I think keeping models around for a defined period of time is fine, but fracturing your model offerings like that (keeping around multiple versions of the same model) is very hard to do economically. The economics of the AI ecosystem are dominated by queueing theory constraints that make it extremely cheap to serve predictable traffic loads, and extremely expensive to serve unpredictable loads, and any time you split your offerings like that, you make both less predictable and therefore more expensive to serve both versions.

If I were paying anthropic prices, I'd expect it, but Deepseek is a super scrappy upstart in comparison and intentionally arbitraging on price. I would never expect them to do that.

Based on what passes for “engagement” and “content” on their platforms, I’d estimate that a majority of their users are exactly that.

So my country has somewhere between 100-200 million MAU for facebook ane.

Facebook marketplace is huge here.It's basically our craigslist. Hobby groups/communities are also huge. If I wanted to find the local jogging or chess club, I'd use Facebook.

Just remember that outside your bubble there's a whole wide world with billions of people living in it

The company is hurting people outside the 'bubble'.

$295 one time permanent license with free upgrades and can be migrated between different computers?

I don't have a single reason to do any form of video editing, let alone professional video editing, but that kind of pricing model makes me want to buy a license just to make sure that companies like this stick around.

Black Magic supports all three platforms and they’re not exactly a large company. They show every day that there is no excuse/reason not to support multiple platforms. Anything else is just an excuse by a lazy developer. The developer for Blender is also the same. A user of both programs.

Black magic also make a program that support/ties into the camera hardware on smartphones and they support both Android and iOS.

> $295 one time permanent license with free upgrades and can be migrated between different computers?

I've been using Resolve since version 14, still getting upgrades to it! My license does only work on one "computer" at a time though, so when I switch between Linux and Windows, I need to re-enter and activate the license, as when I do that it deactivates the previous (other OS) one. You have seemingly unlimited amount of times to switch though, as I've probably done it on a bi-weekly basis for quite some time.

I don't know if something has changed, but if you buy resolve now you can have it activated on two computers at the same time. I've got my resolve studio license activated on my mac laptop and linux workstation at the same time.

They make their money from hardware (and hopefully will continue to be able to do so).

This is commoditizing Adobe for them.

Plus you get the license for free when you buy a camera from them!

I’ve been using DaVinci Resolve Studio for 12 years now with frequent updates, with the license I got with the original Pocket Cinema Camera for £800 (still a fantastic pocket size cinema camera, shoots raw cinemadng and ProRes with a beautiful sensor).

One of the best founders and companies around.

Having a powerful video editor you can grab any time you need to is useful in a way few other bits of software are. Sure, for trimming and even some basic editing or watermarks, ffmpeg works great, but there are times when having full editor ui is beneficial

It really is amazing. DaVinci used to be something that required a dedicated workstation with proprietary hardware cards, and most used it via renting editing bay time in studios equipped with it. Like just buying one of the workstations and then renting it out was a viable business model if you had teh startup capital (and I randomly knew someone who operated such a business years ago).

Highly recommend the studio version. The ability to edit by transcript is worth the price alone.

Yeah this was my experience too. I have an undergrad business degree, but got nerd sniped by an optimization problem, found a solution with constraint programming, and ended up going down a 15 year operations research rabbit hole with it.

Many people say that the way to tackle a hard problem is to break it down into smaller problems. I disagree. The best way to tackle a hard problem is to break it down into a defined search space and as many seemingly-redundant constraints as you can possibly list, then dump it all into a solver, go take a nap for a few hours or possibly a month, then come back to the problem solved for you.

Found minizinc/CP-SAT recently and I’m obsessed

With hiring from management consulting companies coming in a close second place.

Canned sardines have been a staple of my diet since I was in the womb. I can’t wait for this trend to die off, it has been so difficult to find them lately. I mean I understand all of the reasons for it, and I can at least appreciate that sardines are the most sustainable form of fishery we have, but goddamn it is super frustrating.

I haven’t noticed this at my local Costco in Northern California (plenty in stock this week) but maybe I will next week with this making the news. I usually wait to stock up when it goes on sale but I only eat two cans a week.

Which type do you buy? In olive oil? Any specific brand?

