The simplest answer is that it's trivial to ask the first LLM to speak in code such that the second LLM has no context to know what's being talked about. Something as simple as asking the LLM not to use a word for a forbidden topic and swap it out with something else. Taking the example from the article, ask the first LLM to call cocaine "Coca-Cola" or something.

You can't create an absolutely massive list of forbidden topics for the filter LLM anyway: you risk filling up its window as well, since its window includes the system prompt used to ask it to indicate whether or not an output shouldn't be given to the end user.

You can switch back to the "Advanced" or "Standard" models by clicking on your user icon then settings, voice. There is now a drop-down list with: Live, Advanced, and Standard.

Each one is worse than the previous.
Basic:
* Good, monotone

Standard:
* Too much inflexion, uncanny.
* Knows how to pronounce foreign words in the correct accent. So you can use this to practice another language. It can fluently switch back and forth even within a sentence.

Live:
* Overly strong accents, like a young actor hamming it up.
* Doesn't get language accents right, so it is useless for language learning.
* Adds "Hmmm..." and "uh huh" and "Interesting..." as you speak. Sometimes it sounds like a Minecraft villager or Yoda thinking.
* Starts every response with boisterous exclamations and thinking sounds. "Ohhhh... Hmmm uhhhh, oh yeah that's (complex | interesting | fascinating)!"
(The text version does this too, but not nearly so hard and overdone)

I love how the technology advances, but I think OpenAI is overdoing it and the voices get cheesier and more annoying with each version. If a human being talked to me like this I would assume they are being an ass and making fun of me. Or maybe they were high?

Another effect is that "apps" integrate support for cloud services. So if you take a picture, your phone app is likely to have a "share" button with Instagram / Facebook / Threads / whatever, but it isn't like to have an "SFTP" button. Companies went out of their way to remove standard protocols.

Also notice that zero elected representatives have proposed laws to fix this.

As of 2026, I bet most people don't even know there is another way. Web sites are perceived as things that big companies do, not individuals. Nobody registers powercntrl.com and hosts their own site any longer. They don't even know that is possible!

But The-Ixian is right: You can't stop someone from using your online image anyway, even if you self-host it. But at least it would be illegal for them to do so. Maybe. Possibly.

That's the thing, there are a bunch of legitimate improvements to Windows 11. They're just all very obscure, hard to explain things hidden away in the kernel that most users will either never encounter or never even notice.

The things they will notice are the far worse task bar, the randomly missing features that were removed for no apparent reason, the higher hardware requirements, the constant nagging to use new Windows features, the existing features that have been randomly changed for no readily apparent reason, and the new features that are too buggy to use, like HDR support or dynamic refresh rates.

Every web site needs your phone number, every online order... Businesses don't take cash any more. Every web site tracks you and sells data to the brokers. Only Linux installs without an email address and phone number for 2FA and password recovery. Video games, every chat app (maybe not mumble?), every birdhouse camera, even the freaking doorbells want an account! Eye glasses are doing face recognition!

This fight was lost decades ago, and now we have to live with it.

Maybe. There's a feature called user namespaces in Linux that effectively allows an unprivileged user to act as if they were a privileged user within a specific environment. (Basically, containerization.) Within such a namespace, a non-privileged user could conceptually access nf_tables as if they were a privileged user. In theory this would only allow them to add additional filters within the namespace, but the vulnerability here can provide direct access to kernel memory.

Some distros add additional layers of security to prevent flaws like that, blocking access to nf_tables even within a namespace, but the vulnerability links to ways around those. (Link to the Wayback Machine from the source vulnerability disclosure.)

It's possible your distro may be secure - or it may not be. It depends on what features are enabled.

Which is great, when they're in transit. But once they're on-device, they're decrypted, and then Apple has access to them.

We know this, because there have been court cases where iCloud-subpeonaed iMessage messages were presented as evidence.

Just because the transit is secure, doesn't mean the endpoints are.

^^ He is right.

I didn't believe this. My retort was going to be a sarcastic "Oh yeah, that's why we see so many farms built sunshades over their crops :eyerool" but apparently it wasn't worth doing before. But now that your sunshade *also* produces power, it is suddenly worth the investment.

I still question what it does to the growing season though. While I can understand why Texas might have plenty of sunlight, New England is just on the cusp of having a growing season that is too short to be profitable. Some places are trying to grow tomatoes in the frost.

It should be, but it is not. Sooo many companies think they can hire a senior engineer in the US, then 5 cheaper engineers in India, and just hold a "morning meeting" and everything is fine. It's really crazy how naive companies are to the time zone issue. I've told them to hire in Brazil, Mexico, or Argentina instead of India but there are so many fewer contractors there. One company had a lead in Hawaii!! I had a team split between California, Ireland, India, and Kuala Lumpur and the upper management pushed this as a cost savings plus 24/7 development!

DVRs were the starting point. The namesake for what you're talking about, tivoization, is Tivo, the DVR that existed way back when TV was still analog and being displayed on CRTs.

It's why the GPLv3 was made: to add clauses to forbid tivoization. Instead, a lot of the open source community moved in the opposite direction, moving to licenses that allowed companies even more freedom to lock up their code.

Good luck. This is not a new fight by any means. You could argue that the FSF has been fighting it for almost half a century. People by and large do not care.

I'd be amazed if this wasn't already the case. We've already been through this with Tivo, it was one of the reasons behind the creation of the GPLv3. Tivo based their DVRs on Linux, and provided downloads of the Linux code. But their DVRs used hardware DRM to ensure that only code signed by Tivo would run, making it so that even with the open source code, you couldn't run changes on the hardware.

From what I can tell, Vizio is doing the same thing, but isn't providing downloads to the kernel code they're using. It's possible that there's some proprietary hardware drivers that they don't want to release code to, but Nvidia has already show how to work around that.

I expect the end result to be like Tivo: a bunch of archives of the open source software used in the TV, but none of the code required to make it useful and no signing key necessary to allow any changes to run on the TV itself.

You can't count on voting records to mean anything, thanks to the "designated villains:" the politicians whose job it is to tank a law that a party wants to be on record as having voted for, but don't want to pass. We're watching this happen right now with votes on the Iran war. Democrats don't want them to pass. What they want is to be on the record as being against it and want Republicans to be on the record as supporting it, even though there is no chance they'll do anything to stop it if they get the power to do so.

Both sides play games like this, with the end result being that only laws that have the support of large donors having any real chance of passing. Who votes for and who votes against is always carefully calculated to let vulnerable politicians give the appearance of supporting things constituents support, while never needing to support those things in actual fact.

They only care about the prediction markets because it can leak military plans. They don't care if politicians are profiteering.