jonas21

It's perfectly legal under US law to donate to a for-profit corporation. The donor just can't take a tax deduction for it.

The other reason water usage is a bad thing to focus on is that datacenters don't inherently have to use water. It's not like servers have a spigot where you pour water in and it gets consumed.

Water is used in modern datacenters for evaporative cooling, and the reason it's used is to save energy -- it's typically around 10% more energy efficient overall than normal air conditioning. These datacenters often have a PUE of under 1.1, meaning they're over 90% efficient at using power for compute, and evaporative cooling is one of the reasons they're able to achieve such high efficiency.

If governments wanted to, they could mandate that datacenters use air conditioning instead of evaporative cooling, and water usage would drop to near zero (just enough for the restrooms, watering the plants, etc). But nobody would ever seriously suggest doing this because it would be using more of a valuable resource (electricity / CO2 emissions) to save a small amount of a cheap and relatively plentiful resource (water).

I am a human, and I would have done the same thing as Nano Banana. If the user had wanted a strawberry in the skull's left eye, they should've said, "Put a strawberry in its left eye socket."

It seems you may be misremembering. From Wikipedia [1]:

> Past versions of RealPlayer have been criticized for containing adware and spyware such as Comet Cursor. ... PC World magazine named RealPlayer (1999 Version) as number 2 in its 2006 list "The 25 Worst Tech Products of All Time", writing that RealPlayer "had a disturbing way of making itself a little too much at home on your PC--installing itself as the default media player, taking liberties with your Windows Registry, popping up annoying 'messages' that were really just advertisements, and so on."

[1] https://en.wikipedia.org/wiki/RealPlayer

Dario Amodei from Anthropic has made the claim that if you looked at each model as a separate business, it would be profitable [1], i.e. each model brings in more revenue over its lifetime than the total of training + inference costs. It's only because you're simultaneously training the next generation of models, which are larger and more expensive to train, but aren't generating revenue yet, that the company as a whole loses money in a given year.

Now, it's not like he opened up Anthropic's books for an audit, so you don't necessarily have to trust him. But you do need to believe that either (a) what he is saying is roughly true or (b) he is making the sort of fraudulent statements that could get you sent to prison.

[1] https://www.youtube.com/watch?v=GcqQ1ebBqkc&t=1014s

It's also a marketing channel for future flights. The app almost certainly asks for notification permissions, and most people will say yes -- they're useful for knowing if your flight is delayed or there's a gate change.

Now they have a channel where they can let you know about deals, etc. I'm sure they've modeled exactly how much this is worth, and I'd be willing to bet it's a lot.

How did you manage to write off for your expenses for tax purposes? I thought this was not possible for regular employees in the US (your profile says you're in the New York). I'd love to be able to do this.

> Even doing independent contracting on the side seems like a minefield w.r.t. quarterly tax payments, estimated tax, etc.. in the US at least. I'm sure I could figure it out but just earning income for some side work seems like a liability and a big headache.

It's really not hard -- it takes maybe 15 minutes per year, and if you own any stocks outside a 401k, you probably already need to make quarterly estimated payments. If it's too much trouble, you can just hire an accountant to take care of it for you. At any rate, it shouldn't be a significant obstacle to contracting.

San Diego's airport, on the other hand, has the a bustling restaurant district, an interstate with frequent bumper-to-bumper traffic, and a dense residential neighborhood all within a mile off one end of the runway -- and a popular shopping area, an elementary school, and a high school within just over a mile from the other end.

In addition, the terrain rises in both directions (so sharply on one side that planes can't use ILS when landing from that direction).

For $50, kids these days can buy a Raspberry Pi that would have run circles around the best PC money could buy when I was a kid.

Or, for $300, you can buy an RTX 5060 that is better than the best GPU from just 6 years ago. It's even faster than the top supercomputer in the world in 2003, one that cost $500 million to build.

I find it hard to pity kids who can't afford the absolute latest and greatest when stuff that would have absolutely blown my mind as a kid is available for cheap.

The $1.4T commitment is spread over multiple years. Let's assume 4 -- then that's $350B/year. Coincidentally, Google had $350B in revenue in 2024 (and projected to be ~$400B in 2025).

It's certainly possible to imagine OpenAI eventually generating far more revenue than Google, even without anything close to AGI. For example, if they were to improve productivity of 10% of the economy by 10% and capture a third of that value for themselves, that would be more than enough. Alternatively, displacing Google as the go-to place for search and selling ads against that would likely generate at least Google levels of revenue. Or some combination of both.

