nightpool

Do you have any sources for this? My understanding is that all of the discussion about prediction market wildfires are completely speculative, nobody has actually started a wildfire for a prediction market payout yet.

The chat example here doesn't really seem to work for me—I get 32 FPS consistently no matter what the chat settings are. Maybe a Chrome thing? Or an M3 thing?

That might be true if you're applying for a megacorp like AWS or something but hard to feel like that's a valid argument when you're applying to a company with 2 open roles.

Have you interviewed any of these candidates yet? I doubt that you have 500+ real candidates. My experience in the past 12 months of hiring is that you'll have maybe 10 "qualified on paper" candidates, and 490 AI-generated profiles.

Cool, but, uh, this seems really astroturfed? Why are there two anti-Perplexity articles from independent research firms with identical websites on the front-page of HN right now, submitted by the same person? Feels like they should get deleted

(see https://news.ycombinator.com/item?id=49536201)

I don't even know if I disagree with this post, but this seems really astroturfed. Why are there two anti-Perplexity articles from independent research firms with identical websites on the front-page of HN right now, submitted by the same person? Am I going crazy?

(see https://news.ycombinator.com/item?id=49536375, left a comment there also)

Did you read TFA? They're adding zero data retention back for 5.1

One could argue that, certainly, but I think it's a little disingenuous. Pip makes it impossible to use extra-index-url safely, in the ~only use-case that option has. There is no world where you use extra-index-url and you're not vulnerable in some way or another to dependency confusion.

The CVE looks relevant to me. If you're trying to get a package from a private repo, and someone publishes version 999 of that package on a public repo, then you're executing their code instead of your own. Just because the pip maintainers refuse to fix it doesn't mean it isn't a security vulnerability that could lead to remote code execution and needs to be controlled for in some other way.

How do you define a job "practically possible to do remotely"? It is not practically possible to attend in-person daily meetings remotely. While replacing in-person meetings with Zoom meetings is a choice many companies can and have made, I don't think anyone is saying that they're the exact same thing. If your job requires attending meetings to work on a team (and given that almost every job contains an aspect of teamwork), I don't see a clear way for the government to define when having in-person meetings are "required" for better team coordination. I certainly wouldn't want it to regulate by fiat whether zoom meetings were a required replacement for in-person meetings--that would be the same as the government defining Pi as 3.2 (https://en.wikipedia.org/wiki/Indiana_pi_bill)

The stats are from the Wikipedia page for ECT--I'm not sure about the base rate/risk ratio.

ECT can be very, very unsafe. 70% of patients experience some form of memory loss and at least 30% of patients experience self-reported permanent memory loss. The rate of major adverse cardiac events found in a 2019 meta-analysis was 1 in 39 patients. For the people that it helps, I'm happy they've found relief, but it's a tool with significant costs and I know a fair amount of people who underwent it as children or otherwise were pressured to undergo it and I don't know any of them who have a positive impression of it.

Yes, I was actually a part-time Jenkins admin back in high school and college. I use CircleCI more than Actions most of the time (and used Travis before that), especially for higher cost workflows, but there's no beating Actions in terms of its ease of setup and ecosystem advantage. It's way easier to just set up a simple JS file that runs on every commit in Actions compared to anywhere else.

Would be happy to be proven wrong!

Only if you live on Southampton Island. Otherwise, different countries have different DST rules, so it's always important to clarify which offset you're referring to.

