brianolson

Yield. Some fraction of cores had a speck of dust or something, but at 38/40 good cores per chip they got economical yield

Why aren’t these AI companies submitting to the TOP500 to show off their computing prowess?

my knowledge is 10+ years out of date, but once upon a time if they'd chosen to, Google could have had _several_ entries in the top 10 of the TOP500 list

It's just poker, they didn't want to tip their hand

yup. me too. once upon a time that was a 2 week assignment in graphics class in college

I worked in blockchain 5 years trying to build good tech, and speculators ruin everything and don't care about good tech, but the real punch line is that

good tech is like 2% of a real answer

Here's a rant about a bunch of other layers of 'so you want to move money'

https://voidfox.com/blog/payment_processor_fun_2025_making_y...

and different parts of the blockchain ecosystem are working on some parts of that, and all together they're still a long ways off

I worked in blockchain ("builder") for 5 years. I started 'eh, there are speculators, whatever, I build good tech' but finished 'holy crap speculators completely dominate and distort everything, nobody cares about good tech'

"In select GC-heavy microbenchmarks ... we observed anywhere from a 10–50% reduction in GC CPU costs"

- Yay!

"The Go compiler benchmarks appear to inconsistently show a very slight regression (0.5%)"

- Boo

"Green Tea is available as an experiment at tip-of-tree and is planned as to be available as an opt-in experiment in Go 1.25"

I definitely know some application code that spends 30% of CPU time in GC that needs to try this.

I designed and named the Transaction Execution Approval Language for the Algorand blockchain in 2020. I'm partial to the original, but as it grew it got rebranded to be the "Algorand Virtual Machine". Glad someone still remembers it as TEAL!

Great concept. Bring back real small trucks. My grandpa ran a farm with a truck this size.

Disappointed in towing capacity of 1000 pounds ish. I can already do 1700lb on my hybrid rav4

I built a 'slice of function pointers' bytecode interpreter in Go in 2019 for the Algorand VM (Blockchain smart contract stuff) and before that the same pattern in C for a toy JVM around 2005.

It's a good pattern!

The Algorand VM was focused on low overhead running thousands of tiny programs per second. Version 1 had no loops and a 1000 instruction limit.

The JVM was focused on low memory, towards possible embedded microcontroller use.

So, 'array of function pointers' is nothing new, but it is a good pattern.

I think the 'law of large numbers' says that it's very unlikely for you to follow 4k and have _none_ of them posting. You could artificially construct a counter-example by finding 4k open but silent accounts, but that's silly.

The other workaround is: follow everyone. Write some code to get what you want out of the jetstream event feed. https://docs.bsky.app/blog/jetstream

If you want to dedicate a VM to it, bsky Personal Data Server has a pretty easy install

https://github.com/bluesky-social/pds

95% of end users don't care; but Bluesky has the right bits built in anyway. There's a grand central aggregator of all 13 million accounts, but it's not _special_, someone else could run one (several hobbiests are processing this level of data). Migration works* (and works better than Mastodon, all your history and network move even better than a masto server move) (*okay, it's a weird command line tool at the moment, but as soon as someone cares that'll get cleaned up). You can run your own Personal Data Server and hook it in to the bsky network and then everyone can see your posts and interact with them. It's newer, only a couple years old, but all the right parts are headed in the right direction.

atproto PDSes are like blog servers with RSS (but better) and bsky.app is the prevailing RSS reader. It's an open protocol because anyone can host a source and anyone can run a different reader.

OP is a link to the atproto site because it got a major new revision within the last week

2 million new bsky accounts https://bsky.app/profile/bsky.app/post/3l36zm4e6ee2w

Maybe don't finish! Maybe you did the interesting part and that's enough.

Is there something about finishing that's interesting? Some value intrinsic to you in finishing?

Rust async/await is less nice than Go coroutines. There are things you can't do and weird rules around Rust async code. Every Go chan allows multiple readers and multiple writers, but Rust stdlib and tokio default to single-reader queues.

Bitcoin is still wasting gigawatts every year, but later blockchains are literally a million times more efficient. If the current generations of AI turn out to be useful, maybe someone will figure it out an efficiency optimization.

It's not. It's not about efficiency. Compute-per-watt will certainly be better in other systems. This is about pushing a small system as fast as possible because it's easier to program for a small system. A few problems are 'embarrassingly parallel', but lots have substantial overhead as parallelism increases so running each core as fast as possible is a win for some problems.

going way back: Science Fair in High School landed me summer internships that rolled over into my first job out of college. ("Science and Engineering Fair" project was building robots with microcontrollers) I think it was the proof that I could do that kind of work in a self directed way that made them notice me.

there's a bunch of chemical pathways for turning plant matter into jet fuel

https://afdc.energy.gov/fuels/sustainable_aviation_fuel.html

I have heard many times: Google hires good engineers not to _do_ things, but to _not do things elsewhere_. (even if 80% of Google engineers are 'productively' working on products Google cares about, thats thousands of grade-A engineers spinning their wheels, and given the rate Google discontinues and replaces its own software, 80% is certainly high)

but as always: Are you sure your data doesn't fit in PostgreSQL? You should probably try PostgreSQL first.

"as little as $65 USD/month" for GCP Spanner

vs AWS Free Tier:

"25 GB of data storage ... 2.5 million stream read requests ..."

https://aws.amazon.com/dynamodb/pricing/

So, there's probably somewhere the lines on the graph cross, but Google's headline seems misleading.

author is a flakey writer. Writes an article about "TESCREAL"[1] but doesn't define it but links to another of their articles where it's defined buried in the middle of the 4th paragraph. Author notes they come from this crackpot culture, and they're really still in it and they do a weak job of organizing their thoughts for any audience not in their head.

[1] "transhumanism, Extropianism, singularitarianism, cosmism, Rationalism, Effective Altruism and longtermism" being major themes and key words in the crackpot intelligentsia scene the article is about.

The homepage is all hype, I'm still not sure what Astro really _is_. From clicking through to GitHub I know you install it with npm, so I guess it's written in JavaScript. I guess it's focused on browser side but for 'all in one's I'd expect something server side too?

ActivityPub has been ignoring privacy since at least 2017 https://github.com/w3c/activitypub/issues/225

W3C/ActivityPub has been ignoring privacy since 2017 https://github.com/w3c/activitypub/issues/225

If that was true we would have already all switched to Java

I've been using math and C++ to do redistricting since about 2005 ( https://bdistricting.com/ ). It definitely occurred to me along the way that for the mere cost of a year or two of software engineer time, I could totally build an _evil_ version of my impartial redistricting software that optimized for evil goals. But I obviously didn't want evil ends, and sadly no one even offered to pay me to be evil ;-)