That’s exactly the same as going into a furniture shop and going “if this furniture was so good, then why would they want to sell it so badly”. They’re a furniture shop - they make and sell furniture. There’s only so much furniture you need for yourself after all no matter how good it is.
Investment banks make investment products for investors. They do keep risk for themselves so you will find when they sell these loans they will often keep a slice, but they only need so much Oracle datacenter loan risk - they want to have free capital to find other opportunities to fund.
If a loan is supported by a known lease it is more attractive to investors than a regular loan, so I somewhat doubt they are thinking about it like this.
The real reason they will be doing this is this is what they do in general. They don’t hold on to huge amounts of risk with predictable returns (because it consumes their operating and regulatory risk capital inefficiently) they will instead find buyers for that risk, which frees up capital they can then use on their next thing. That basically gives up some of the long term, low return, low volatility risk for shorter term, higher return, higher volatility risk.
Speaking from personal experience which is admittedly quite old at this point, but it used to be the case that Debian definitely didn’t go out of its way to try to pull in devs. When I had a few small open source things myself I packaged them for RedHat (this is pre-fedora when RedHat didn’t have a commercial and free version, they just had “RedHat Linux”) and looked to package them for debian, given I was actually using debian for a few of my personal servers. I made the .deb packages just fine but found the Debian community were definitely not trying to attract devs[1]. I couldn’t get anyone from Debian to sign my gpg keys which if I recall correctly was a necessary part in getting my package upstreamed[2] and in the end I just gave up on it because I’m really not interested in joining a community that is so unwelcoming.
[1] although it was maybe specifically just me they weren’t trying to attract.
[2] to the point where I actually worked with someone in my day job who was a debian dev and he wouldn’t sign my key without me producing physical official ID like a passport or something. Just really bizarre level of paranoia like a government kyc process or something.
It’s two sentences for goodness sake. Why on earth should anyone spend their time reading your “thoughts” when you don’t think it’s worth spending your time to write them?
More substantively, I don’t believe it is true, but hypothetically why would “the real” problem be having 100 distros on distrowatch with little to choose between them? The benefit of open source for the individual is supposed to be the opportunity to learn by doing and to have the freedom to do things however you want, and the benefit to the community is this extreme Darwinian process where lots of crazy people try all sorts of things and the community as a whole picks results they like. All of these benefits arise precisely because there are a lot of projects, many of which don’t achieve mass adoption.
I think they should have included the official NYC procedure, which is:
1. Dig out around the affected area
2. leave massive dent in the surface for what seems like years
3. Maybe cover it with a few janky bits of wood and/or metal sheets that make a hideous clanking noise all day and night and have the same approximate surface friction as an ice rink so are pretty murderous to any 2-wheeled road user
4. Leave this solution to mature like a fine wine
5. I really mean single malt whiskey. You can leave it basically as long as you like
Whoa, fancy. All we get are open pits with a few barricades around them until the news finally starts talking about the tree growing in the middle of the road.
‘Course where i grew up we had it ‘ard. Aye…
We’d get woken up 2 o’clock in the mornin’ after sleeping in a road crater filled with sulphuric acid, race down the street naked in the dead o’ winter to distract the hungry velociraptors while one of us used a dull butter knife to chop off a limb we’d throw into the road hole so the steamroller could come by, flatten it to patch the hole, then use the stump to tamp the edges into place.
And drivers these days keep complaining how tough it is to get a pothole fixed…
I like Atlanta’s solution even better: after becoming egregious enough sloppily bolt down a much too tall metal plate over the pothole and those that have proliferated nearby. Cross fingers and hope they won’t coalesce as a sink hole and as with everything else: “Go Dawgs”
My understanding is the DOT gets pissy if locals fill in the pothole themselves, but I imagine that there's enough interested people to do vigilante road repair if they weren't subject to government harrassment.
nobody leaves a dent in the surface on purpose, the problem is that whatever caused the pothole is almost certainly still causing it to sink under the surface. A patch doesn't fix the problem, it just makes it less bad.
There's a section of town that was developed on an old swampy bit of land. They drained the swamp but did not let the land dry/settle long enough after draining the water. This caused parts of the road to sink, but not enough to break/crack the roads. Instead, they just have swells as you're driving along. It's actually impressive on how much sinking happened without breaking the road itself.
The local obscenely named utility company dug in our road a few years ago, necessitating a large patch. They proceeded to drive through their patch as they left. The tire dents & ridges are still there years later. You're right, of course, but I think you may be overestimating the concern patch crews give to their craft.
That's a method in the manual - "roll-and-go". The fact that it's still there years later is actually proof that the patch was effective, if maybe not well applied.
When someone says the error is of a certain magnitude they mean the absolute value of the difference between the the two things, so what they're saying is that the values they produced with their approximation are about as wrong as the difference between the actual values and those values cast to float16. The exponent is most definitely important and would be included in that.
I read that too, but I wondered whether elementwise error is the right metric. Surely the actual error metric should be to evaluate model performance for a conventional transformer model and then the same model with the attention mechanism replaced by this 4th order Taylor approximation?
To spell it out for myself and others: approaching equivalent calculations for each individual attention block means we also approach equivalent performance for the combination of them. And with an error bar approaching floating point accuracy, the performance should be practically identical to regular attention. Elementwise errors of this magnitude can't lead to any noteworthy changes in the overall result, especially given how robust LLM networks seem to be to small deviations.
Me too. Love my DM42. I also have an HP35s and two HP12Cs still on the batteries they were rocking over 10 years ago when I used them daily in my finance days.
To answer the op's question, any RPN calculator is wildly better than the default calculators on things like iOS once you take the time to learn to use it[1]. If you want a good calculator app, Thomas Okken's free42 is a great, free HP42S emulator which runs on iOS. I still prefer a physical calculator (eg the swissmicros has more screen space so you can see 4 lines of stack and dedicated function buttons) but it's much better than the default.
[1] Great support for complex numbers in polar or rectangular form, fewer keypresses for multi-step operations, no ambiguity (whatsoever) about operator order without any need for parentheses, vast number of memory slots (not just one), the ability to write a program to do anything you can do manually on the calculator so you can automate anything you do frequently, numerical integration and root finding etc etc there are many many benefits.
Investment banks make investment products for investors. They do keep risk for themselves so you will find when they sell these loans they will often keep a slice, but they only need so much Oracle datacenter loan risk - they want to have free capital to find other opportunities to fund.
reply