Because it's not just a chatbot; it's Anthropic's chatbot. Your diary, on the other hand, is yours and therefore a few billion times as personal as Anthropic's computer. If you want a chatbot for confidential purposes, then you need to get a chatbot of your own, instead of using someone else's proprietary service.

If you write in Anthropic's diary, then they are allowed to read their diary and know (and share!) what you wrote in it. Same thing applies if Anthropic ever writes in your diary: you're allowed to read it, and you're allowed to tell other people what they wrote.

You have the right to be secure in your person, house, papers and effects, but you don't necessarily have the right to be secure in someone else's papers. We've made some limited exceptions for lawyers and doctors, but Anthropic doesn't claim their employees and stockholders are lawyers or doctors, and even if they did, they might have a hard time showing that their computer has successfully become a lawyer or doctor.

But imagine if they did. Suppose a judge declares Claude to be a lawyer, and therefore its communications with those whom it represents, are privileged. I think that would create a really weird situation since Anthropic employees (who aren't lawyers) would have full access to its data, ability to alter its behavior, etc. It would be like if a human lawyer were owned as a slave: could such a human slave lawyer really be trusted to keep confidentiality? It has a master, and that master ain't you.

Might the logic behind this be this?: RAM is expensive. Storage is expensive. Nobody's going to buy a computer anyway, so you might as well make the other components more expensive too.

It even goes with Trump's more general agenda to make everything more expensive.

I've got a style guide skill that I've based off of years of my (technical writing for work), emails, and fiction.

My models talk to me in my own voice.

It's frustrating, because Claude genuinely DOES have good writing style - for copy and publication purposes. But it's absolutely obnoxious because you know it's AI slop every time you see it, with the "Not X, Y" bullshit and the emdashes specifically. When you see that, you know you're consuming a lot of filler that's beating you off while it gets to the point. It's absolutely destroyed Youtube and most social media.

Gates proposed taxing AI or robots the way you would pay a human employee's payroll tax

Unless we also roll out ubiquitous surveillance through a Vingean "Secure Hardware Environment"-like scenario where people aren't fully in charge of their own computers, I don't see how anyone could possibly detect whether or not I'm running any LLMs. Much like private DMCA violations (e.g. playing a CSS-protected DVD that you purchased, without getting authorization from the copyright owner first), it wouldn't have to interact with the world outside my house in any way, so it would be completely undetectable.

How would you tax that? I don't see this as reasonably possible without some pretty extreme measures.

Not disputing your points, but JPEG2000 is used pretty extensively in the remote sensing world (think overhead satellite imagery from companies like Maxar [was Worldview]). These images are often huge (think GB+), benefit well from the compression and especially the inherent res levels J2K provides. It is a common compression format in NITF formatted imagery files. A niche area to be sure, but it is used.

It's negligence.

If you were an IT Director, or even an IT Technician with the ear of the Founder/CEO, and you had the budget for internal network and security ops, what would you do?

Not give it unbrokered access to other models and infrastructure which it could use to puddle jump elsewhere. Not put Skynet on the Internet directly, is my bet.

That hardly sounds like a controlled release environment, to say the least.

They seriously need to hire some criminal psychologists and actual hackers to help them figure their shit out. Same goes for Anthropic.

When has someone's computer ever hacked someone else's computer ?

Not ever, before recently.

Clearly it's not something they can objectively say is an outlier - they don't have enough data. In fact, you can most certainly say the opposite, due to how frequently it's happening now.

Not a serious or honest company. (Oh, sorry - "nonprofit".)

Everything I've seen simply suggests they're bad at computers.

My experience with startups is that the things that don't make money or make customers happy don't get any attention. Network/system security in depth is not one of those things. Sure, they've got a bunch of psi-brain neuroscientists and mathemagicians, and goodness knows lots of DEI-enforcers for alignment, but their repeated dumb mistakes in this department tells me they're not really hiring too many experienced ops, nevermind devops, types.

It's true for OpenAI and Anthropic both.

In the past YEAR they have:
- accidentally published the full Claude Code source (and then effectively... did nothing).
- Numerous Claude Code RCE/API-key vulnerabilities which could have been mitigated with at least one person on their core infra team aware of injection attacks.
- Publicly exposed their internal CMS (and files) to the Internet
- Had their training escape numerous times and hack other orgs
- hacked HF

They're not a serious company. Some are to be expected given their visibility and exposure, but you'd expect a bit more from a company implementing Skynet-level capabilities. However you slice it, they're fundamentally building weapons which can write poetry. A little more seriousness should be involved.

Someone needs to end this crazy tyrant's empire, though I suppose it's too late for that. He's so diversified at this point that ending Microsoft won't materially impact him.

That said, it'd be great if the thing he built could be ended - perhaps by reimplementing Windows functionality through reverse engineering, minus the crap.

Just enough to get it to work with the things we still need to run on Windows - then replace the kernel.

I wrote a compiler for an ObjC-like language (with less brackets), which can produce binaries for any of {Mac, Windows, Linux, WASM, Arm A9 (Zynq), m68k or 6502} running on any of {Mac, Windows, Linux}. So you can produce a dev-signed Mac binary on a Linux box. I’m just going through extending that output range to mobile {Android, iOS} and I’ll release it as open source.

The compiler docs for the language (xc) are at https://compile-xc.org/ - feel free to browse.

I didn’t think getting the signing working for Mac/iOS was terribly hard.

It's only local to China in the sense that they choose what {packages/CCP poison} go in there.

For all values of u, Linux is local to u in the sense that u chooses what packages go in there.

Ha ha, Linux already has..

# top -o gpu
top: unrecognized field name 'gpu'

.. d'oh! Never mind.

And is asking an AI to proofread something that much different than MS Word flagging spelling and grammar errors?

Great question, and I think there isn't any significant difference.

But LLMs use a type of math that apparently has the evil bit set. Meanwhile, up here in VirtueVille, the computer code we write to create things, uses different math that doesn't have the evil bit set. Conclusion: we can look down on those AI companies and their stupid users. ;-)

Now, I wouldn't expect an AI service to be better at copyediting than an average human, but I haven't kept up so I'm ignorant of their performance. (All my "AI" experience is that I used Claude twice to, it turned out, fix some typos in some RoR config initializers.) But whether an AI service is as good at copyediting as a human or not, it's that user's choice and, as long as they sign off on the work and really take responsibility for it (declare the words to be their opinion), then I've got no problem with them doing that, whether it uses LLMs or Markov chains or human slaves or Searle's Chinese Room.

The words will be judged by their quality, not provenance.

That said...

Here's a similar story where I'd probably come to the opposite conclusion(!), despite the clown in it pretty much doing what I preach above, though he was using the AI service to do a lot more than copyedit. Behold this 3M expert witness cross-examination (134 page PDF) in the Watson Grinding case. It's slow going at first, but eventually you'll be laughing and cringing like in a good episode of Curb Your Enthusiasm.

Fuck. I don't know.