I have friends at companies that used to do BS jobs. Then, AI came in and took over the BS, and now they're doing real, meaningful work. One friend was told her team (3) would be eliminated once AI took over. Now that the AI automated the teams' job, they do proper finance work, and they've expanded the team to 5 people. Turns out the limiting rate was a bottleneck in the paperwork that the AI eliminated, meaning they can do proper forecasting analysis more effectively. Each additional person increases the set of product markets they can effectively forecast for. With effective forecasts they can time the market better, and they've already saved the company significant money — way more than their salaries, combined.
> The big difference is that with a debit card, it's your money that is hit by fraud.
I kind of disagree. I think what's happened is the bank would prefer you to believe that. Imagine I kept my money at the bank, and deposited $10000 with the teller. Immediately afterward a robber follows in and steals that $10000 from the teller. Does the bank say "oh no Mr. TheChao! A robber stole your $10000!". I mean, no? The bank got robbed. Just because the bank's digital security is more tied one-to-one to dollars and its easier for a robber to steal from "my till" doesn't mean it was me who was robbed. It's the bank's job to stop that.
But that's not what happened when someone steals your debit card number. Once the teller enters the deposit and you walk away from the window, the money in your account is yours, the cash the robbers stole is the banks.
When someone steals your debit card number and uses it, it's your money that is debited from your account, not the banks, and debit card transactions have much different rules than credit card transactions where you're generally not held responsible at all for fraudulent transactions)
But with debit card transactions, you can be responsible for up to $50 of fraudulent transactions reported within 2 days, and $500 for those reported within 60 days. And the bank is not required to return the money while they investigate the fraud.
Though many banks waive those liabilities and treat debit card disputes the same as credit card disputes (but still may not return your money while they investigate, so you could be bouncing checks when someone makes a $1000 fraudulent transaction if you can't float the $1000 yourself while they investigate)
I refused to allow the bank to issue me a debit/ATM card against my primary checking account, fortunately they are one of the few banks that still offer a dedicated ATM card so even if I lose the card, it's useless without the PIN.
Eh when you deposit money with a teller, it gets added to your account, which is in physical terms probably just a big storage and a database who has how much.
But in the case of debit card, the card ties the money to your account. It is actually that.
It's as if someone would steal from a personal safe at the bank.
Except that you have strong statutory rights and commercial agreements between your bank and the card networks making it very likely that you’ll be made whole.
Eventually you'll probably be made whole (unless the bank isn't one of the ones that waives your $50/$500 liability). But usually they won't return your money until after they resolve the dispute so your account is still missing the disputed money which could make you bounce checks or be unable to use your debit card until the dispute is resolved.
Which is why the credit card is still the better option. Especially given that the max liability on a credit card is always $50 if caught within the first 60 days, while the max liability on a debit card is $50 only if caught within the first two days, then up to $500 if reported after up to 60 days.
They may post a provisional credit when I report the fraud on the debit card or they can wait up to 10 days to do so. With a credit card, no money has left my account and I get the final say on whether I want to part ways with my real money. If the bank really wants to fuck me over by saying it's not fraud, I get to make it as unprofitable as possible for them, which includes forcing them to sue me if they really want the money.
I will take the ding to my credit report and a lawsuit over actual money taken directly out of my account any day.
> while the max liability on a debit card is $50 only if caught within the first two days
Within two days of learning of the fraudulent use or the loss of an "access device" (i.e. a card). Otherwise you have 60 days from the statement date as well.
> I will take the ding to my credit report and a lawsuit over actual money taken directly out of my account any day.
That's definitely true for many people, but probably not for many others, e.g. anyone wanting to buy a house or even rent an apartment.
I hope I'm remembering this right: a mathematician claims to have a proof for the ABC conjecture, but can't conceive any other mathematician it's right — it's "too weird", so the proof is rejected?
The consensus is that proof is in fact incorrect. People tried really hard (like putting in a year of effort) and most converged to the same place, that proof of 3.12 is incorrect or has a gap. Peter Scholze (who won Fields Medal) and Jakob Stix did a writeup. People seem to think Shinichi Mochizuki correctly reduced ABC conjecture to 3.12, but didn't prove 3.12, and also are doubtful about the whole program because 3.12 doesn't seem any easier than ABC conjecture while complicating everything.
