I will add that nodes don’t stay still, either. A 2025 run on a node may have a bunch of improvements over a 2023 run on that same node.

And Google’s jump from Samsung to TSMC itself might be a bigger jump than a typical year over year improvement. Although it could also mean growing pains there, too.

Fun coincidence from Newton’s not quite right theory of gravity combined with his not quite right particle theory of light.

Well, by the time the Pixel 10 comes out, it’ll be 2 generations after the iPhone that used a SoC from TSMC’s 3nm node (the A17, used in iPhone 15 Pro, launched September 2023). I’d imagine it’ll have caught up some, but will still behind while Apple is presumably launching something from TSMC’s 2nm or A14 node at the same time.

Their algorithms are mostly public. Their training data/procedure, and the trained model’s parameter weights, are not.

Price per kw and price per kwh stored. And price per kwh over the expected lifetime of the battery itself (longevity and reliability and safety and disposal will have to be factored into total cost of ownership).

The ceiling is going to be lower than with lithium. Sodium ions themselves weigh about 3 times more than lithium, for the same +1 charge. So it’s not just that sodium is a certain number of years behind lithium. It’s that it’ll likely plateau at a point permanently behind where lithium will likely be.

Safari support means there’s benefit to web server support. Server support means there’s benefit to browser support in other browsers. Apple can kick start the network effects necessary to get this standard adopted.

Webp and heic are fine for web, but JPEG XL is special in that it actually has use for print-based and other ultra high resolution workflows, while also having the best path forward for migration from JPEG.

Yeah, I’m not a fan of AI but I’m generally of the view that anything posted on the internet, visible without a login, is fair game for indexing a search engine, snapshotting a backup (like the internet archive’s Wayback Machine), or running user extensions on (including ad blockers). Is training an AI model all that different?

Coal companies are literally going bankrupt as coal plants get decommissioned. When it comes to actual political power, the fossil fuel industry you want to watch out for is oil and gas, not coal.

Mine all the coal you want. If you don’t have anyone willing to buy from you, at a price that covers the cost of extraction, you will fail.

So even though the coal companies’ bankruptcies are getting them out of their cleanup and decommissioning obligations, the root cause of that is that coal just isn’t competitive as an energy source.

until these get produced for real in mass quantities, they are vaporware

The world is already seeing exponential growth in annual completion of grid scale battery storage. Here’s some recent data in the US, as products and projects mature from theoretical to small scale prototypes to full scale pilot projects to full production.

And author should compare winter moths

There’s also significant developments being made in geothermal, which is actually dispatchable. Plus we actually still produce more grid-connected wind than solar right now, it’s just that solar is so damn cheap it makes sense to install capacity well beyond matching peak demand.

Some combination of overcapacity, demand-shifting, and storage will go a long way in reducing the amount of dispatchable fossil fuel capacity that is necessary.

The problem is that we’re not getting rid of the other stuff

We are, though. Coal use in the United States has cut in half in the last 15 years, and it’s still on a steep downward slope. Even as natural gas (which emits roughly half the CO2 per unit energy as coal) increased over the same time period, our total emissions from energy consumption has dropped from about 6 billion tons to 4.8 billion tons.

The progress we’re making might be slower than many of us would like, but we’re also at a tipping point where we’re making many fossil fuels simply uneconomical. And that’s the key: to make polluting costly enough that big businesses won’t want to.

I’m not sure that’s true.

Well, I’m sure it’s true. I’ve started and stopped Prime benefits multiple times.

Don’t set a reminder, just cancel now. If you cancel, you get the rest of the time you paid for and it just doesn’t automatically review, so there’s no penalty to canceling early versus right before the deadline.

Wait is he fucking them? I thought most of these children were born via IVF.

Realistically, I would grieve the loss of my children, who would never be born if I didn’t line things up just right to cause them to happen again. I’d spend more time with my parents, who are getting along in the years, and I’d make the most of my time with them while they’re healthy and happy.

There are a few specifics where I’d try to get some loved ones out of trouble before some critical tipping point that would later cause a bunch of heartache and stress.

There are general things about money and politics I’d probably do differently, knowing about how stocks have performed and what not, but that’s not super interesting to me, because I’m mostly content in my personal life (including my career) and wouldn’t want to upset that balance by doing anything too different from what brought me here.

Well they can pay compensation to people who do work for them: employee salaries, contractor work, etc. So the nonprofit structure might prevent them from paying dividends or stock buybacks or other ways of transferring directly to shareholders in their capacity as shareholders, but nonprofit structure alone isn’t a guarantee that the organization won’t steer excess cash into someone’s pocket.

No reason to believe this is true of this non-profit, but that’s the reason why it’s important to look at the books of nonprofits that you donate to.

I’ve been a general skeptic of exactly how much the power and performance to power stats are attributable to the ARM instruction set or architecture versus the fact that Apple just locks up TSMC’s latest and greatest node for a year before everyone else. AMD’s CPUs are still x86_64 but achieve similar performance per watt as the Apple silicon on the same node and similar TDPs.

So if it turns out that TSMC has the secret sauce, then maybe we don’t need to move laptops over to ARM at all.

Don’t expect me to pay 2 grand for a laptop with no external USB or HDMI ports

HDMI is a trash standard and we should move everyone to Displayport at a minimum, or Thunderbolt. But also, all the MacBook Pros have HDMI.

I was even more frustrated by how bad the built in trackpad

Weird. I think Apple’s trackpad is head and shoulders better than anyone else’s.

The New Yorker’s style guide requires markers for coöperate, coöpt, etc., but it’s non-standard outside of that one particular publication.

Yes, but the tokens are more than just a stream of letters, and aren’t saved in the form of words. The information itself is organized into conceptual proximity to other concepts (and distinct from the text itself), and weighted in a way consistent with its training.

That’s why these models can use analogies and metaphors in a persuasive way, in certain contexts. Mix concepts that the training data has never been shown before, and these LLMs can still output something consistent with those concepts.

Anthropic played around with their own model, emphasizing or deemphasizng particular concepts to observe some unexpected behavior.

And we’d have trouble saying whether a model “knows” something if we don’t have a robust definition of when and whether a human brain “knows” something.