recursivecaveat

It's weird how consistent the 'ai art style' is for each medium. You really only need a glance at the right frame in these to know immediately. (https://www.aventos.dev/our-work) Not looking for errors or anything, just style. The ramen one is markedly subtler about it than the others at least. I always assumed this was due to people using 'default' styling by not specifying anything in particular, but these people claim to have a pretty in-depth process that presumably would nudge away from that. When I look at old pictures of dall-e 2 or midjourney v3, the hands are all mangled and stuff, but they're not in the 'ai style'.

A few months after Andrew Wiles presented the proof for Fermat's last theorem, some mistakes were found that subsequently took him over a year to patch over. Imagine being 12 months into trying to put the pieces back together, after all the years of work and announcing your finding publicly.

In medieval Europe they would do "recoinage". They would declare all the old coinage no longer legal tender (also outlaw hoarding and transacting with it or foreign currency). The mint would take all your old coins off your hands and give you back 75% as many new coins.

Extremely similar to iRobot (Roomba). They both practically genericized themselves by inventing and dominating the product category, then just couldn't keep up with the competition. GoPro does feel more self inflicted though. Their drone was a failure, and they burnt a lot of money trying to do some kind of pivot to being a media company.

We are still so slow and have had space travel for so little time, we are almost certainly on the "wait" side of the wait equation: https://en.wikipedia.org/wiki/The_Wait_Equation

This is definitely the guy if anybody wants to peruse for themselves: https://www.amazon.com/stores/author/B0F2RNXR5R The most reviews I could find for any book was 6, most with 0. They cluster around 'dad' topics so I assume he selects the topics himself. I wonder how many he has actually read any substantial amount of. I'm always curious to what degree these people are conscious spammers vs koolaid drinkers. I find it funny that he considers it a ridiculous idea to have his picture actually taken once for an author photo lol.

When the LLM decides to drive-by rephrase me when making a functional change it drives me up the wall haha.

Woah woah woah human, you can't just say there are "far too many" pipes with similar names to abbreviate their labels, the most I'll allow you is a "large number".

I wish they were just regular inputs. A significant portion of them I encounter have some kind of cursor navigation or paste bug. I think this mechanism is the perfect Goldilocks-zone control of simple enough that people rewrite it from scratch, but text input has enough hidden complexity that it's easy to get something wrong.

If a given paragraph gets encoded into twice as many tokens, that means the model gets twice as many matmuls to process it. The amount of compute thrown at the problem is increased (everything else constant), which may improve the quality of the result. This is believed to be one of the reasons that 'thinking' tokens improve quality. For long tasks it will lead to more context compactions though which will harm the quality to some degree as well.

I still don't like having explicit conversions everywhere like in rust. Either you're not thinking too hard about it and the explicit conversions are not really doing anything for you, or you are, meaning you need to be reasoning about it every time and justifying why it can never fail and/or injecting error handling. I would be a much happier rust user if index/length types were are i64 and we relegated unsigned types to serialization almost exclusively. I have other gripes for unsigned types btw, those are just my complaints why explicit casts are not a panacea.

I know the Three Gorges Dam added ~0.06 microseconds on its own, so the effect of reservoirs definitely is noticeable. https://www.jpl.nasa.gov/news/nasa-details-earthquake-effect...

The encyclopedia of integer sequences can be quite useful: https://oeis.org/

Not necessarily. Looking before crossing the street is inversely correlated with getting hit by a truck. Getting trucked is inversely correlated with getting mauled by a lion (most places with wild lions are light on road traffic). Doesn't mean that looking both ways will increase your odds of becoming lion chow though.

I remember from moltbook, all AIs ever talk about is AI haha. I don't know if it's intentional or more that the fact that all the models are presumably system-prompted and post-trained to be cognisant of their AI-ness, so it's already in the context. They probably beat them over the head with the idea that chatbots are friendly and helpful and would never hurt a fly trying to align/safety-ify them, so that could lean into the theme of AIs being trustworthy with autonomy?

I would consider it extremely obscure overall. A large majority of programmers would not be aware of its existence. At the same time there are clearly much less popular languages with articles so it is kindof weird to push to delete. (eg: random scheme implementation w/ no releases in 20 years https://en.wikipedia.org/wiki/SISC) I would say that wikipedia broadly favors programming languages as far as notability. Like most nerd/geek things their footprint skews toward the internet, and people who enjoy geek stuff are more likely to be wikipedia admins than the general population.

