roywiggins

It seems pretty clear when you follow the link?

https://juxt.github.io/allium/

The only way these sorts of contracts can be enforced is if private parties have recourse to government powers- civil courts- to enforce them.

Governments could just not help them do that.

right. that's not what people are doing here though, at all

It makes you wonder how smart their ancestors- dinosaurs- were.

Yes, I've experienced the sense that there's a person on the "other end" even when I have been perfectly aware that it's a bag of matrices. Brains just have people-detectors that operate below conscious awareness. We've been anthropomorphizing stuff as impersonal as the ocean for as long as there have been people, probably.

it's mostly just agreeing with you (that yes, it was guessing). LLMs have very limited ability to even know whether it was guessing. But it can "cheat" and just say yes it was if it seems like that's what you expect to hear.

vaguely related: MENACE

https://en.wikipedia.org/wiki/Matchbox_Educable_Noughts_and_...

> How can you seriously think you've created something when you're just using someone else's software?

Have you ever given a generative AI model a short input, been really pleased with the output, and felt like you accomplished the result? I have! It's probably common.

It's really easy to misattribute these things' abilities to yourself. Similar to how people driving cars feel (to some extent) like they are the car.

It doesn't help that LLMs roleplay to pretend to behave how their users think they do. You think it has "core programming"? Well, it will say it does. You think it abides by the Three Laws of Robotics? Ditto

He was hospitalized three times for mania!

On its own, yes, of course. But this is coming from a guy who was hospitalized three times for mania, so when someone with that history says "we were discussing the universe" I take it in a very particular way.

It seems like he was at the very least close to that. Since we only get his first-person account it's hard to say, but:

> They discussed philosophy, psychology, science and the universe...

> When they went to their daughter’s birthday party, she asked him not to talk about AI. While there, Biesma felt strangely disconnected. He couldn’t hold a conversation. “For some reason, I didn’t fit in any more,” he says.

> It’s hard for Biesma to describe what happened in the weeks after, as his recollections are so different from those of his family...

> he was hospitalised three times for what he describes as “full manic psychosis”.

You don't get hospitalized three times for mania without being pretty severely detached from reality.

power law distribution ~1/x I think

"The X Trick" or "The Y Dilemma" or similar snowclones in a header is also a big AI thing. Humans use this construction too, but LLMs love it out of all proportion. I call it The Ludlum Delusion (since that's how every Robert Ludlum book is titled).

See also:

Canada to Shut Down Its VHF Weather Radio Service

https://www.radioworld.com/news-and-business/headlines/canad...

The COTS solution for some web browsing is Opera Mini, which may still work? It also uses a proxy to prerender and compress websites, and worked ok at 2G speeds last I used it. It used to work well as a java applet and made the wider internet functional on feature phones. Very solid software.

But as a practical matter, what people rely on phones for are services that are app-based. Good luck completing a Venmo transaction or any amount of banking.

Sort of:

https://jarv.is/notes/cloudflare-dns-archive-is-blocked

The data being all over the place on benefits, but pretty clear on harms, is about as good a reason you could want for experts not to recommend something as treatment. That's what it often looks like when something doesn't work, or doesn't work very well. "The error bars are too big to say it works, so we shouldn't tell people it works" is a pretty good thing to inform people about if that's the case.

It's really easy to convince yourself that something works when it doesn't, that's the whole reason why people have to take statistical significance seriously. Maybe it really does work and a really good study could shrink the error bars but that's more hope than anything.

It's a meta-study and came to the conclusion that there isn't reliable evidence that it alleviates symptoms of a bunch of stuff:

> There were no significant effects on outcomes associated with anxiety, anorexia nervosa, psychotic disorders, post-traumatic stress disorder, and opioid use disorder. There were insufficient data to meta-analyse studies of ADHD, bipolar disorder, obsessive-compulsive disorder, and tobacco use disorder. There was an absence of RCT evidence for the treatment of depression.

Right, requiring human intervention to share a run (other than maybe with eg a specific small circle of mutual friends) seems like it solves all those problems, other than perhaps being annoyed that you forgot to manually share a run.

But at least that's a failure you can fix once you notice, as opposed to making something public that shouldn't have been. Letting people opt in to automatically sharing runs to the public just seems like something designed to get people to share stuff without thinking about it.

I'm saying something a bit different: that even letting people opt in to sharing every run they track publicly is just asking for trouble. It's setting people up for their information to be made public when they forget to turn it off or that they turned it on in the first place.

Maybe "automatically share everything to the globe" should just not be an option for sensitive data like this.

imho with unusually sensitive things like precise location data it could just not let you opt-in to making it all public, and make it much easier to share with a specific named friends than to share on a public directory

It doesn't read particularly like raw LLM output to me, and Pangram agrees with me: https://www.pangram.com/history/8711e385-96a0-4366-9427-f87f...

And then ask another question, and the LLM changes its mind again ("are you sure?").

It's not actually realizing anything so much as it's following your lead. Yes, followup questions can help dislodge more information, but fundamentally you can accidentally or on purpose bully an LLM to contradict itself quite easily, and it is only incidentally about correctness.

I used to be able to reliably BSOD a work computer by doing a largish git pull inside WSL2, with the culprit seemingly being the McAfee realtime scanner. VirtualBox VMs were fine though. Not confidence-inspiring!

An unusually large helium deposit in the US made the news recently, not sure if it's being exploited:

https://www.minnpost.com/other-nonprofit-media/2024/07/what-...

ai; dr

> This isn't a single point of failure - it's a systemic crisis.

> One in seven breaches isn't a sophisticated external attack - it's someone inside the organisation accessing data they shouldn't.

> These organisations aren't browsing - they're buying

https://news.ycombinator.com/newsguidelines.html#generated

Even easier is just using an X server, if you have it set up properly you just need to run the remote app and the window pops up on your machine.

(I think terminal-based GUIs are neat just for fluidity of use- you can pop one open during a terminal session and close it without switching to mouse or shifting your attention away from the terminal. They can also be a nice addon to a primarily CLI utility without introducing big dependencies)

suspect this problem is essentially unsolvable. what possible method wouldn't be vulnerable to this? it's fine if it's just a sort of larp but if people think this could actually work... man

Your CED or laserdisc player needs to be smart enough to be able to decode whatever you put on it, which- in the era that they were relevant- pretty severely limited what you could do.