It’s one of those things where periodically someone gets sanctioned and a few others get scared and stop doing it (or tone it down) for a while.

I guess SHEIN are either overdoing it or they crossed the popularity threshold where companies become more scrutinized

I’ve never used SHEIN so I can’t tell if they are using these practices or how bad they are, but from the article I see they allegedly use fake urgency messaging, which I know has been sanctioned before in the EU (the company I used to work with had to rush removing it from our eCommerce site).

A company can tell you that the item you’re looking at happens to be the last one in stock, if it’s true. But if they lie about it, so you rush into a decision to buy it before it’s gone, then it’s a deceptive practice.

Depends what you mean by “valid”. If you mean “profitable”, sure: Fraud has always been a profitable business model.

But if you mean “valid” in terms of what Microsoft got out of their $455M investment, not so much, as they wanted a great new AI model, not the output that the “human-powered” model produced pretending to be an AI.

I agree. I was almost skipping it because of the title, but the article is nuanced and has some very good reflections on topics other that AI. Every technical progress is a tradeoff. The article mentions cars to get to the grocery store and how there are advantages in walking that we give up when always using a car. Are cars in general a stupid and useless technology? No, but we need to be aware of where the tradeoffs are. And eventually most of these tradeoffs are economic in nature.

By industrializing the production of carpets we might have lost some of our collective ability to produce those hand-made masterpieces of old, but we get to buy ok-looking carpets for cheap.

By reducing and industrializing the production of text content, our mastery of language is declining, but we get to read a lot of not-very-good content for free. This pre-dates AI btw, as can be seen by standardized tests in schools everywhere.

The new thing about GenAI, though is that it upends the promise that technology was going to do the grueling, boring work for us and free up time for us to do the creative things that give us joy. I feel the roles have reversed: even when I have to write an email or a piece of coding, AI does the creative piece and I’m the glorified proofreader and corrector.

cover letters, meeting notes, some process documentation: the stuff that for some reason “needs” to be done, usually written by people who don’t want to write it for people who don’t want to read it. That’s all perfect for GenAI.

the point I was trying to make is that the reason both pro and anti-AI sentiments are blind is because “AI” companies are purposely mixing up things that don’t belong together, in order to attract investments.

If you wrote “cruise ships are generating a lot of pollution” and someone answered “but it Magellan or Columbus hadn’t had ships, our knowledge of the World wouldn’t have advanced”, you’d think they are gaslighting you, right? You wouldn’t say “this blind anti-ship sentiment is going to hurt geography”

Machine Learning Models have existed for a long time. They are at their core predictors: you give them data, you carefully tweak the model’s parameters for a long time and you can finally train a model that can make predictions in a specific domain. That way you can have a model trained specifically to identify patterns that look like cancer on medical imaging or another one (like in your example) to predict a protein’s structure.

LLMs are ML models too, but they are trained on language. They learn to identify patterns in human language and predict long pieces of text that are similar to those language patterns. They also accept input in natural language.

The hype consists in slapping a new “AI” marketing label onto all of Machine Learning, mixing LLMs and other types of models, and creating the delusion that predicting a protein’s structure was done by people at Google casually throwing prompts at Gemini.

And as these LLMs are exceptionally power-hungry and super expensive (turns out that predicting human language based on a whole internet’s worth of training requires incredibly complex models), that hype is to gather all the needed trillions of investment. GenAI is not the whole of Machine Learning and saying “Copilot is not worth the cost of the energy that’s needed to power it” doesn’t mean creating obstacles to ML used for cancer research.

In other news: AI is a better human than Duolingo CEO

You are right. Bunch of incel 19-year-olds… This is probably more about hiding their browser history from their moms

Easy, we just give AI access to all our files and personal information and it will know our age!

Look up stuff where? Some things are verifiable more or less directly: the Moon is not 80% made of cheese,adding glue to pizza is not healthy, the average human hand does not have seven fingers. A “reasoning” model might do better with those than current LLMs.

But for a lot of our knowledge, verifying means “I say X because here are two reputable sources that say X”. For that, having AI-generated text creeping up everywhere (including peer-reviewed scientific papers, that tend to be considered reputable) is blurring the line between truth and “hallucination” for both LLMs and humans

Basically, model collapse happens when the training data no longer matches real-world data

I’m more concerned about LLMs collaping the whole idea of “real-world”.

I’m not a machine learning expert but I do get the basic concept of training a model and then evaluating its output against real data. But the whole thing rests on the idea that you have a model trained with relatively small samples of the real world and a big, clearly distinct “real world” to check the model’s performance.

If LLMs have already ingested basically the entire information in the “real world” and their output is so pervasive that you can’t easily tell what’s true and what’s AI-generated slop “how do we train our models now” is not my main concern.

As an example, take the judges who found made-up cases because lawyers used a LLM. What happens if made-up cases are referenced in several other places, including some legal textbooks used in Law Schools? Don’t they become part of the “real world”?

I tried reading the paper. There is a free preprint version on arxiv. This page (from the article linked by OP) also links the code they used and the data they tried compressing, in the end.

While most of the theory is above my head, the basic intuition is that compression improves if you have some level of “understanding” or higher-level context of the data you are compressing. And LLMs are generally better at doing that than numeric algorithms.

As an example if you recognize a sequence of letters as the first chapter of the book Moby-Dick you’ll probably transmit that information more efficiently than a compression algorithm. “The first chapter of Moby-Dick”; there … I just did it.

Especially right now, I’m feeling lots of things but “lucky” ain’t one…

These are more realistic. I hear they can form a flock and shit all over you. Scary stuff!

Can’t wait for Trump to do his thing and backstab them the second this publicity stunt loses media attention in 3…2…1…

They’ll probably get off the plane, take a couple of selfies and board another one directly to El Salvador.

When we say “Ye did this”, how many people actually recorded, played, produced, published this? I know next to nothing about modern music production, so maybe now he can do all of that at home with a laptop and an internet connection, but if there are others making money out of a clearly unstable person and getting none of the hate, I’d like at least to try and avoid giving them my money

Still apparently no recorded kills, even if it’s not something he would have had interest in hiding

“you had me at extraordinarily expensive” DJT