Not quite?
It is more that
1) someone has gone through it, identified a step he thinks isn’t a valid step, and the author hasn’t been willing to work with that person
2) most consider the proof, due to its length combined with those doubts as to its validity, not worth their time and effort to work through and understand (because it would take a lot of time, and they have jobs to do, doing research and teaching, etc.)
Some of the best mathematicians in the world tried to study his work, found flaws he did not address, and somehow there’s someone every week suggesting there’s a conspiracy against this guy. It’s really baffling. AI will probably help him move on by lean verifying his proof is wrong…
By this point, he is very much nutso enough that a Lean certified counterexample to his theories would not dissuade him. His response would be either that the formalization is incorrect (with no coherent insights on how to fix it), or worse, Lean itself is a tool of Western imperialism and incapable of properly explicating his ideas. He has, in the past, ranted against such things as monotheism and English grammar as being the reason for his theories' lack of popularity.
IIRC he has expressed support in the past for attempts to formalize IUT in Lean, but we'll see where that really goes, because he's absolutely not clearheaded enough to lead such a project himself.
Yeah, models aren't good at code any more than they are good at people. They turn it out very fast, but that isn't actually useful when the code needs to be carefully reviewed to find the land mines and questionable design decisions.
Housing values skyrocketed due to low interest rates. The prices are sticky and won't correct just because mortgage rates have gone up. That means no one can move. What we need is a large number of small houses (or owner owned multitenants) and a population willing to move into them.
Gasoline has 10x the energy (MJ/Kg) than TNT, but it's not a "bomb". Being a "bomb" is about energy release rate. A battery is still controlled by its chemistry.
Liquid gasoline does not have anywhere near the energy of TNT, or even a battery for that matter. It contains basically zero releasable energy. It needs oxygen or another oxidizer to react with to actually release any energy. TNT and batteries' energy density calculations include the oxidizer and the oxidizer is in close proximity to the fuel (molecularly so in the case of TNT). If you 10x the "energy content" of gasoline it's still rate limited by access to oxygen. If you 10x the energy density of a battery (the type with the oxidizer contained within the battery, not a fuel cell or metal air battery) you've got 10x the energy ready to be released quickly if oxidizer and fuel mix in unfortunate ways
The previous point is fair though. What is the energy density of gasoline in a stoichiometric mixture with its oxidizer and not already detonated? That's where TNT beats it by a long shot. The mixture with any chance of being stored would be some relatively low pressure gas.
Edit: Or, a fair closed-system comparison to a battery would need to include a liquid oxygen tank or similar so you have the two components stored but can control the delivery...
In the context of the conversation, it doesn't actually refute the point though, which is that explosive danger is contingent on a much broader set of relations than simple energy density. Gasoline just demonstrates the disconnect. Tacking on additional conditions to make high-density batteries dangerous is exactly what the engineering goal should be.
As a stupid example, if we have a high density battery technology that never explodes unless, say, under magnetic forces only seen being produced by magnetars, that's a pretty safe battery.
Gasoline + LOX beats TNT on energy density per unit mass and per unit volume, no?
There's a reason that kerosene+h2o2 is a classic rocket fuel, and TNT is not.
It's usually kerosene + O2, not H2O2. But H2 + O2 is another classic liquid rocket fuel that has higher specific energy (energy per mass), but lower energy density (energy per volume) than kerosene, and is used mostly as a 2nd stage fuel (where mass is most important). Its lower density and the need to keep it very cool mean heavier tanks, and it's also hard to work with due to being such a small molecule that likes to leak, which causes development costs to be higher.
> The previous point is fair though. What is the energy density of gasoline in a stoichiometric mixture with its oxidizer and not already detonated? That's where TNT beats it by a long shot. The mixture with any chance of being stored would be some relatively low pressure gas.
eeeh that still doesn't matter really. Gasoline is safe(r) because it needs conditions to explode (spreading into area then ignition), instead of just detonator. Diesel even more so
A damaged gasoline tank that's near empty might explode but more likely it will just leak and start burning, and probably at that moment be still extinguishable. Explosives, well, you won't have time to blink.
