HN user
Total EU defense spending is around $450M USD. The US defense budget, prior to 2027, is about $950M USD. Are you saying the US could have all those social policies for $500M USD?
Whenever somebody makes a benchmark, people complain that the benchmark results are meaningless because they’re gamed. I don’t know why those same people don’t understand that grading on vibes is strictly worse.
Google stupidly positioned their service as if it was a separate console you had to buy games for, which then couldn’t be played anywhere else. The successful streaming services sell you games for non-streaming platforms and then just allow you to stream them as an option.
The purpose of a federal government should be to grant rights, not restrictions.
If the federal government is about granting rights, does that imply the default state is “no rights”? That seems objectively worse.
Good gaijin are welcomed, bad ones need to leave.
This is always the rhetoric in anti-immigration movements. You may find that the definitions of “good” and “bad” vary wildly.
A lot of people are missing the fact that the Steam Frame is Valve's attempt at staking a position in the wide-open and malleable VR space.
It is their third attempt.
which would help reinforce the idea that what we sacrifice as the price of scientific knowledge, is absolute knowledge.
I don’t think it is possible to have absolute knowledge of anything. Scientific knowledge is the best (only) thing we have.
I enjoyed it!
Admittedly, I went in with extremely low expectations. It was fun, though, and I liked the visuals and the music. The plot was … something.
The link says:
Some teams in the Google Cloud org just laid off all UX researchers below L6
That’s not all UX researchers below L6 in the entire company. It doesn’t even sound like it’s all UX researchers below L6 in Google Cloud.
The notion of “PhD-level research” is too vague to be useful anyways. Is it equivalent to a preprint, a poster, a workshop paper, a conference paper, a journal submission, or a book? Is it expected to pass peer review in a prestigious venue, a mid-tier venue, or simply any venue at all?
There’s wildly varying levels of quality among these options, even though they could all reasonably be called “PhD-level research.”
buying an EV does not actually reduce emissions like magic, unless the owner drives that car for a looooong time. Like 10-15+ years.
I find this timeframe surprising. I did some quick searches and there are models like GREET that suggest the break-even point is much sooner than that in the US. It is difficult to know for certain, of course, as there are many variables.
Regardless, it is of course better to incentivize long-term ownership as well. I think of HOV access as similar to a tax deduction on purchase. It’s a cheap way to provide a carrot for initial EV adoption.
It makes sense if you view the HOV lane primarily as a way to reduce emissions, not traffic. This is also why e.g. single-rider motorcycles are often allowed to use HOV lanes as well.
Google spends something around 30 billion dollars a year to be the default search engine across many platforms. You can spend the same amount and tomorrow your search engine will have 88.9% of searches.
It is a widely held belief that users don’t change the defaults, and I’m not asserting it’s wrong in general, but why doesn’t it apply to web browsers?
As an (unhappy) Windows user, I note that Microsoft pushes Edge aggressively, with each major Windows update “helpfully” offering to “optimize my computer” by making it the default browser again. However, Edge market share is only ~12% on desktop [0], despite the fact it is significantly more work to install Chrome than it is to change a mere default setting. Is that just because desktop users are more willing to jump through hoops?
[0] https://gs.statcounter.com/browser-market-share/desktop/worl...
The rejection message doesn’t seem to be accurate. I tried “happy person” as a prompt in AI Studio and it generated a happy human without any complaints.
It’s possible that they relaxed the safety filtering to allow humans but forgot to update the error message.
If you look at the side-by-side [1], it looks pretty plagiarized to me. They changed the words a bit but the overall plot and themes are essentially identical.
Trump won the majority vote in an election that brought out a huge amount of voters compared to previous elections, and Republicans won every other government body.
For the curious, based on [1], turnout in 2024 was 63.1% of the voting eligible population, compared to 65.3% in 2020, 59.2% in 2016, and 58.0% in 2012.
[1] https://www.presidency.ucsb.edu/statistics/data/voter-turnou...
Their methodology seems reasonable to me.
To clarify, they look at the probability a model will produce a verbatim 50-token excerpt given the preceding 50 tokens. They evaluate this for all sequences in the book using a sliding window of 10 characters (NB: not tokens). Sequences from Harry Potter have substantially higher probabilities of being reproduced than sequences from less well-known books.
Whether this is "recall" is, of course, one of those tricky semantic arguments we have yet to settle when it comes to LLMs.
I learned a while back that Google Maps was moved from maps.google.com to google.com/maps so that when people gave location permission to Maps, Google Search could also use that permission.
This does not appear to be the case, at least on iOS Safari. I went to Google Maps, gave it permission, then went to Google Search and searched for “delivery near me.” It again asked me for permission.
In the spirit of good science and as a happy taxpayer for the cause of these organizations, we should still be open to their scrutiny. A simple question we should ask, after all we're good scientists, is whether these groups are at their appropriate funding-to-success level or not, particularly in an era of a spiraling debt crisis.
