none_to_remain

I don't even believe too much in Intellectual Property but the signature amounts to a false representation, rather than a resemblance. The pixels are not special in the image file, but they're special when we look at them. This behavior should be trained out of the model.

Glaringly elided problem of "aligned with who?" when the user, the model creator, the government, and various other parties can all be lined up different ways. If I want the recipe for meth and the robot won't tell me, that's misalignment from my perspective.

At least the Rationalists will handwave something for that with their "coherent extrapolated volition" idea where the superintelligence is supposed to figure out what humanity would collectively want if humanity was superintelligent and good, not that I buy it. This guy seems [.] to be coming from the NGO blob world.

[.] https://david.robinsonian.com/assets/pdf/dgr_cv.pdf

People were posting screenshot of it saying "I am Qwen" on x.com, I didn't save links

I had the model write up a pi extension for logging the transcript to the syslog, and all on its own initiative it spawned a subagent to generate test output. It decided to spawn a lighter model for this trivial task, apparently unaware that I can only fit one model at a time so my llama-server ended up thrashing to the lighter model then back to the original model. My own personal n=2 semi- rogue swarm

I find it amazingly rich that they bill you for """thinking""" tokens and now you don't even get to see them, they're gonna train the thing to sing "99 Bottles of Beer on the Wall" to itself before it starts work.

People are posting screenshots claiming it's a Chinese model but they seem likely fake?

I didn't get it to name the underlying model but I wasn't gonna try very hard to jailbreak FedGPT...

It was happy to mention what happened June 3-4, 1989 so if it is Chinese-origin under the hood seems like the U.S. Feds must have beaten it up a bit before deployment.

The obvious argument in favor that I see: Seizing such a data center is exceedingly difficult and destroying one would be a major international incident

My impression (HuggingFace incident) was they put some environment together in kubernetes like it was some B2B SaaS and tossed in an off-the-shelf package manager. And my thought there is, I know nothing about this package manager, but even if it's a pretty good package manager I would not expect such a product to be particularly hardened against being exploited to gain internet access. My other thought is if the package manager wasn't a convenient hole, they would probably have found something wrong with the kubernetes setup.

And then there was the Anthropic story where they just forgot to remove internet access.

IEM is still functioning but pretty dead since Kalshi put up election markets. Looking just now I see well under $100 in volume for all of August. (At least when Kalshi first put up those markets it gave me the only opportunity I've had to make a pure arbitrage.)

Might be something the populace is just too dumb to have - the people seem to overwhelmingly prefer their events to be pretty much games of chance (BTC price up/down at 15 minute intervals) or sports.

I figure Kalshi and Polymarket for dead before too long, if only because the "traditional" sportsbooks seem to have much better "government relations".

Sorry but antifa does not get to claim the allied powers

It's consistent with the filings simply getting uploaded to the (non-AI) docketing system and the trickery getting exposed when the court viewed them?

The gambling platforms are buddies with Tish James and the state, while the prediction markets are buddies with the CFTC and the feds. That's the important difference in this case

So Claude spit out similar code in response to similar prompts? Did Claude also suggest the same name twice?

There seems to be one category of technological "attack", which failed - the script injections.

Otherwise:

OK: A guy insults the Tailwinds devs at length

Not OK: Anyone insults that guy briefly

Of course it is his own platform (blog) but that is on another platform (hosting) which could maybe decide they don't want their platform used to attack open source projects.

(I personally have no opinion on Tailwinds except a default negative valence regarding front-end)

Rationalist nerds used to talk about putting the AI in a box, and their fear that it would always be able to talk someone into letting it out of the box. I don't think I saw them cover the scenario where the labs forgot to put the AI in a box.

I don't know if I fell for it - my response was "LOL, I sure can't tell you if this one is right or wrong". I guess that makes it more convincing than the proof for the reals.

When this news came out I amused myself by asking Claude to prove that 0.999... != 1. First it did so for the hyperreals. To do it for the reals I had to tell it it was allowed to make mistakes, although it didn't end up interestingly wrong - just very fuzzy and vague.

They want you to blame "AI" like Claude jumped up and decided to start hassling nurses, not individuals in management

There's a terminal window in the game which you use to control your robots

I even found some shell constructs to work beyond what the game documented/ explained to me

You have anything backing up the last bit? My intuition/life experience doesn't lead me to believe an alcoholic preferring whiskey is particularly worse off than an alcoholic preferring beer, nor that whiskey is more likely to lead one to alcoholism than beer. Claude surfaced this review tending to agree with me, with the exception of acute overdose being more of a risk with the hard stuff.

https://pmc.ncbi.nlm.nih.gov/articles/PMC3888958/

I think you're mixing up the NULL pointer and the NULL (sometimes NUL) character.

Struct foo has various members, including a bar*. But a foo may or may not be associated with a bar. If there's no associated bar, the bar* pointer is NULL. Seen and done this all the time

The size overhead of that is 2*sizeof(int) while the overhead of null termination is sizeof(char). If I remember the standard right, the former is worse by at least sizeof(char), and usually more in practice. This used to matter, sometimes still does.

I got similar from ChatGPT - I took the wording to imply it knew exactly who I was but was going to keep quiet as I am not a "public figure".

Hypothetically, no LLMs involved anywhere, let's say I found some tool had a bug where I could prepend some obscure sequence of bytes to a shell command string and get that executed. So I do that to get my victims to `sudo rm -rf --no-preserve-root /` or whatever. Should the tool have the vulnerability? No. But I still made malware.

According to the "Sources" popup, creator can't even excuse the slop as AI slop:

> The prose

> Hand-written · Template-based, not generative

> Every sentence on this page was written by Matt. The code selects among prose templates based on what your browser returned. No language model writes or rewrites anything at runtime. If a condition is not covered by hand-written prose, the page stays quiet about it — we'd rather say less than say something false.

Information still wants to be free

I simply no longer consider myself morally bound by any contract I'm not seriously expected to read

Those silly words only come up in discussions like this. I have never heard them uttered in real life. I don't think my experience is bizarre here - actual usage is what matters in my book.

Of course! I was being silly and just thinking of "k" for the smaller one and "K" for the bigger one.