Batteries ? Zero chance to extinguish, gonna let it burn thru. THAT is the problem
To stoichiometrically burn 1 gram of gasoline requires 3.5 grams of oxygen, so an energy density of 9.5 MJ/g, still more than double TNT. The state of the oxygen doesn't impact its mass.
That's a huge assumption that won't always be the case though. A gasoline tank with a small hole in it is a lot safer than a liquid-electrolyte battery with a hole in it.
I can remember several situations where “a little bit of gasoline” caused a lot of problems. It’s a relative term. You need a lot of air to make it really shine, but “a little bit of gas” goes a long way lol.
Energy capacity vs. energy release rate, or energy vs. power, tends to follow an inverse relationship, at least within similar domains.
The fuels which afford the highest rates of storage tend to release it more slowly. Chemical explosives typically have about 1/10 the energy storage capacity (by mass) of fuels, as several people have pointed out.
Similar relationships exist for other forms of energy storage. For electricity, fuel cells, chemical cells ("batteries"), flywheel / reaction storage, and capacitors tend to offer a trade-off between total storage capacity and reaction time. Capacitors can react at intra-phase rates (that is, within a single 60Hz power cycle), whilst batteries and flywheels can take longer to come online.
Gasoline, when in liquid form, is not mixed with oxygen and burns, releases energy slowly. In comparison with TNT which is both the fuel and oxidizing agent releases energy very fast, it has detonation velocity of 6,900 m/s.
There is a way how to use the energy density of gasoline for explosive purposes, Thermobaric weapons.
"A fuel–air explosive (FAE) device consists of a container of fuel and two separate explosive charges. After the munition is dropped or fired, the first explosive charge bursts open the container at a predetermined height and disperses the fuel in a cloud that mixes with atmospheric oxygen (the size of the cloud varies with the size of the munition). The cloud of fuel flows around objects and into structures. The second charge then detonates the cloud and creates a massive blast wave."
I'm still surprised we haven't figured out a bioreactor you can feed your scrap food and turn it into energy, methane combustion notwithstanding. Like imagine feeding a roomba some sugar and it's set for a few days lol.
But only if said Roomba is unable to find enough sugar on the ground for a few days. Otherwise, it's completely self-sufficient, you just have to leave enough crumbs on the floor.
One of the problems with gasoline is its explosive nature, which the fuel storage and delivery system of a car attempts to mitigate. It has a better profile for burning off than a battery though since it will spill out and burn away from the vehicle.
Gasoline isn't that explosive. You can shoot up a gas tank and it will just burn. What you see on TV is movie magic. Gasoline vapor can explode, but you need the right air to fuel ratio and a spark.
Requires a good mix though. The parent is still correct.
Mythbusters spent a long time trying to get cars to blow up in a dramatic hollywood fashion in various episodes and determined that under most conditions it wouldn't happen.
They did get it to happen, they just had to try really hard.
And was still not really possible without some really unlikely conditions. Not that car fires aren't dangerous, regardless of whether the vehicle is ICE or otherwise, but the fireball thing doesn't really happen.
"Fast X: why cars don’t really explode when they crash"
"Petrol and diesel can only explode when under pressure and mixed with air and in the case of petrol, have a small amount of energy added in the form of a spark or a flame. Engines pressurise the fuel/air mixture in the cylinder and so produce small, confined explosions which turn a crankshaft and drive the wheels."
"When cars are involved in collisions, fuel lines are often torn and petrol leaks out onto a hot engine. Liquid petrol can catch fire in the presence of air. But it can’t explode because it’s not under pressure and is in the liquid phase rather than a vapour."
My teen years would beg to differ, when some neighborhood friends and I decided to experiment with gas and bleach. The neighborhood residents, fire department and local police also had an opinion, too. (We all got into BIG trouble. Definitely one of the dumber things I did as a teen.)