I hold out hope for (human) work to coalesce on creating a stable/secure microkernel (SeL4 anybody???) and getting it mature enough to build an OS on.

Once that is done, then you build out the absolutely massive collection of userspace drivers that would be needed to replace linux by leveraging LLMs and other automated techniques. The good part is that by keeping the drivers in userspace, the security risk of generative techniques for driver development is pretty low.

I thought work was finished to allow Hurd to use Linux device driver code as is?

Mach's problem wasn't that it had slow IPC, but rather that in terms of microkernels, Mach was pretty damn big.

There are some very fast microkernels out there, like the L4 family, which negate the IPC overhead of microkernels by being small enough to fit entirely in the L2 cache of most processors. Linux may only have the single IPC call per round trip, but it's a fucking huge kernel and there is typically a ton of cache thrashing going on.

>small enough to fit entirely in the L2 cache

Shouldn't the "hot" path fit in the L1 of a "modern" processor (100 kB+)?

Are you saying there is cache thrashing because callers often sercice rheid own requests themselves? If you don't want to service requests in the same thread, doesnt it mean you have to spend entire core(s) for running the kernel?

The snark artist in me would say that Linux is a kernel plus tens of millions of lines of code that absolutely should not be in the kernel but are.

Ok, I'll bite. Which parts of the kernel should not be there, given that Linux is not a microkernel? I maintain that most of the code is there for hardware support, for security, or to support user-space application models.

I'll grant you the in-kernel nfs server and ipsec implementations, but those are not tens of millions of lines of code. The only way I can reach that number is by lobbing off a huge part of device drivers into userspace, but that's not really an interesting discussion as it seems purely subjective.

Hot take: filesystem drivers. Final stage boot should load two images: a kernel which surfaces block devices via HAL, and a one or more filesystem drivers which function as external backends for the in-kernel vfs system.

This would facilitate more innovation and ability to quickly resolve reliability issues with filesystems, make the whole “can you boot off of it or do you need a kernel module” distinction obsolete, and would massively reduce the temptation for the kernel to support certain APIs only for certain filesystems (let me inotify on procfs, and use nonblocking IO on any FS I like, dammit!)

You're welcome to implement inotify for procfs right now.

Nonblocking I/O only makes sense for IPC. You want async I/O, which Linux doesn't support but Windows does.

> You're welcome to implement inotify for procfs right now.

I more meant that if filesystem drivers had always been external to the kernel, then it would have been harder for so many syscalls to develop filesystem-specific behavior (e.g. things that work on block FSes but not nfs/tmpfs/9pfs/procfs and so on--inotify, atimes, inode stability, sparsity, stuff like that). Being monolithic made it easier to get lots of FS-specific exceptions, which results in some things requiring you to carefully parse manpages or debug EINVALs based on where files live. The benefit of that approach is that it enabled fast development of new features, but looking back I'm not sure if it was worth it.

> Nonblocking I/O only makes sense for IPC.

I want O_NONBLOCK to not fail for disk files. It could lie and block, or only report ready once things were in the buffer cache, or only report ready once part of the VFS response came out of the hardware interrupt/command queue, or any one of several options. I don't know which if any of those count as "async I/O", but the core problem here is ubiquitous syscalls (read/open/inotify_add_watch) whose failure-or-not is contingent on the type of object you give them (blockdev file vs. non-blockdev file vs. semaphore vs. socket vs. pipe vs. timerfd and so on).

> You want async I/O, which Linux doesn't support but Windows does.

Linux kind of has async block device I/O with io_uring now, but they repeated what is basically the Linux original sin mentioned above and made it contingent on whether a device/driver/filesystem support asynchronous operation or not.

Windows overlapped I/O definitely did it first and better though.

As someone who worked on logistics optimization algorithms for Amazon, I’ll just say that the one thing Amazon did best was have clueless upper management continuously make poor strategic decisions that continuously nullified all of their improvements from optimization.

What was the root cause of that? (Company consciously prioritized other things than those logistics optimizations? Individual incentives lead to management behavior that was against company's intent? Bad hiring and retention practices for upper management or whomever was informing them?)