What has changed is where the durable value actually lives. It is increasingly useful to separate the stack into a few layers:
The computing, IO, and compiler kernel libraries based on CUDA, compiler frameworks like MLIR or JAX’s XLA, and of course Apache Arrow.
The database systems and caching layers, ideally connected with ADBC’s zero-serialization connnectivity.
The language bindings and orchestration layers that expose those capabilities.
The application or agent interfaces that sit on top.
When viewed this way, most of the long term value clearly resides in the first two layers (compute and data access), not the last two.
I don't think OP is looking for context from the AI model perspective but rather a process for maintaining a mental picture of the system architecture and managing complexity.
I'm not sure I've seen any good vendors but I remember seeing a reverse devops tool posted a few days ago that would reverse engineer your VMs into Ansible code. If that got extended to your entire environment, that would almost be an auto documenting process.
> I would be surprised if AI prices reflect their current cost to provide the service, even inference costs.
This has been covered a lot. You can find quotes from one of the companies saying that they'd be profitable if not for training costs. In other words, inference is a net positive.
You have to keep in mind that the average customer doesn't use much inference. Most customers on the $20/month plans never come close to using all of their token allowance.
Very good question! I believe they are simultaneous positions. Skyfield has facilities to calculate the light propagation adjusted position but i didn’t use them. Would you have? Is one more “correct” or more likely to be anticipated by future sentients? I’m always unsure about ther design details.
Also there is an other skewiness. Because obviously the drawing is not to scale the moon position can be correct from the sun’s coordinate frame or the Earth’s coordinate frame, but not from both. I choose to make the moons “correct” in the sun’s coordinate frame. Meaning that if you were hovering over the ecliptic frame looking down at the Jupiter during the wedding and rotating the pendants so the sun is in the direction the real sun is, then you would see the moons under you in the same arangement as they are on the pendant. But if you would stand on the surface of earth (during the wedding) and look at Jupiter you would see the moons in a different arangements than a tiny human standing on the earth dot looking at the jupiter dot. (And not just because of the time delay difference, but because the coordinate systems are different.)
Which is weird. Because the wedding happened on Earth, not hovering over the plane of the eliptic over Jupiter. So maybe that was a weird choice. (And not even talking about how north-centric it is that i decided to draw the diagram from the “north” looking down at the eliptic, instead of from the “south” side. These are all kinda culturally driven arbitrary choices. Would love to have none of those present but I haven’t found a good and principeled way yet.)
I think it depends on what you mean by "is a function". You can think of a constant, `x` as `x_: () -> {x}` (i.e. everything can be indirected). It could be argued that this is "philosophically" "useful" since getting (using) the value of `x`, even as an actual constant, requires at the least loading it as an immediate into the ALU (or whatever execution unit).
You have a laundry list of complaints about Canada's action wrt the US. What would someone like you on Canada's side of the border offer in response, do you think?
All of these issues go back long before Trump, who has made things uniquely worse. But any two countries with as long (and tightly bound) a history as ours are going to have constant points of friction. Are you suggesting Canada is uniquely a "fake friend" in this equation?
> They may not be able to register with Google, but they can make an RCS app nonetheless.
They can make a "standards compliant" RCS which will be able to connect to literally zero carriers or servers on the planet.
In fact Google Messages "RCS" doesn't even implement the RCS standard. They use a proprietary protobuf api exclusive to Google. Google messages can't connect to any "RCS" servers except Google's.
> Twilio has RCS too
Twilio, Bandwidth etc. are gatekept by Google and are basically just a middleman reselling Google's product.
Yes, but some of them run countries where we live in (and therefore working for military contractors in this country is literally helping kill people to take their stuff). This includes US where tech is so heavily concentrated.
Went to Badwater Basin in Death Valley last week and there's miles of (bad) water. Unfortunately the Park Service but the kibosh on paddle boarding, etc. Should be a good bloom this spring.
An illegal immigrant commits a crime, say vehicular homicide. ICE lodges a detainer against this person, and the local PD refuses and instead releases the offender. As a result, ICE runs a tactical team out to go pick him up.