My high-school crowd stopped just short of building a multistage FAE device. I left early for college and everybody else was short of cash, and that was the end of our neighborhood R&D program. All pre-9/11, needless to say. It's a lot less fun when the possible punishment goes from detention at lunch to waterboarding in Gitmo.
We used to take a half-full can of gas, then get a ping pong ball, drill a hole in it, fill it with bleach, then put tape over the hole. Drop the ping pong ball in the gas can, screw the lid on really tight, then wait. It used to take about 10 minutes for the gas to eat its way into the ping pong ball, but when it did, and the bleach mixed with the gas... boom! The scary part is only worked, maybe, half the time. So there would be cans of gas/bleach just sitting in a field, waiting to explode or release really harmful chems into the air. We were kids, young and stupid.
No it can't. To combust it needs ~ 500F. If you left it out uncovered in the open and allowed the it to vaporize with a spark nearby, it might explode, but thats not spontaneous combustion. Gasoline sitting in a sealed container wont spontaneously combust.
Oh, it was most definitely gasoline. As long as there is no vapour hanging above the container, it works just fine... as the neighbour post says, the first 50 times. Then one day the old timer blows up himself and the shop
Gasoline doesn't do that. Perhaps you're thinking of the following?
What it does do is slowly polymerize, becoming useless as an ICE fuel in time, typically in a year or two. This is why backup generators should run on propane, which has no degradation mechanism.
> The attachment contained 3,000 English-language titles organized by ISBN number, including books such as Distinct Element Modelling in Geomechanics by K.R. Saxena (1999); Barrett's Traditional Fairy Tales (2021), an academic study of Irish folklore; and Laser Shock Peening of Advanced Ceramics by Pratik Shukla (2018).
How is a book from 2021 considered rare in this context? There's almost certainly a digital copy of it in existence prior to Anthropic purchasing a print edition.
Niche text. It's not impossible that there was only ever under a thousand of them printed and released into circulation.
A digital copy would exist somewhere, of course. But for us, that only matters if we can buy or download it. And for AI companies, that only matters if they can get a digital copy DRM-free and licensed permissively enough.
This argument doesn't make any sense. All manner of AI companies just ingest whatever random text they can find on the internet to train their data, including copyrighted publications. Why would DRM on a digital copy of a book matter?
> Why would DRM on a digital copy of a book matter?
Because DRM is just a way to make "breaking copyright" more practically cumbersome. What's easier, breaking digital DRM for each and every E-book you find, or just establishing a single pipeline for scanning physical books?
Again, what's more technically feasible? Setting up a DRM-cracking LLM and manually verifying that it's actually succeeded for each and every DRM scheme you encounter, or just throwing the books in a regular old office scanner and being sure it works without even checking...
I would be delighted if AI companies got together and thoroughly dismantled DMCA.
That atrocity of a law was a blight upon digital freedom since the day it came to exist. DRM should never have been given any legal protection - and I would push for numerous forms of DRM to be outlawed instead.
so if I base64 encode my blog, have some Javascript that 'validates' an authorized viewer and then decodes the base64 into HTML which is added to the DOM does that constitute DRM ?
The DMCA doesn't care exactly how the authorization works, it only cares that it's needed.
Quote from Wikipedia[0] of DMCA section 103:
> No person shall circumvent a technological measure that effectively controls access to a work protected under this title.
> "circumvent a technological measure" means to descramble a scrambled work, to decrypt an encrypted work, or otherwise to avoid, bypass, remove, deactivate, or impair a technological measure, without the authority of the copyright owner; and
> a technological measure "effectively controls access to a work" if the measure, in the ordinary course of its operation, requires the application of information, or a process or a treatment, with the authority of the copyright owner, to gain access to the work.
Rare or not, destroying books is in itself a morally repugnant act. I don't know man, it's not so long ago we used to view nazi book burnings as an archetype of evil. Today companies are offering book-burning-as-a-service and it hardly causes a stir.
Is it destruction or format shift? Technically the books still exist; just that they're now digital. There's no content loss (what I imagine comprises a "book") unless any digitization is then deleted.
