bubble. AI is also being used as a scapegoat to cover the results of previous bad management decisions.
Show threadcan everything be outside now?23h ago
@metaphase @gabrielesvelto yes, to a certain extent a much larger segment of tech is subprime than just AI.
SaaS companies whose valuations are based on subscriptions by other companies owned/funded by the same VCs/investment groups are a huge vulnerability for example.
Show threadGabriele Svelto7h ago
@metaphase absolutely. This is a playbook that has already been tried by the same people with other technologies: the internet of things, the metaverse, cryptocurrencies, NFTs, etc... they've never been so successful before.
Show threadJames Kemp23h ago
@gabrielesvelto there's an argument that it's sub-prime MBAs underlying the sub-prime computing!
Show threadFinal Round Player 😷🇪🇺🍸22h ago

@gabrielesvelto I look forward to investing in AI Backed Securities.

And the Quantitative Easing when the bubble bursts.

Show threadBornach8h ago

@jbond @gabrielesvelto
You can do that right now on the Polymarket

AI written article on AI prediction of AI model performance based on bets placed by AI on prediction markets that AI tech bros are claiming is the next great thing
https://www.ainewshub.org/post/ai-powered-forecasting-how-prediction-markets-like-polymarket-are-enhancing-tech-trend-predictions

AI-Powered Forecasting: How Prediction Markets Like Polymarket Are Enhancing Tech Trend Predictions

Discover how Polymarket and AI-powered forecasting are reshaping tech trend predictions. From real-time market signals to autonomous trading agents, learn how prediction markets are becoming the smartest tool for investors and tech enthusiasts in 2026.

AI News Hub
Show threadДими́трий22h ago

@gabrielesvelto

throw in an energy crisis and you've got a deal!

Show threadBornach8h ago
@saltywizard @gabrielesvelto
Recalling back in 2007 when $100 oil suddenly made living in the ex-urbs unviable
https://news.oregonstate.edu/news/economists-say-high-gas-prices-triggered-housing-crisis-2007
Economists say that high gas prices triggered the housing crisis in 2007 | Newsroom

CORVALLIS, Ore. - A new study by economists at the University of California, Berkeley, and Oregon State University suggests that the U.S. housing crisis that started in 2007 and eventually led to a worldwide financial crisis was triggered by rapidly rising gas prices. Economic bubbles have a long history - from the Dutch tulip mania in the 1600s to this latest housing bubble - and many factors can cause them to inflate, the researchers say. But little is known about what actually triggers a bubble to burst.

Newsroom
Show threadTheolodian8h ago
@bornach @saltywizard @gabrielesvelto 2027 says hold my beer…
Show threadBritstralian20h ago
@gabrielesvelto 100%
Show threadBarry Cook 🇨🇦16h ago
@gabrielesvelto - beyond Artificial Intelligence is the realization that no Intelligence was involved.
Show threadR.L. Dane    🍵 12h ago

@gabrielesvelto

Holy crap, that's brilliant.

Show threadGabriele Svelto7h ago
@rl_dane thank you!
Show threadNtropic4h ago
@gabrielesvelto Marketing being scammy is nothing new.
The US will probably not be in a position to help the companies very much, as the US can't borrow as foreign governments have little interest in US bonds. Since AI companies aren't relevant for the wider economy (t's not housing or energy or food), this will mostly just correct peoples retirement funds after their explosion over the last 4 years. A market correction.
Show threadNtropic4h ago
@gabrielesvelto The comanies are currently operating at a loss, but that's not necessary. Due to the exponential scaling laws, distillation and quantization, we can run 100x smaller models at 100x lower price with almost the same performance. As investor money runs out, companies will simply have to run models at a size that covers the costs. Google will likely win that race, since they can continue operating at a loss for longer due to their remaining revenue streams and in house tensor chips
Show threadGabriele Svelto4h ago
@Ntropic assuming that happens it still doesn't solve the issue with training. Training is a constant unavoidable price, and the presence of LLM outputs in the inputs means that filtering is becoming more and more expensive, in turning making training more and more expensive. The alternative is accepting at least a measure of model collapse. It also assumes a favorable legal environment and - at least in the EU - that's not gonna last for long.
Show threadNtropic3h ago

@gabrielesvelto I can keep using existing models, I don't see why one would "need" to train new models over and over.

The training for scaling to larger models will not be maintainable at the current rate, but it is also not necessary.

For model knowledge to not be outdated, new information can be introduced with many much more efficient approaches, from old school LoRA to SotA adding Engrams. We mainly train to implement new tricks, such as more efficient architectures and add new concepts

Show threadNtropic3h ago
@gabrielesvelto in the mid to long term RL post training will likely keep increasing, as this has shown the bigget benefits and its scalablility has been much improved. Model decay would come from pre-training data, and that is already available and only slightly increased between model versions - probably because they want to avoid model decay.