> I'm not sure how they count 6000 citations, but I guess they are counting everything, including quotes by the vicepresident. Probably 6001 after my comment.
The number appears to be from Google Scholar, which currently reports 6269 citations for the paper
Are there vector DBs with 100B vectors in production which work well? There was a paper which showed that there's 12% loss in accuracy at just 1 mln vectors. Maybe some kind of sharding is another option.
Immediately makes me remember what happened in Cuba back in 2016, where there were some theories about a "sonic" or "directed-energy" weapon being used, could this in any way be related? https://en.wikipedia.org/wiki/Havana_syndrome
> Since 2016, numerous American and Canadian diplomats and secret (intelligence) agents in Cuba, China, and other places in the world have experienced an abrupt onset of unusual clinical symptoms including, tinnitus, vertigo, visual problems, and cognitive difficulties, after they encountered strange sounds or sensations; this has been called “Havana syndrome”
> In a comprehensive report, the National Academies of Sciences, Engineering, and Medicine concluded that the most plausible mechanism for HS is a directed, pulsed radio frequency (RF) energy source
> This also describes transit and describes getting internet service at home.
Well no. Transit means that you use another AS (usually by a larger ISP) to get connectivity to a certain AS. And as for your internet service at home, unless you announce an AS, you are not peering with anyone.
Mildly related: In North Philadelphia there are a few blocks known as the Logan Triangle that were abandoned once it was discovered that the topsoil was not stable.
It’s actually designed for your own gameplay—it scans hours long raw session to find the best highlights and clips them into shorts. It's more about automating the tedious editing process for your own content rather than generating "slop" from scratch.
"-4 makes sense if you consider that the calculator is so damn stupid it ignores every convention every single calculator has made in the past hundred years and instead copies behavior of a dumbass language" isn't exactly the praise you think it is.
It's absolutely marketed as an investment, and solely used and referenced by people saying it is an investment. This is like saying those cannabis paraphernalia shops are marketed as only for tobacco.
Yeah, it's a bummer. As someone who doesn't vote for the right I sincerely feel bad for the sane and respectable people among the right who have watched their parties turn into this and have lost their voice and representation in the media.
Not enough is understood about the replication crisis in the social sciences. Or indeed in the hard sciences. I do wonder whether this is something that AI will rectify.
In a liberal democracy the "will of the masses" is applied indirectly, through the election of representatives, making laws, and then applying those laws and governing in accordance to those laws. To get elected, politicians and aspiring politicians tell electors all sorts of things. Some of them tell electors that their problems have simple solutions, which go against what the intellectual elites (scientists, doctors, engineers, lawyers etc.) recommend or say is doable. Those are what are usually defined populists. Some of them actually believe that "experts" lie for some agenda. Most of them know perfectly well that those simple solutions won't work, but say what they think electors want to hear. Not all politicians/parties act like that, even if it's common to have some populists in all most parties - because populism works.
I'm not sure if you're still reading comments, but in my experience the people who are elated at how well agents perform, versus your experience:
> but the dissonance between what I'm seeing online and what I'm able to achieve is doing my head in.
... could be due to your choice to use Codex, which is OpenAI's coding agent. The stellar reports you're reading online are mostly about Claude Code. (People say so.)
Try your next project in Claude code. I am on the Max plan, which includes Claude code at the flat rate, and have had just as good an experience as the online reports I've read.
Yes, it actually uses a private API: CGSSetWindowBackgroundBlurRadius. There's public APIs, but they do not provide as gentle of a blur as I was hoping to achieve. iTerm actually uses this same API:
https://github.com/gnachman/iTerm2/blob/master/sources/iTerm...
White smoke is water vapor. It's a normal byproduct of hydrocarbon combustion and tends to condense in the exhaust at low loads or immediately after exiting the exhaust, especially in colder temperatures, so you'll see a lot of it in stop and go traffic.
You've cited one example and claimed it to be evidence of a trend. But you haven't shown evidence of a trend. One-offs happen all the time on HN for all kinds of reasons – randomness as much as anything else.
They were saying "the same" in context of how often you have to replace the tires. Now, EV tires are often a slightly different compound (and more expensive) to deal with the higher weight and torque. I don't know how that plays into the particle emissions from those tires though.
If there is a single policy change I could pick for public spending on IT it would be to forbid outsourcing to “contractors” and thinking of software delivery as “projects”
There is also a reading device with a single page in the 1961 Lem novel "Return from the Stars":
> Lem predicts the disappearance of paper books from the society. Lem even describes a reading device very much like a tablet computer that the main character Hal Bregg gets familiar with when he tries to find paper books and newspapers.
Sir Mark Tully, the journalist and broadcaster who has died aged 90, was, for a quarter of a century, the voice of India for BBC radio listeners; in India his name became a synonym for the quintessential foreign correspondent.
> But 50 years from now most of that ICE infrastructure will have disappeared.
I'm guessing it will be already in 20-30 years from now. In 5-10 years from now, no-one will buy an ICE vehicle. Add to those 10 years a lifetime of 10-20 years for the last sold ICE vehicle and you get 20-30 years. So 20-30 years from today there will not be many ICE cars rolling on the streets and most gas stations and other needed infrastructure will be gone as it is not economical to stay in business.
For original research, a researcher is supposed to replicate studies that form the building blocks of their research. For example, if a drug is reported to increase expression of some mRNA in a cell, and your research derives from that, you will start by replicating that step, but it will just be a note in your introduction and not published as a finding on its own.
When a junior researcher, e.g. a grad student, fails to replicate a study, they assume it's technique. If they can't get it after many tries, they just move on, and try some other research approach. If they claim it's because the original study is flawed, people will just assume they don't have the skills to replicate it.
One of the problems is that science doesn't have great collaborative infrastructure. The only way to learn that nobody can reproduce a finding is to go to conferences and have informal chats with people about the paper. Or maybe if you're lucky there's an email list for people in your field where they routinely troubleshoot each other's technique. But most of the time there's just not enough time to waste chasing these things down.
I can't speak to whether people get blackballed. There's a lot of strong personalities in science, but mostly people are direct and efficient. You can ask pretty pointed questions in a session and get pretty direct answers. But accusing someone of fraud is a serious accusation and you probably don't want to get a reputation for being an accuser, FWIW.
It would be trivial to automate this: block HN in the hostfile and only unblock for an hour or two on Saturday evening, for example.