It is my understand that there isnt a lot of innovation here. Rather the (broad) field of mathematics got mined for solutions that already existed within the corpus.
It implies "We didn't know we already knew how to solve these problems". How is that a good look for mathematics as a field?
None of them have pointed to any of the solutions and said "this is real innovation" - because it's all been mining existing ideas and applying them elsewhere.
Imagine you spend years working on a solution to one of these problems. Only to realize after you publish that more than half your answer already existed in another branch of mathematics and you spent a bunch of time re-inventing it.
Yes the first big solve by OAI was "stolen" and that is tragic, and pretty bad. But it looks like this new batch, the one people are crying about, has none of that stink.
> the only other "language" that I ever felt this way with was probably PHP
I made a lot of money on PHP and more on Golang.
PHP had warts (variable order) but much like go it came with just about everything you wanted.
It isnt just the functionality - it's the fact that both of them had amazing documentation with the functionality. PHP manual was an amazing reference and the "comments" section always had a bit of solid wisdom for ya (and you needed it cause of the warts). Go's manual (and the blog) are such a fine resource that I rarely need to reach further than what it contains.
It's much easier to use what's included when the documentation is accessible and clear.
There's no reason to talk about PHP in the past tense, it's still around and actively being maintained. Of course, it has fallen out of favor and been replaced mostly with Node et al (which is IMHO a change for the worse, but opinions may differ).
PHP's "batteries included" approach is probably a product of it being launched before package managers became ubiquitous (the PHP package manager Composer was only released in 2012), Go's is a conscious decision.
FTA: At home, I run Voxel labs on a Minisforum MS-A2 mini-PC with a 32-thread AMD Ryzen 9 9955HX and 96GB RAM.
I have 128 gb dd4 system sitting at home - Emulation (mostly of clients nonsense) has been massively insightful over the last few years. And as a dev, being able to do all my work "not on my laptop" has been a major blessing.
When prices normalize build a couple bulky machines and a have a stack of small and cheap (I have a bunch of 8th gen i5's I got used). You can do a lot (and give yourself room to grow/upgrade) on a pretty tight budget.
I think there is a quantifiable measure-create a test dataset containing questions and human expert-written answers; give the questions to an AI to get corresponding AI-generated answers; then ask humans to guess which answer was human-written and which was AI-generated; now you can measure how much better than chance the humans perform at AI output detection, which gives you a quantifiable measure of the human-likeness of the AI responses.
> This doesn't seem like a story at all given how robust this debt is trading.
FTA: That is exactly why CoreWeave-leased paper prices at 7.00% to 9.875%, against the 5.70% to 7.875% that hyperscaler-leased bonds command, per bond market data compiled by CCIR.
Just so were clear: Core weave paid Micron and NVIDA 80 percent margins to buy equipment at a 9 percent lending rate so it can sell it to providers who are subsidizing tokens at 20x their own API prices (that are likely also subsidized).
Could it all work out? It could. But it does not look like there is the financing to keep building - thats going to impact prices and the moment that happens this debt looks even more risky.
What you've written is the demand for debt. What's the supply like? You can't conclude demand exceeds supply without knowing both sides, and the US has printed more money than any other time in history.
The debt (bonds basically) is sitting on the books of investment banks. The article here is 5 days old - that "9 percent" figure is already out of date.
Here is today: Shares of AI cloud provider CoreWeave (CRWV) dropped over 15% Wednesday, a day after venture capitalist Chamath Palihapitiya posted on X a chart that showed the yield Nvidia-backed firm's 2032 bonds hit 15% yesterday as a sell-off brought prices into distressed territory. SOURCE: https://finance.yahoo.com/markets/stocks/articles/nvidia-bac...
I can conclude that supply is far exceeding demand because of the rates attached.
Bonds are a weird product. As long as the issuer honors them (remains in business) they are a guaranteed return.
In this case CoreWeave has sold ~4billion in 5 year bonds at a 9% rate. If you buy one for 1000 bucks you will get 2x45$ payments a year for 5 years. Then at the end a 1000 cash payment.
For the sake of argument let's say you bought this earlier this year. You now have an asset worth 1000 dollars. If it is a good bond (A US treasury or Municipal Bond) you can borrow against it from lots of place - you have something worth 1000 dollars on your "books" (accounting). For the earliest sales (you) core weave has already made a 45 dollar payment.
You now have a 1000 dollar asset and 45 bucks in cash. IF you hold to maturity and keep all the money you get 1000 bucks back and 450 bucks in interest.
What happens if you want to sell it? Maybe rates went down a lot, so that 9 percent is huge, and the AI boom is going gang busters so that 9percent looks like a good deal. I offer you 1100 for the bond. You now have 1145 dollars in cash.
What happens if you need money, and rates are the same and CoreWeave looks the same? I will offer you 980 for it... You now have 1025 bucks in cash... and I get the rest of the payments (45$x9) and 1000 bucks back at the end. (This has tax implications).
