This decision has, effectively, turned LMMs into a supply chain risk.
Before this incident I’d gladly use any anthropic LLM in production features. Right now, this has become a risky decision that can tank my business overnight.
> Right now, this has become a risky decision that can tank my business overnight.
If your business-critical systems rely on SAAS that doesn't have a solid SLA and breach-of-contract provisions that more than cover the damages in the event, you've made "a risky decision that can tank [your] business overnight".
If the software your business depends on can't run indefinitely without getting permission to operate from someone else's systems, then you're perpetually at risk of someone else tanking your business because they decided that you can no longer use that software.
Anyone doing product integrations should recognize it’s a perpetual risk but why stick to the platform that will require US citizenship demands for future models especially when there are other labs with reasonably comparable performance that don’t require this?
Anthropic didn’t have to beg for the government to deem their models a security crisis.
> ...especially when there are other labs with reasonably comparable performance that don’t require this?
This access restriction could be
1) A legitimate concern about capabilities.
2) A hype-generation scheme entered into by both the Federal Government and Anthropic.
3) Pettiness.
If we're in situation #1 or #3, then there are no other LLM providers that provide comparable services that are safe from this.
If we're in situation #2... well, I guess how safe you are from this is a factor of how desperately the company needs "wau, look how scary our software is!1!1" hype.
> Right now, this has become a risky decision that can tank my business overnight.
If your business-critical systems rely on SAAS that doesn't have a solid SLA and breach-of-contract provisions that more than cover the damages in the event, you've made "a risky decision that can tank [your] business overnight".
I'm pretty sure that a US government export restriction / ban / etc. would count as a force majeure invalidating all the fancy wording you could wish for on a piece of paper.
The only way to actually control is to self manage in an environment you control.
The long term goal of LLMs is to automate most white collar work, so wasn’t that a risk you faced anyway? Like Amazon basics, which took all the easy to replicate goods they saw on their marketplace.
As long as E2E encryption is not guaranteed and we rely on id verification, the only thing this can do is to limit the 3rd parties that can easily access your data.
Everything else is in the air
As someone whose mother died to pancan, I could really care less on any of the brainwashed old farts in their churches or parliaments.
None of that matters to me or the people suffering from cancers, it’s al Knut a selfish obstruction attaching religion to the research material
Win insists on bootlocker/secure boot, meanwhile most of the Linux doesn’t boot with it or you have to go though hell and back to install unsigned drivers (nvidia, gentle-yall).
I’d all say that Linux is like living in a car with 0 euros and saving up for a house. Simple user can scrape by, but mowing dev work life to Linux is much harder than to Mac. VPNs, inconsistent distro support for weird work stuff and such will make you spend days to weeks of unpaid overtime to get comfortable
Linux can handle BitLocker & Secure Boot just fine. The problem with dual booting in that configuration is rather that every time Linux updates the boot loader, Windows will freak out and stop booting until you enter the recovery key for BitLocker. This can be prevented by first booting into Windows to disable BitLocker until the next reboot and then installing the Linux updates, but in practice I find that I forget about it all the time with my dual-boot laptop (which spends most of its time booted into Linux).
This is a solvable problem and there's even pacman hooks around to do it for you
But also don't blame Linux. Even your comment says the problem is Microsoft. We need to be collectively mad at the right entity if we're going to get them to change. Otherwise they'll keep bullying people and they've found that they can bully people so much it gives them Stockholm Syndrome, where they feel they can't leave.
For quite a while I've found that the much easier answer is to have a physical drive per OS and make sure it's the only drive connected during install, or at least one for anything that doesn't play entirely nicely with multi-boot. Obviously there's downsides to that, buying another drive or you might be using something like a laptop which is less friendly to extra drives, dis/reconnecting M.2 drives isn't as trivial as SATA either.
Secure boot mainly gets annoying if you have an Nvidia card, since the akmod needs to be self-signed. It's not insurmountable but you have to load your keys into the UEFI before it'll work.
Bazzite builds the Nvidia driver into its kernel, so you don't need to do anything special. Post installation it prompts you to do key enrollment, so all the user needs to do is select "Enroll MOK" and type "universalblue".
Microsoft went all in on do more with less and fired/reorged significant part of the company.
Wouldn’t be surprised if the outage is caused by new team taking something over with near zero documentation while all the tribal knowledge was torched away
I’ve tried multiple versions when trying to move away from windows, but was always stuck with random inconsistencies everywhere.
Eventually I had to choose a larger evil and choose Mac after paying for a week of lost productivity installing, setting up, fucking up, wiping l, installing random Linux distros.
They are like Vim. “Alt,letter,letter,arrow,letter,letter,arrow,enter”, etc. Rather than a single combination of keys, it is a series of key presses.
I agree that it might be trivial to set up for spreadsheets, and it would be really useful for other spreadsheets, and many other applications. I suppose a hurdle is how context sensitive the commands are depending on the cell or range of cells activated, and their contents and data type.
Before this incident I’d gladly use any anthropic LLM in production features. Right now, this has become a risky decision that can tank my business overnight.