natnat

What a depressing outcome. This could have powered hundreds of thousands of households, cheaply, without adding any CO2 to the atmosphere.

It's the severance cost, mostly.

This really isn't true. You should use different parameters (specifically, you can reduce the random_page_cost to a little over 1) on a SSD but there isn't a really compelling reason to use a completely different DBMS for SSDs.

Are there decent US based alternatives to Hetzner? I'd like to have my servers located in the US for a variety of reasons, but most of the alternatives I've seen to Hetzner seem to be pretty fly-by-night shops.

I think this has a lot to do with the mechanism of RoPE attention, where physical closeness in the code is a signal of relevance.

This reminds me of the "Natural engagement pattern" sketch that Mark Zuckerberg made about facebook content back in 2018: https://www.facebook.com/notes/751449002072082/

Content gets more engagement as it gets closer to the "policy line" of getting banned, and in a competitive information environment (an engagement maximizing algorithm) you end up with a lot of content close to the border of what's allowed.

I remember software working really badly in the early 2000s, when Microsoft had an unassailable monopoly over everything. Then there were a bunch of changes: Windows started getting better with Windows 7, Firefox and then Chrome started being usable instead of IE, and Google and Apple products were generally a huge breath of fresh air.

Since then, Google and Apple products have become just as bad as Microsoft's. I think this is because the industry has moved towards an oligopoly where no one is really challenging the big players anymore, just like Microsoft in the late 1990s. The big companies compete with each other, but in oblique ways that go after revenue not users.

Can someone link to a real source for this? Like, a paper or something? This seems very interesting and important and I'd prefer to look at something less sketchy than venturebeat.com

It's unfortunate people expect you to have social media like a girl asks me if I have Instagram and I'm weird to not have one, I get it they can scope you out too for safety but when I tried using that stuff I felt this pressure to post about something

Probably worth Googling something like [men who don't have social media] to think what women think about this, it's more positive than you might think :)

You're right that a hobbyist couldn't build something like search or maps or docs in a weekend, but a lot of what Google ships is boring webapps. And I promise that they are really hard to ship: approvals are annoying and time-consuming, and the infrastructure is designed for scalability and performance, not flexibility or iteration speed.

There's a joke inside Google that Google infra makes easy things hard and impossible things possible.

Approximately what proportion of the US federal budget is spent on scientific research? What proportion is spent on foreign aid? Looking up these values is a useful exercise.

I don't honestly think that technology is meaningfully downstream of money. A startup or hobbyist can build something that costs Google several million dollars in a weekend. Most of these systems are complex, but not as complex as e.g. an operating system.

But upgrading technology requires government administrative capacity. That's generally cheaper than outsourcing technology development to third parties, but does require a commitment to try to understand the thing you're managing.

Politicians don't hire competent administrators because they believe that building a solution yourself and buying a solution from a contractor are basically equivalent, which anyone on this website can tell you is not true. This is an easier problem to solve than most think, but it's not trivial. And it's really hard when you have clowns like Elon Musk purposefully destroying institutional knowledge for no good reason.

Starlark is nice but Google also has GCL...

Nuclear is really terrible at peak loads; it's a very reliable base load. It's basically impossible to quickly increase or decrease the output of a nuclear plant.

The PISA scores are below average in 2022 but not significantly so. Most countries have had a pretty bad decline in test scores due to covid, but that's (hopefully) a one-time problem and future cohorts will do better.

I'm not saying that's a good thing. We're a very rich country and we should do better.

But the US also has a lot of systematic problems that peer countries don't, like child poverty, that are likely a bigger cause of subpar test scores than bad schools.

I'm not trying to say that the US educational performance is great, just that it's not terrible. I think there's a lot of room for improvement, but there's a narrative that US schools are failing horribly and that just isn't true.

In what world is the US education worse-than-third-world? The US has universal literacy, phenomenal higher education institutions, and primary and secondary schools that are comparable to peer countries on average.

