I agree with the idea that it's not probable but I do think it's important to point out-- we only need these elements to run with the bandwidth they have and the token rate because we want to see things in human-scale time. But slow things down to a 1tok/sec doesn't matter to this hypothetical anti-aligned LLM. Time is, after all, relative, and it's not like LLMs give a shit how long something takes. They don't have squishy stupid organs that fail after a certain amount of time, or those pesky glands that emit impatience hormones.
But yeah you're not going to be able to shard out the terabytes of Fable weights that are needed to run inference without addressing some fundamental physics problems.
Companies are amoral. If killing people boosted their share price they would issue every employee a machine gun in a heartbeat. The only reason this doesn't happen is it's illegal.
One possible solution is to only be able to contact someone if you have received an invitation code from them out of band. E.g. "scan this QR code to add me on signal". Such an invitation code should default to single-use but users should be allowed to generate standing invitations so that businesses and the like can print and post one in their store or whatever. Start getting spam from one of your standing invitations? Just revoke it and make a new one. Presumably the Signal folks can come up with more alternative solutions than the half baked one I came up with after thinking about it for a minute, they're clever cookies.
Nothing is the hard part. AI has very large working memory, no 5+-2 issues, no cognitive boases. AI has ingested every word of text in the world, it has all the design patterns and syntaxes in its weights. It's the worst it will ever be.
> watermarks never took off because the technical complexity and lack of standardization
Both audio (sports) and video (netflix et al) watermarks are alive and well. They are used to trace piracy (ie account ID), not for content ID (at least AFAIK).
Somewhat off topic but also ironic - I recently upgraded my version of Android, and when I opened YouTube for the first time on this video it asked for permission to detect nearby devices. I get why it would need that to cast, but at this point it's not paranoia to think that granting that permission just means it would scan whenever I have the app open and upload the data to YouTube servers for no good reason.
Yeah, I mean my slice of the internet is full of video shorts of people using dead ballpoint pens/tetra paks/kitchen ingredients/Temu pasta makers, or silicone baking mats/budget acrylic paint/leaves/laser printer copies to make really cool, unique prints.
The idea that anything about contemporary art needs further democratisation is kind of insane; there is no art form out there that you cannot find someone more or less begging you to learn for next to nothing.
> This was also a time when Flash content was supported by Apple QuickTime
While QuickTime supported some Flash content, my recollection is that it only supported simple animations which could be mapped onto QuickTime's own interactive video features. I don't think it supported ActionScript, so this kind of interactivity would have been beyond it.
yep one of the first rules of any market should be
you sell a product -> you are responsible about the product
with a ".. -> you must be honest about the product (/what the products does, provides, etc.)"
to see how absurdly not followed this common sense(^1) is in the us look at supplements
1. provider are not hold responsible, non small parts of sold supplements are harmful in one way or another. Even cases like supplements advertised for pregnant woman with highly dangerous components for pregnant woman existed, no major change followed.
2. high amounts (AFIK 80% or so) of supplements sold in the US are severely dishonest and don't quite contain what is advertised. And this isn't just about it technically containing the supplements, but in a form the body can't absurd. But it goes so bad that "it's just sugar pills" is more on the harmless side... (and even that is dangerous given that many people take supplements for fixing health issues caused by to little of what the supplement).
not only has the US not fixed it when it started going out of hand, they went explicitly out of their way to not fix it.
(^1): (Common sense for anyone looking a bit into market dynamics/consumer protection etc.)
That is tiny. I just checked Uk legislation and it seems to require £1.2m cover for damage to property and unlimited cover for personal injury and deaths.
Depends on how you use your computer. If you're mostly a developer/analysis terminal jockey with some browsing-- Qubes is a tremendous upgrade even if you don't care at all about the security properties:
The normal qube model of template OS vms + ephemeral app overlays makes it a cinch to troubleshoot complex issues because you can scribble all over the VM (e.g. go ahead, monkey patch your system LIBC if you want!) and all those changes will be gone when you restart the VM. Once you do find a solution you like, you can apply it cleanly and intentionally to the template. If all you were doing was a one-off, then no need to go make it permanent.
I've had friends screwing around with AI agents get their systems really screwed up because running the agent in a VM was work and requires maintenance. .. in qubes it's just the natural way to run it, a few clicks and you're good to go. And the maintenance overhead of running in a VM is mostly non-existent.
However if you're doing stuff that is Video heavy-- particularly gaming, and to a lesser extent CAD, video editing, etc. Qubes really brutally hurts video performance. It can be somewhat offset by running on higher end hardware (and then getting performance of a few year older system). Even just watching youtube videos is obvious impacted.
Similarly, it dents battery life. This is addressable via additional batteries given that now laptops are usbc powered and 100wh external batteries are readily available. But this is something of a lifestyle question.
Even before the AI-apoclipse I considered qubes to be non-negotiable on laptops-- there are just far far far too many browser RCE vulnerabilities to consider anything less for any computer that isn't a total security write-off.
