There is really no way they did not notice that before. And then they obviously decided to let it run.

We do know that generating more "slop" type bullshit is what LLMs are really good at. Roslansky is perfectly right about that, if you want to obscure that you add no value, have no ideas and no insights, LLMs are the tools for you to hide that. Probably explains why so many people like them.

We still have to see any other uses outside of isolated and probably carefully crafted "examples" to the contrary. For example, apparently, that Hugging-Face stunt did cost them $20M in tokens, two or three orders of magnitude more than a competent black-hat hacker would have charged. And the math-proofs they apparently stole are not something they can repeat because mathematicians doing research will not give any more proof ideas to LLMs. They have all learned that lesson.

Wait, that is not right. Somebody automated an attack campaign with LLMs recently, at about $25 (!) per target: https://www.heise.de/en/news/A...
So for attackers, LLMs are bringing in the big bucks. But only for them. And they do so because lots of IT security is so abysmally bad that even slop can get in.

Indeed. On the rare occasions I till use MS office, I am constantly surprised (and annoyed) how dysfunctional MS office is compared to LibreOffice. They should have stopped "improving" it 20 years ago, before the atrocity of the "Ribbon" that only makes doing anything harder and wastes vertical screen space on top of that.

That seems very likely. They certainly have burned all goodwill they could burn and they essentially committed suicide in the EU. Because their crap is so deeply embedded, it will take a while to get rid of them, but the process is well underway and this time there is only one way this can end.

Indeed.

But, to be fair, LLM-type "AI" does enhance Bullshit generation by 10x or more, just as he pointed out. His mistake was not recognizing that MS products are mostly used for bullshit generation and hence the gains are real.

It should really not be needed to say this, but apparently it is: Do NOT trust your phone. Your phone is not your friend. Unless you are an expert user of encryption, whatever is on your phone is not secure. Your phone will record your location and track you. Your phone is an attack vector.

The cost of LLM type "AI" is getting completely unsustainable, while the benefits refuse to materialize and the problems are getting more and more obvious. It is time this ends and sanity returns.

Exactly. He spoke truth ti power and power (the cretin at the top of MS) cannot have that. ne wonders how somebody like that managed to raise at MS though.

And he apparently did some good work, like finally getting Teams a way to test Audio, the absence of which causes problems time and again due to the crappy Windows audio set-up. I am now investigating alternatives as enshittification and sloppification is to be expected.

Actually two. The EU is just more quiet about its investigations.

I was mentally in the wrong thread. My apologies.

Well, it certainly can replace management consultants, were "high confidence while being clueless" is basically the job description. But these people do not have positive effects on their customers or on society. Jenny Tian did sum it up very nicely in this skit: https://www.youtube.com/watch?...

It is still astonishing how many people do not know how meaningless and easily manipulated benchmark results actually are and not only in this space.

Personally, I find the existence of these people just sad. But yes, pissing them off is a service to society. The last thing we want is have them accepted and accommodated. These people are the enemy of anything civilized and they serve only their own narrow interests.

There are definitely multiple constrains and most are hard. One we are seeing now is that they obviously have trouble financing the training (which is the obvious real reason they are talking about "slowdown"). And sure, if there was some fundamental idea to accelerate that aspect on the level of what the transformer idea did, sure. But that is fundamental research and you cannot force that. Somebody needs to have the right idea. You can try to get exceptionally smart people to work on it (but it looks like the exceptionally smart people are al leaving...) but that does not assure results either. And using tons of not-so-smart people does absolutely nothing. Money cannot really accelerate fundamental research, unlike applied research or product design, where money can do a lot. Hence it may happen tomorrow, it may happen in 10 years, it may happen in 100 years, it may happen in 1000 years or it may never happen, because the limit is fundamental or the idea is too complex. There is no way to know.

So they essentially made some unsupported claims? Well.