The American average is actually ~$35,000 apparently. Though that comes from the wedding industry so they probably undercount small ceremonies and have some incentive to juice the numbers besides. I couldn't really find a good clean table that went back further than the 90s, but qualitatively it sounds like spending spiked in the 80s / early 90s and has mostly stayed with inflation since. Personally, even though I could comfortably afford it, the financial aspect of such an expensive affair would completely suck all the joy and life out of my wedding for me.

https://www.nbcnews.com/id/wbna47681295

People are pretty expensive (they're literally worth a lifetime of free labor). 1 slave would've been like 3 complete years income for an average free white southerner to purchase, plus ongoing expenses obviously. So they basically end up in the hands of upper class people who have a steady need for lots of manual labor. Doesn't mean that everyone else around was not benefiting from the economic surplus, or was not supportive of the institution.

1000 people is what they usually do for political polls. I was taught that a sample size of 1000 is sufficient for pretty much any purpose. At some point what becomes more significant is the differences between the poll-answering population and the general population. For eg if you call people at random on the phone, you're biased towards people who are likely to pick up the phone and talk to a pollster (eg retirees are home more often and not busy). Of course nobody is that naive nowadays, but that doesn't mean systemic error is completely eliminated.

100s of millions have definitely been exposed already. The best defence is probably to be a baby so your risk window is minimal. I haven't been able to pull that off personally, so I follow the other recommended piece of advice which is to keep your credit checks permanently frozen with the agencies and only temporarily thaw it for specific usages.

https://www.upguard.com/breaches/social-insecurity-billions-...

If you have no requirements for accuracy, you can just advance 35% of applicants at random.

If the first 50 people who apply are all bots, why are you reading resumes in order of submission?

Starcraft 2 has the same overall design. If you disconnected and rejoined in an online match you could watch it scrub through the entire game to get back to the synchronization point (not theoretically necessary I guess, but probably avoids a lot of headaches). 10:1 for simulation to real-time feels pretty on-par with what I remember blizzard could achieve.

If you remember that hotel chain Sonder which went bankrupt last year, they had a zero-local-employees model: no front-desk, outsourced maintenance and housekeeping. I think they made the same mistake. Your typical interaction with the hotel receptionist is extremely formulaic. Many other hotels have replaced the sign-in process at least with a machine. That's most prominent in your mind, so its easy to assume that's where most of the value comes from.

Apparently the distribution of patterns (whorls, arches, loops, etc) varies by ethnicity, but not that much. So for an individual set of prints you can't determine anything reliably. It's not really genetic, rather determined mostly by the blood temperature and flow while you're a fetus, like a animal's precise fur pattern.

Given that 20-90% of people don't receive one across various modern countries, I think we have enough regular data. Otherwise like 80% of Germans would be champion athletes if a no-epidural birth really did that for you.

I believe this is the actual audit report here: https://media.api.sf.gov/documents/TTX_Business_Tax_System_P...

Page 27 / exhibit 4 is the most interesting piece of evidence for steering to me.

It seems kind of silly to tout the 5% "not medically necessary" line when 7 times as many were denied for "a reason the insurer never specified". I wouldn't really describe claim denials for reasons like administrative or missing referrals as value neutral either. These are roadblocks controlled by insurers that waste patient and provider time, and reduce access to care.

To lend credibility to the false figures, Fairfull forged bank documents using PDF-editing software and uploaded them to the data room investors used for due diligence.

Wait, when they're doing due diligence, nobody actually checks anything with the bank? I think it might be harder to fake eligibility for a mortgage than $107 million ARR.

I mean, they claimed to be for browser diversity when it was not them on top lol. Underdogs want the race to tighten up, 85% market leaders want to stay out in front.

Magic the gathering has this problem. You have 2 types of cards, and drawing an imbalanced mixture is pretty detrimental. During play you tend to sort them into 2 piles though. Consequently it's a not uncommon sight to see people manually interweaving their cards after a match, then shuffling. Logically, this is either pointless or cheating depending on the quality of that shuffle, but people do it anyways haha.

I feel like there's some hidden complexity there. For any finite flat surface there's a point where not all the spaghetti will fit on it simultaneously. So you have to do O(N) compression steps to find the bundle with the long strand. Locating the strand within the bundle also seems non-trivial if it's big enough. Both are easier to see if you start thinking about scaling to sorting like, square miles of spaghetti at a time.