You only need to clip triangles is you're worried about attribute interpolation for very large triangles. There's two ways to handle this: (1) discard (fast but not a great user experience); or, (2) primitive synthesis. Just frustum clipping is enabled by point picking in the local tile. Primitive synthesis requires some FP kung fu; but, is easiest done in barycentric space against a reverse transformed clipping rectangle. This lets you carefully control clip rounding error using either doubles or (better) fixed point. Abrash likes to use integer fixed point, but that is historical — modern fixed point can be handled with careful control of the fp unit in the mantissa. The major issue is regenerating the Z and the 1/Z values for the new vertices of the synthesized primitives. Everything else should flow down the pipe naturally, assuming a deferred attribute synthesis rasterizer.
There are examples in the open source version of my rasterizer: OpenSWR.org.
I'm clearly super old-school when it comes to rasterisation, and a lot of this has flown above my head. I have a ton of questions; I hope you can answer them.
> (1) discard (fast but not a great user experience)
What are we discarding here, and why is it fast but not a great user experience?
> (2) primitive synthesis
I assume this is retriangulating clipped triangles that are now no longer triangles?
> reverse transformed clipping rectangle
Which spaces does this reverse transformation map from and to? I assume the clipping rectangle here is the triangle's AABB in raster space (or as you say, barycentric space).
> integer fixed point, but that is historical — modern fixed point can be handled with careful control of the fp unit in the mantissa
So we are no longer doing 16.16 fixed point, but tweaking the FP representation itself?
> The major issue is regenerating the Z and the 1/Z values for the new vertices
Why is this a major issue?
> deferred attribute synthesis rasterizer
I assume this means attributes are perspective-correct interpolated in raster space.
Sorry I was so terse! Let's ignore guard band, for now. Our rasterization surface is a rectangle. We don't need to "geometrically clip" a triangle to the rectangle's surface. Instead, we just walk the surface of the rectangle and ask two questions: (1) does the triangle capture this (sub)sample; and, (2) what's the interpolated value of the attributes at this (sub)sample. In practice, for software rasterizers, we're working on tiny subrectangles (the tile), e.g., 4x4 or 2x8, whatever. So, we can be a little "inefficient" with our walking with respect to the edge testing for interiority. (This is roughly how HW works at the 2x2 level, as well.) Because the (sub)sample is "pulling" the attribute, we don't need to geometrically "clip" the triangle to the tile: the tile's (x,y) subsamples do that "for free".
On the flip side, if a triangle is 'really big' we need a guard band to either reject or subdivide triangles. The first one is fairly cheap -- we're throwing away the triangle! -- the second one is a better user experience, but requires synthesizing primitives. (The worst case is that a single triangle becomes 5 triangles, I think.) Each of those triangles needs its Z and 1/Z calculated in fixed precision. The precision of that fixed precision (though) can be clamped to the local tile; so, even though the global precision might need to be 25.25 (or whatever), the tile-local precision is only 4.9 (or whatever), with an intermediate 24.24 that can be handled with a float-float patch-up. The computation should all occur in the triangle's barycentric space: that means you need the inverted barycentric mapping to invert the guard band into the triangle's barycentric space. You do that because it lets you control the fixed point calculations better. (You can leave off the inverted determinant multiplication until the last moment.)
When I say "deferred attribute synthesis" I mean that we don't calculate attributes in the vertex shader. Instead, we calculate the barycentric, Z, and 1/Z values and pass those along. When we fire up the tile walker for the triangle (in general we only need 1-3 tiles), we calculate the attributes "on the fly, as they're used" and then let the compiler do CSE to fold down the replicated constructions.
My own knowledge of GPU rasterization may be dated, but IIRC GPUs tend to rely on guard band clipping up to a guard band threshold before using geometric clipping. The guard band clipping involves rejecting 2D coarse rasterization blocks that are fully outside of the scissor rect. This is just a quick rectangle check, but the tradeoff is that a larger guard band means more GPU time lost in over-rasterization and potential higher precision requirements for rasterization values (which could be fixed point). Beyond the guard band, the triangles are clipped in floating point against the frustum clip planes.
reply