We spend lots of money on schools in the US and generally get pretty good outcomes. Our math scores leave something to be desired, but average reading scores are better than peer countries. There are plenty of exceptions, especially in poor neighborhoods, and chronic absenteeism is a real problem, but it's worth keeping sight of the fact that the vast majority of American schools are quite good.

Some examples:

* PISA scores: https://www.oecd.org/pisa/OECD_2022_PISA_Results_Comparing%2...

* PIRLS/TIMMS scores: https://nces.ed.gov/fastfacts/display.asp?id=1

This is insane conspiratorial thinking? How would being the only host that happens to get DDOSed constantly be a good business proposition?

We use a kind of funky BigQuery setup at my shop to reduce our spend, and I think it's based on a quirk in BigQuery pricing that Google doesn't explicitly recommend.

BigQuery has two completely separate pricing models: on-demand, and slot-based. The two models bill on completely separate things:

* On-demand pricing bills per GB of data processed by your query.

* Slot-based (or editions) pricing allocates your project a number of CPU/memory slots, and you pay per CPU-second.

You can find the costs each query would have by looking at the total_bytes_billed and total_slot_ms columns in the INFORMATION_SCHEMA.JOBS_BY_ORGANIZATION table, and multiplying those values by the slot-ms cost (total_slot_ms * 0.01111) and the bytes-billed cost (total_bytes_billed * 0.0000059). Then you can go through your queries and allocate them to either on-demand or slot-based pricing, depending on which is cheaper.

Usually slot-based is cheaper, but queries that do a lot of internal joins can have really huge CPU costs but cost very little in on-demand if they're not reading a lot of bytes.

Somewhat annoyingly, these billing models are configured at a per-project level; you can't switch between the two of them in a single project. Fortunately, you can query tables from other projects easily.

If you don't want to do very much, you can try adsense. If nothing else you'll get a good sense of what the floor is.

PAPERCLIP

There was a really good episode of the Odd Lots podcast where they discussed this. Arizona is a very favorable location for semiconductor fabs because there's a very low risk for natural disaster (no hurricanes or earthquakes), the temperature changes are relatively predictable, land and electricity are relatively cheap, and Intel already has a large fab there. Water is a little bit of a struggle, but they can buy water rights from farmers, who use most of Arizona's water.

As an xoogler, I don't agree at all. Google has a ton of problems, but a reluctance to think deeply about and fix tech debt isn't one of them. In fact, I think the eagerness to fix fundamentally broken abstractions is a big reason Google has so much code churn and is so eager to irresponsibly deprecate things that work!

Completely! I'm not trying to disparage the progress that's being made.

This is a great start, I just wanted to make it clear that there's a long way to go :)

This is great, but keep in mind this is like a third derivative metric. North America is way behind China in number of batteries owned, batteries produced, and factories being built. This is just the growth rate of factories being built.

This a really classically difficult political problem of concentrated benefits with diffuse costs. The complication of the tax code comes from a zillion little exceptions added over the years, and every little exception has a beneficiary who doesn't want the existing system to change.

Imagine you're a congressperson trying to simplify the tax code. If you want to remove a single little exception, you're going to piss someone off a whole lot for not much benefit. If you want to do a big refactor and lower everyone's rate while cutting a ton of exemptions, you need to fend off zillions of loud, empowered interest groups.

As recently as 2017, congress tried to simplify the tax code with the TCJA. Paul Ryan and co tried very hard to make the tax code simpler while also cutting rates, and the process failed under the enormous pressure from interest groups. They ended up leaving pretty much every exception in place, and only cutting a few high-value tax benefits (like the SALT deduction) that benefited people from the other political party.

When there's no clear path to actually solving the problem, a hack that makes things better in the interim can be a really helpful solution.

It basically does show everyone the same results. There's some session-based reranking and ads are heavily personalized, but there's basically no personalization in regular search.

Regardless of the merits of the case, this is a really funny exploration into how bad Google's chat products are.

Also worth reading: http://en.wikipedia.org/wiki/Bus_bunching

For some businesses, Yelp puts a big consumer alert on their business page when they catch the businesses soliciting reviews on craigslist or whatever: http://officialblog.yelp.com/2012/10/consumer-alerts-because...