This is the outcome that you appear to believe is optimal, and you are intentionally using emotionally loaded words like "lenient" to attempt to guilt me into retreating from my position that this is, in fact, not an optimal outcome. In many cases like this additional crimes are committed before the offender is apprehended, crimes which are of course 100% preventable, without you and your "leniency".
>Tell me you at least recognize the difference between actively doing something against some group, and merely not helping them?
Technically speaking, you are right. These states are actively working against their own citizens, not the Federal government.
That's not really an error, that's a fundamental feature of unit economics.
Fixed costs can't be rolled into the unit economics because the divisor is continually growing. The marginal costs of each incremental token/query don't depend on the training cost.
I’d recommend providing a lot more screenshots and information about how the core DAW functionality works in comparison to other DAWs. As is I can’t see enough about what this would feel like to spend my time downloading and trying it
I suspect it will falter with or without wider adaptation. I’ve been happy to adopt the free offerings and still engage my brain. I don’t see how they get enough revenue to pay for the massive outlay of all these data centers.
Functionally, it is very similar to Flatpak. The main reason people do not like it (for reasons independent of sandboxed applications in general) is that Canonical controls the store and that it is not open-sourced, and that it is very difficult to remove it on Ubuntu setups (a major pain-point for people who need an unsandboxed Firefox setup).
>40% of international students were cheaters. When some were caught they fell back on cultural norms as their defense. The university never balked because those students, or their institutions, paid tuition in cash.
Twenty years ago, at Vanderbilt, this would have been an understatement — particularly among non-citizen asians.
I remember in organic chemistry an instructor attempted to re-give the same examination ("because ya'll did so terrible") and it was struck down by a dean as not allowable simply because the Honor Code was to be invoked that nobody/groups would share answers (yeah sure okay).
The minority following the Honor Code ended up getting into lesser graduate schools (e.g. myself) — because most courses didn't curve and VU didn't give out A+ as a grade. I have specifically not mentioned the specific country which cheated most-blatantly... but everybody from back then knew/knows.
I've been using it exclusively for a few years now, but I'd still say: it depends. e.g. we've yet to see an actual photoshop replacement in OSS (Krita seems to be on a good trajectory now that they've pivoted to a broader scope). First stop is checking whether you have viable replacements for things that you use.
Ubuntu is going to strong-arm you into Snaps, the snap-ectomy is nontrivial, and they have a habit of reappearing. Some people don't have a problem with Snaps - so non-issue if you don't care. Otherwise I'd go with a downstream distro that removes them: pop os, mint, or even upstream (Debian).
XFS is an extremely mature file system if you don't need anything fancy, and you're probably less likely to lose data compared to $proprietary. The other major ones (ext4 and btrfs) are probably just as good, but XFS honestly does stand out in terms of maturity and simplicity.
A common trap is trusting the installer with partitioning. My last Ubuntu installation ran out of space on EFI. 5gb is overkill, but given how abundant disk space is, who cares. Separating / and /home is a good idea for rescue/reinstalling but without btrfs subvolumes (Ubuntu uses btrfs subvolumes by default) it becomes a bit challenging to figure out how to dice things up: e.g. docker containers are stored in /var, so they can deplete your system drive space. Last time I didn't use btrfs, 200gb for / never caused issues for me.
Oh, and Windows has a habit of removing other boot loaders from its drive. If you dual boot, use a different disk for the entire installation.
That's really the extent of the gotchas I'd give to a person literate enough to install an OS. I would slightly urge towards immutable (Silverblue), but Ubuntu is just fine.
“””
What has changed is where the durable value actually lives. It is increasingly useful to separate the stack into a few layers:
The computing, IO, and compiler kernel libraries based on CUDA, compiler frameworks like MLIR or JAX’s XLA, and of course Apache Arrow. The database systems and caching layers, ideally connected with ADBC’s zero-serialization connnectivity. The language bindings and orchestration layers that expose those capabilities. The application or agent interfaces that sit on top. When viewed this way, most of the long term value clearly resides in the first two layers (compute and data access), not the last two.
“””