I agree, in principle. However, this is a trap.
Here’s a playbook:
1. Declare, loudly, that a problem exists. The problem doesn’t have to be real, but it’s better if it is.
2. Announce, even more loudly, that you are going to address the problem in a way that’s suspiciously self serving.
3. Implement your preferred solution as rapidly as possible. The “solution” can be as flawed as you like. It may or may not actually fix the original problem; that part is unimportant.
4. When people react to your implementation, they sort themselves into three buckets: supporters (partisan or otherwise), detractors (partisan or otherwise), and “reasonable people” who “see both sides.”
5. While the “reasonable people” are still debating whether it was a good idea to cure the patient’s brain tumor by decapitation, move on to the next “problem” that needs to be “fixed.”
My hypothesis: it is about control.
Education level is correlated with voting preference [0].
US universities are funded in large part by scientific research grants. Cutting funding directly damages these institutions, and it gives the current administration leverage they can use to influence university political policies through either overt ultimatums (cut DEI or we don’t give you the money) or indirectly by funding professors with pro-administration viewpoints while defunding professors with anti-administration viewpoints.
[0] https://www.statista.com/statistics/1535279/presidential-ele...
I suspect you have to choose the right numbers but not the right order. That makes the numbers work out.
Medium is a publishing/hosting platform for authors. This Medium account is (or purports to be) the official account of the Stanford alumni magazine.
it'd hardly be prudent for the world's largest software company NOT to have it's own SOTA AI models.
If I recall correctly, Microsoft’s agreement with OpenAI gives them full license to all of OpenAI’s IP, model weights and all. So they already have a SOTA model without doing anything.
I suppose it’s still worth it to them to build out the experience and infrastructure needed to push the envelope on their own, but the agreement with OpenAI doesn’t expire until OpenAI creates AGI, so they have plenty of time.
You can submit a blank ballot. Since the ballot is secret, compulsory voting merely verifies that you got a ballot and submitted it, not that you filled it out correctly/completely.
It is difficult to investigate since the original statement of “$21 million for ‘voter turnout’ in India” doesn’t give any reference to the actual grant.
Some sources suggest that the referenced grant is [0], as this is allegedly the only grant for $21 million by USAID that matches the description. It is for Bangladesh, not India, and the money appears to have gone to various pro-democracy advocacy groups. While I suppose you could argue about whether the US should “spread democracy” around the world, I don’t see that you can say this is fraud rather than merely government spending one disagrees with.
I don’t think it’s a good strategy in this case.
Some of the people you fire will not come back, even if you try to rehire them immediately. Some of the firings will not result in problems in the near term, but will cause problems later on. Some of the firings will cause problems that only occur under certain conditions, like catastrophic events and natural disasters.
Furthermore, you cause real harm to people who depend on these services in the interim period where you’re figuring things out. In some cases that harm cannot be undone.
The stated policy objectives to reduce waste could be achieved with much less disruption and cruelty simply by doing them more slowly and thoughtfully. There is no need to rush everything through in six weeks.
I mean, Altman also wrote this (https://x.com/sama/status/1882234406662000833):
watching @potus more carefully recently has really changed my perspective on him (i wish i had done more of my own thinking and definitely fell in the npc trap).
i'm not going to agree with him on everything, but i think he will be incredible for the country in many ways!
Assuming the numbers I linked above are correct or at least in the ballpark:
At the federal level, pretty much anything else, since it’s already not really spending any money on libraries, relatively speaking.
At the state/local level, it’s harder to say since there are many more administrative units involved, each with their own budget and operating model. (This also makes it hard to optimize in general, since you’d have to apply the optimizations independently across many polities.)
If I use my city as an example, library funding is 5% of the city budget. The state provides no funding to any library, so this is the entire amount the libraries get. It’s the second smallest spend by category as the city tracks such things —- though there is an “other” category that represents 10% of the budget. The bulk of the spending is on police (23%), fire (17%), and public works (11%). Obviously these are also critical services (basically everything in the city budget is!), so it’s not easy to do cost cutting there either, but there’s proportionally more room for improvement.
It's not even hyper optimizing, just basic optimizing would be nice.
It feels a little bit like hyper optimizing. According to [0], the US spent $14.6B on libraries in 2020. The vast majority of that was from local municipalities, with state funding accounting for ~$1B and federal funding only $80M (disclaimer: I have not put any effort into verifying these numbers). It seems like our time would be better spent optimizing larger, more expensive programs before really pushing to make improvements here.
X is the only corporate social media platform that offers this. You don't like the algorithm? Don't use it, and use X just like everybody else for the first 10 years of its life.
Speaking as a non-X user, doesn’t YouTube offer essentially the same thing with the “subscriptions” tab?