HN user
Full disclosure: I work at Etched.
Weights are not burnt into silicon per-model. They're in SRAM/HBM. There's some more info on the website (etched.com) and we'll be sharing more details about model benchmarks this summer.
As far as I can tell from these articles, driving into a flood has happened twice to Waymos, once in Texas and once in Atlanta? It does seem like it's pretty uncommon.
It used to stand for "[R]adeon [O]pen [C]o[m]pute", but since it's not affiliated with the Open Compute Project, they dropped the meaning of it a little while ago, and now it doesn't stand for anything.
Looks like the shelves were custom-built for those machines. I wonder what the monitors were hooked up to, or if they were just spares.
My first thought was that this was built someone who clearly cared about the system they were running.
This isn't new.
"The Productivity Paradox" is what they called it when people were skeptical that computer would end up finding a place in the office. There are articles from the 90s complaining about how much people are spending on buying computers for no real impact on productivity https://dl.acm.org/doi/10.1145/163298.163309
My favorite example of this was the Pew Research study: https://www.pewresearch.org/short-reads/2024/03/05/online-op...
They found that ~15% of US adults under 30 claim to have been trained to operate a nuclear submarine.
Procedural generation can use a fixed seed, it's not too uncommon. For instance, Elder Scrolls 2: Daggerfall's map was procedurally generated, but is the same for every player.
Computer hardware isn't trying to be currency. Bitcoin was supposed to be, but hardly anyone who uses Bitcoin these days is using it to buy things--it's used as a store of value or a speculative asset, not a means of transaction.
The other thing about Bitcoin is that it's deflationary, which leads to people holding the currency rather than spending it, as predicted by Econ 101.
The satellite view shows this off much better than Wikipedia's ground-level picture. It Really is just a long band of holes dug into the side of a mountain.
https://www.google.com/maps/place/13%C2%B042'20.0%22S+75%C2%...
The big problem with MUMPS is that as the "Massachusetts General Hospital Utility Multi-Programming System", it does not work well for development in other states. There's been some experimentation with using it in Wisconson, but a W is not an M.
Only going off when you want it to actually happens to also be requirements #1, #2, and #3 for grenades
I asked it to create the kind of storybook my toddler would have asked for ("create a storybook about a music truck and an ice cream truck and a mailman and a carwash", inspired by his request for a story last night), and the results were certainly... interesting.
Obviously Gemini doesn't know that "music truck" is another name for "ice cream truck", but more concerningly, the illustrations it made for the trucks were this kind of eldritch amalgamation of Cars-movie style cars and people driving cars. The story was just OK, I don't think it would have kept my toddler's attention for the whole ten pages. Plus, the mailman is barely involved.
It has some ancient history as a morse code distress signal: https://regulatorylibrary.caa.co.uk/923-2012/Content/Regs/03...
And it shows up in some old BSD code: https://www.snellman.net/blog/archive/2017-04-17-xxx-fixme/
But... I think repeated letters are just easier to type than any other string, and since X looks like the classic "marks the spot" logo, it's what people jump to.
I've recently started using Github code reviews for a lot of C++, and one thing that I wish it would do is show the header (.h) files before the implementation (.cc) files.
Small PRs help, but I often end up just opening a handful of windows to have everything open at once.
For better or worse, Google gets more bad press when their models get things wrong compared to smaller AI labs.
People love to dunk on hacky looking gamedev code. Some of it is pretty ugly, e.g, VVVVVV's famous gigantic switch statement, but if it works and makes a fun game, that's the actually important part.
The parting of the waters in the USA is such an interesting landmark for just how uninteresting it actually is in person. Looks like any other split in a forest river, marked by a sign on a tree that you can only see after a long wilderness hike.
https://en.wikipedia.org/wiki/Parting_of_the_Waters
https://www.anotherlongwalk.com/2022/07/day-119-parting-of-w...
Yeah, YouTube's UI lets you set where the ads go. The creator tools let you set how many, and where midroll ads will play. However, most creators just click the "insert automatically" button.
Non-centralization is enough for me (I love archive.org, but always worry about them flying too close to the sun one day and getting shut down), but this does also have a different UX, and has a more direct focus on videogame history, which should help them surface things in more interesting ways than a general purpose archive.
Personally, when I send an email, I feel less time pressure to respond, so I more carefully craft my responses. The metadata is similar enough, but the actual data in email/forums is usually better.
Too much iodine can lead to thyroid problems, and this has been a problem in some subgroups in Japan: https://academic.oup.com/jcem/article/107/6/e2634/6516999
The most annoying bots are the ones that mindlessly slam sites over and over, without doing any filtering. Having these kinds of tarpits out in the wild forcing people to be better behaved with their crawling bots is a feature, not a bug.
They presented at HotChips 2024: https://www.youtube.com/watch?v=4_R0S6piLA0&t=2114s
There, one of the things they mentioned was that they were hoping that RISC-V in general, and their cores in particular, becomes a platform for academic research. That way, cutting-edge things are shared in the open and the industry benefits globally.
The optimistic view of this is that it's hoping for a better future for all. The cynical view is that open source is much harder to sanction/export control. The truth is probably somewhere in the middle.
Pew Research found that 24% of Hispanic adults are licensed to operate a SSGN submarine: https://www.pewresearch.org/short-reads/2024/03/05/online-op...
The margins of error on online opt-in polls are massive.
For what it's worth, conviction rates are not in the high 90%s.
The DOJ has a high conviction rate, at 93%, but Federal cases are the minority of cases in the USA. Most cases are state-level, and the conviction rates vary pretty significantly (E.g., California has a conviction rate of ~65% for property crime: Table 6 of https://www.courts.ca.gov/documents/lr-2019-JC-disposition-o... )
It's kind of interesting when you look back a year or so to when ChatGPT came out, and the prevailing HN sentiment seemed to be that ChatGPT would be a Google Search killer, even back when it was even more prone to hallucinations and stating wild nonsense as fact.
The N64 has a GPU, which is not the same as modern GPUs, and you need that GPU to render graphics.
Skimming the Github repo, looks like this uses rt64 under the hood for the raytracing support ( https://github.com/rt64/rt64 ), and that uses the ubershader technique to emulate the GPU. This excellent article talks about the what, why, and insanity that are ubershaders: https://dolphin-emu.org/blog/2017/07/30/ubershaders/
To be fair, N64 is not the PS2 or Gameboy--there are only 388 games that were released for it. Going through all of the games that were released and shepherding them through this process is feasible.
I have a Roomba, it works pretty well for doing 'maintenance level' vacuuming--keeping the level of cat hair to a manageable level, etc. For the most part, robot vacuums have succeeded in becoming boring, which is what I really want in an appliance.