KurSix

I think the weirdest part of that scenario is that science might become something closer to archaeology

I'm not sure the scarce resource will be people capable of understanding the AI's work. It may be institutions willing to tolerate the cost of understanding it

HBM isn't dramatically better in every dimension. You get huge bandwidth and good energy efficiency per bit transferred, but not necessarily a meaningful latency improvement and capacity expansion becomes tied to the package

Yeah, the scale is kind of mind-bending. You're taking something that's already extremely fragile, grinding most of its thickness away, handling it through several more process steps, and somehow doing that cheaply enough for mass-market products.

Sanctions can absolutely create the incentive to build a domestic supply chain, but incentive doesn't automatically translate into catching up on the same timeline

China has had a very strong incentive to recreate lithography and fab equipment for years, but a lot of the specialized HBM packaging chain was probably much lower priority until relatively recently

You probably can't make this reliable across every modern car without breaking something, but that's different from technically impossible

It should be as mundane as pulling the SIM card from a phone (or must be)

At that point "opt out" starts to feel like a UI feature rather than an actual description of what the car is doing

If AI makes code more stylistically uniform, it might actually make those deeper choices more visible

If software becomes dramatically cheaper to produce, that may just give investors even more software to invest in

Japan is probably evidence that the surrounding food culture matters at least as much as the existence of UPFs. You can sell plenty of processed food without getting the same outcome if portion sizes, eating habits, school meals, and daily activity are different. The marijuana comparison feels like a non sequitur, though.

Shelf-stable hummus and Doritos can both be "ultra-processed", but that doesn't tell you much about why one might encourage overconsumption. Hyperpalatability seems much closer to the actual claim the article is trying to make.

It's probably less "modern antidepressants are uniquely bad" and more "we got much better at starting people on safer drugs than we did at figuring out how to stop them"

For something prescribed to get someone through a temporary crisis, "how and when do we get you back off this safely?" should probably be discussed at the beginning, not years later when the patient decides to stop.

What seems most broken is that medicine apparently didn't have a great way to predict which group you’d fall into, or much of a plan for the people who had a really bad time stopping. Also, RIP the anti-spider benefits

The sound especially feels almost impossible to date

I wonder if part of that is that Blade Runner doesn't optimize for the first viewing. Once you already know where the plot is going, your attention is free to wander into the production design, sound, background activity

Wow... The venue almost adds another layer to the movie. You're watching a 1982 vision of 2019, accompanied by live electronic-orchestral music, inside a theater built a couple thousand years ago...

The sound is probably why the visuals feel so much larger than the sets actually were

Soon the GPUs will just be the promotional gift you get for opening a sufficiently large Nvidia financing account

I think they just mean "is expected to guarantee". Since the deal hasn't been signed, Reuters is hedging, but "may guarantee" is definitely awkward wording for something whose entire purpose is certainty...

The permanent jobs number really puts it in perspective. You're not just dropping a giant data center into an existing community at that point, you're potentially reshaping the county around it

The $500B number is hard to even contextualize. Calling it a "data center" almost undersells what is being proposed

The numbers have become so large that normal corporate risk management starts looking quaint

The harder test would be whether they can come up with the right invariants and abstraction in the first place

"The market chose it" doesn't mean it was the best solution, just that it fit the incentives

The problem may not be that LLM taste is bad, but that it is incredibly easy to converge on

Yeah once taste can be fully reduced to rules, it stops being taste and becomes something you can optimize for

Generating the first version is almost free now. Understanding why that version is subtly wrong, debugging it in production, maintaining it six months later and deciding whether the abstraction was a mistake are all still expensive...