I'm not sure that doing these folks the favor of assuming there must be rational thought behind their assertions is the right strategic move.

is this a number that has any reference or are you just caught up in the exuberance?

He used a genome language model...? Like a statistical model that shows the relationship between different parts of the genome...?

How did you guys think they did this in the past...?

There's... nothing new here. This is relying on people not understanding what these terms mean, including perhaps the author.

I'm pretty sure what's actually happened here (that is, I heard through the grapevine, read some of the notes, have seen conversations between established mathematicians who are using it at the moment) that someone has built a natural language interface to Lean.

This is only relatively true.

Nothing will make a general purpose statistical neural network more efficient at arithmetic than just doing it directly on the CPU.

That's the rub innit

The business case has always had the unstated assumption built into it that curating the data properly was waaaay too expensive. Additionally though... even doing that wont get rid of the potential for invalid or unwanted associations and overtraining.

hrrrm.... Softbank are up to their eyeballs and represent a fair chunk of Japanese capital

It will primarily hit the US, that's true... but it wont be confined there.

Based on how my own indexed fund is set up (what we call 'superannuation' in Australia), I expect that I'll also take a hit... I'll just be insulated by my other investments (hopefully).

I can see some research institutes snapping up some cheap just-out-of-warranty Blackwell 200s and running them until they die. I've got proximity to that, so it's visible to me... I can't see a lot of other use cases. I imagine they're going to be similarly niche.

Most HPC systems use commodity hardware, just a lot of it in cluster. The most recent nodes we have on the system I help to manage are equipped with either ECC DDR5 5600 or 6000, depending on which batch they came in.

Zitron's analysis definitely agrees with you about who will be hurt if/when this all crashes... and I can't see how he's wrong about OpenAI's commitments vs their income. I can't see how any of it go to this point unless the market operation selected for people who were irresponsible (but lets face it, that's not exactly a new idea).

Yeah this isn't new either. I first remember reading this as speculation over a decade ago.

Oh I'm not disagreeing with you.

And yet.

Please allow me to direct your attention to the entirety of the medical device industry. Also the automotive industry.

et cetera

I'm not sure I understand your point. I'm not really concerned with the label, more the function. I don't agree that gaslighting oneself is sufficient to solve physical problems.

This.

Additionally, there are some people in the comments here who are straight up inventing things in order for the story to make any sense.

I wouldn't at all be surprised if this was yet another medical startup where there's an aim and a goal and a business plan that involves the technology coming into existence or gaining capabilities surreptitiously due to the business plan having a need for it.

The fact it's running on a statistical system doesn't automatically mean it knows which of the token interactions it has represent, for example, fractions. There's no reason to believe it'd be better at actually performing deductive statistical analysis any more than to to believe it's using actual integers to perform arithmetic.

That is to say... even if it is incidentally ok at it, there are going to be some absolute financial disasters in the foreseeable future that are due to humans making bad decisions on the back of inferred data that they believe is deduced.

In summary, no. You aren't missing anything.