Is this guaranteed to happen? Of course not. But it's not in "bonkers" territory either.

Well, yes. Compared to most countries that have signed this treaty, the US has excellent protections for PII and data privacy.

But that's beside the point. The most objectionable parts are about state surveillance and the potential for human rights abuses.

For example, here's what the EFF had to say about it:

https://www.eff.org/deeplinks/2024/07/effs-concerns-about-un...

FAIR has its own biases, and these can be quite strong (have a glance at the studies on their website and judge for yourself).

IMO, the Columbia Journalism Review (https://www.cjr.org) is a better source for media criticism.

I feel like it was all pretty obvious by late 2017. Prototyping and development in PyTorch was so much easier - it felt just like writing normal Python code. And the supposed performance benefits of the static computation graph in TensorFlow didn't materialize for most workloads. Nobody wanted to use TensorFlow - though you often had to when working on existing codebases.

I think the only thing that could have saved TensorFlow at that point would have been some sort of enormous performance boost that would only work with their computation model. I'm assuming Google's plan was make it easy to run the same TensorFlow code on GPUs and TPUs, and then swoop in with TPUs that massively outperformed GPUs (at least on a performance per dollar basis). But that never really happened.

Out of curiosity, what field are you in?

You can turn it off in Chrome settings if you want.

It's unlikely that it would be accepted for publication in a top peer-reviewed journal if there wasn't something novel.

Even if we assume you're correct and every company's true mission is to maximize power and money, the stated mission is still useful in helping us understand how they plan to do this.

The questions in the original comment were really about the "how", and are still worth considering.

The NTSB's original report has more detail on how the SD Card was encrypted and how the NTSB managed to decrypt it:

https://data.ntsb.gov/Docket/Document/docBLOB?ID=18741602&Fi...

The percentages are added, not averaged. Each category sums to 10%, and the General Knowledge category has 5 equally-weighted subcategories, so 2% is the best possible score you can get in the social science subcategory.

I don't know why they decided to do it this way. It's very confusing.

I assume the author is reading the comments here and decided it would be in their best interest to make the repo private.

How so? It sounds like a pretty normal merger to me.

I don't really see the issue. If you don't want the face recognition feature, then you'll turn it off once, and that's that. Maybe if you're unsure, you might turn it off, and then back on, and then back off again. But what's the use case where you'd want to do this more than 3x per year?

Presumably, it's somewhat expensive to run face recognition on all of your photos. When you turn it off, they have to throw away the index (they'd better be doing this for privacy reasons), and then rebuild it from scratch when you turn the feature on again.

Yes, exactly! These are all examples of markets where a handful of winners (or sometimes only one) have emerged by investing large amounts of money in developing the technology, leaving everyone else behind.

No, it's more like a winner take all market, where a few winners will capture most of the value, and those who sit on the sidelines until everything is figured out are left fighting over the scraps.

Nice writeup. All of the technical detail is great!

I'm curious about the process of getting colo space. Did you use a broker? Did you negotiate, and if so, how large was the difference in price between what you initially were quoted and what you ended up paying?

Heinz, Hilton, Dell.

> eventually the list will become silly and the function will earn a # pyre-ignore annotation. This defeats the whole point of the pointless exercise.

No, this is the great thing about gradual typing! You can use it to catch errors and provide IDE assistance in the 90% of cases where things have well-defined types, and then turn it off in the remaining 10% where it gets in the way.

People love to bring this up, and it was a silly thing to say -- particularly since he didn't seem to understand that radiologists only spend a small part of their time reading scans.

But he said it in the context of a Q&A session that happened to be recorded. Unless you're a skilled politician who can give answers without actually saying anything, you're going to say silly things once in a while in unscripted settings.

Besides that, I'd hardly call Geoffrey Hinton an AI evangelist. He's more on the AI doomer side of the fence.

You'll probably find a high correlation with the street cleaning dataset (if there is one). Nearly all of the tickets at the top of the leaderboard are street sweeping violations.

I wonder if street cleaning is net profitable for the city once you factor in tickets. That would make cutting the cleaning frequency [1] a doubly bad idea.

[1] https://sfstandard.com/2025/02/18/san-francisco-city-hall-st...