Then you should change your behavior, because probably one day you will talk to someone in Arizona

Getting rid of Actions and Copilot and other secondary services almost halves Github's incident rate: https://i.imgur.com/XPcMIFr.png

I'm a big fan of Github Actions and I think people are often a little too harsh on it, but it's clear that it's sad that it's come at such a high cost to the platform's stability

I get gas at a very busy Costco regularly and even though the lines look bad they're actually very efficient, because Costco 1) has longer gas hoses allowing you to reach both sides of the car 2) has a middle lane allowing cars to enter / exit the gas pump even when there's someone in front of them. So even with ~50 cars waiting the average wait time is only like 5 minutes

Are you comparing cash-only prices to card prices? Some gas stations advertise cash-only prices with 20-30c added on for card processing fees—I've never seen Costco be more expensive then the card price

No, the East Coast observes EDT until November

This used to be true on older patches but it's not at all how top players play these days. See e.g. https://www.youtube.com/watch?v=ejt52UgnhAo for an example of the meta that prevailed for top-level versus play around the end of MK8's lifespan (Yoshi Biddybuggy Roller or Yoshi Teddybuggy Roller depending on your preferred drift style) or https://www.youtube.com/watch?v=vvrCP50Gfj4 for a breakdown of some of the latest time trial world records. At the beginning of the game's lifecycle top speed was the most important stat, but as more tricks were developed and the game's balance was adjusted, having faster and more consistent miniturbo boosts became much more important.

(Mario Kart World, which isn't covered in this article, has a more standard speed vs acceleration trade off but in that game high-max-speed (heavy) combos have even worse problems due to the way the rail grind mechanic was implemented and what seems to be an unforeseen interaction with the coin curve see e.g. https://www.youtube.com/shorts/sPyur6BkiOs)

I mean, the code is unappreciated because the fundamental premise was broken. Locking users out of their paid accounts if they don't respond within some X deadline (no matter how long X is), without a refund, is a completely unacceptable "solution" to technical debt. The fact that you tried your best to give users a good experience doesn't matter if it's serving a fundamentally broken goal.

You're comparing the "score percentage" (e.g. out of the total number of partial points available, how many did the agent achieve) to the "completion percentage" (how many tasks does the model score 100% on). The paper says "Claude Opus 4.8 with maximum thinking and batched tool calls scores best but still completes only 20.6% of tasks at a 54.8% partial score", which is ~the same number that Anthropic reports here (55.7 vs 54.8).

That is—the agent scored 100% on 20% of tasks, but on average it got 54% of the "score" awarded in the exam. One number reflects partial progress, the other one doesn't. The authors of the benchmark prefer you to look at the lower number (because they want to show their benchmark as capturing useful gaps in capabilities and with a lot of room for improvement), the authors of the models want you to look at the higher number (because they want you to think of their models as capable)

It's like 3 lines of standard CSS. If you have an issue with the way your browser is handling it, you should probably take that up with them.

The scrolling behavior on this site worked great for me—it let each example stand on its own and meant I wasn't spending a lot of effort trying to figure out when a given example would end and the next example would begin.

It doesn't look like there's anything custom going on here either, it's all just standard scroll-snap-* CSS properties. if you have an issue with the way your browser is handling the scroll-snap CSS properties, I recommend you take that up with your browser? After all, you should be using a user agent that works for you, not against you.

Is it? When they could be having you running an even more powerful model on a $50k machine they sell by the pallet-load to enterprise consumers? We already see RAM manufacturers abandoning the low-end market in favor of server support. It's not clear to me that Nvidia sees personal GPUs as their best long term investment compared to selling millions of server-farm class machines

Unfortunately, I just had this experience with Google Maps last month—I wanted to go to downtown to buy cheese at Pike Place, but I didn't realize that 6th avenue was closed off for the annual Pride Parade. Traffic was horrible, and Google Maps kept telling me to turn right onto closed off streets until I just had to abandon my trip entirely. Definitely some lack of communication between Seattle city planning and the Google maps team, but even the existing Google Maps systems weren't able to react fast enough to give me any warnings and the time estimation was laughably incorrect.

good luck miniaturizing that! :D

Needs (2020) I guess?

This doesn't say anything about which providers are the majority in Europe, which is what OP's study is about. Your findings (non-US companies are "more likely") and OPs findings (US companies are the majority) are both compatible.

StreetComplete doesn't have satellite overlays, but both Vespucci and EveryDoor support viewing satellite overlay tiles!