What happens if you want to sell cause you need money AND, you think CoreWeave is gonna burn, or this "ai thing" is a bust. Well it's likely that I think that too. the bond is only ever going to pay 45X9 more payments but I want a higher rate. I say to you "I'll give you 800 for it" (a 15% rate), cause CoreWeave is fucked. You need the cash so you sell it for that much and have 845 bucks. Thats a loss of $155.
Let's say you bought a billion of those bonds, and expected to walk away 1.45 billion in 5 years - but you sold them all off today and LOST 155 million.
Someone owned a LOT of 9% Core Weave bonds and either found a BETTER investment -OR- they wanted out of CoreWeave and no one had the money or desire to own them. The bond buying market is announcing that "We don't want any more CoreWeave Debt".
There are a lot of problems with CoreWeave: those 5 year bonds it sold, well the contracts to support them (15 billion a year in revenue on 60 billion of debt) only run for 3 years. At current rental rates for long term commitment NVIDA hardware had a 3 year pay off term... But thats not power, the building, staff, the business was (is) marginal.
Who was selling the debt? Well my guess would be a chunk of it was JP Morgan (who set up the bonds to start with). Dumping it right after you announce your numbers (they were a minor miss) and lower your target for CoreWeave's stock would be an ideal time. But that is PURE speculation.
So 9% is the yield and people demanding 15% for new bonds instead of the previous 9% means they lack confidence you'll be around to pay back the original principle?
Are you saying there are existing bonds at 9% being sold for some amount of money (possibly even less than the original principle?)
How does selling the $1k bond at $800 translate into 15%? The bond was originally worth $1k+$45*10=$1450, now $45*9 ($1405?), so if you buy it for $800 that's approximately 85% of its original value ($800+(9*$45)/$1405) and thus a 15% decrease? How does that relate to the original 9%?
Is this about what existing bonds are being sold for or new bonds?
Presumably a us treasury bond yield is basically the same as inflation?
> Are you saying there are existing bonds at 9% being sold for some amount of money (possibly even less than the original principle?) ... Is this about what existing bonds are being sold for or new bonds?
The bonds were issued earlier this year. Their sale is a loss for someone, and NOT a small one.
> How does selling the $1k bond at $800 translate into 15%?
1000 5 year at 9 percent 5x90(450) + 1000 back.
If you need to dump it for 800, the bond still pays out 1000 in the end, The issuer still pays out 5x90(450). The 200 bucks I pick up at the end gets spread out over the 5 years so it looks like 5x130 (650+800).
That would be 16 percent, but all the numbers for the sake of sanity got rounded from their real ones. 9.xx rate, 8xx resale price, and 14.9 (closing on 15) rate.
It gets calculated that way because if you need to sell more new debt it only makes sense to know how to price it.
> Presumably a us treasury bond yield is basically the same as inflation?
Sometimes, often not. But Treasury Bonds have quite a few advantages. No state tax on the interest payments. They are almost as good as cash, so you can borrow against them at a pretty good rate (Better than a HELOC and way better than a credit card). Lots of institutions will give you a line of credit for up to 90 percent of the value of the bond. You can also borrow against them for margin trading (60percent? better than stocks) -
Cryptography, DeCSS ... we have a long history of trying to block things like this and it simply not working.
There is a long, convoluted set of legal arguments that are going to make the weights of a model "facts", rather than "code" making it even harder to "ban". Furthermore having them not be "facts" is going to open up a whole other legal argument about what they are and what the attribution is.
I suspect that they are fishing for a case, so a judge can set some precedent.
You're going about this backwards. They don't have to ban models. They ban the consolidation of compute without a license. This will be far easier and already has legal precedent.
At all. It's easy for registered business to license compute for example, under laws like this. But if a user like you gets too much unlicenced you could be prosecuted under munitions laws.
With a stroke of a pen the us.gov could stop GPU sales tomorrow unless you register and keep track of your compute. Manufactures would have no problem integrated systems to monitor users cards if they get to keep access to the market.
All going in the same direction? With a sun? With a full mouth (pelicans dont look like that, but a cartoon one might - why no reference to realism). Most of them have clouds.
There are helmets (makes sense) but just as many scarves.
Is any one going to draw a gray pelican? Does it matter?
The question is designed to be as generic as possible, and were converging on the most generic and safe answer ... the question is WHY (more so when humans would have much higher diversity, even if wrong). We're asking the generic question because no one wants to repeat the "German solders from 1940's" fiasco: https://archive.is/5l66U (archive link to skip pay wall NYT).
This is what "alignment" looks like? Are we reaching some "ground truth", some platonic form around pelicans on bikes? Are we just back feeding the best output into the training data and this is the convergence we ended up at? Is someone who LOVES pelicans deeply offended that all of them have full mouths and it's just not anatomically accurate and we are all missing it?
I realize that some of this seems over the top, but there are foundational conversations to be had around this without the conversation devolving into someone accusing someone else of having an "-ism" style bias.
I have to wonder if in the long term this bot is an uber product or a restaurant owned product.
reply