yk

The approach, detailed in a paper presented at the International Conference on Machine Learning, achieved 100% accuracy in identifying models specialized for CSAM generation.

I know some reasons for 100% accuracy in machine learning, first of all the test set leaking into training data. Or you just accept a silly high false positive rate.

When I was an admin I liked to joke that if you guarantee more than 5 nines, then you are an insurance company and you are planning to pay the penalty instead of actually fulfilling your promise, here the principle is probably the same.

Up to a conformal factor.

Seven LLMs were arranged in series. Six assumed another had read the code; the seventh read it and apologised.

And this is why management assumes that one can just automate software developers.

Pretty sure the reason is, that anybody who could actually do the changes thinks works for me, the alternative is a lot of work for little technical gain.

And besides once you start your tablet for Linux Projekt you have to touch everything, so that is a nice opportunity to finally refactor the wacom_debug_2 mess and pretty soon you're drowning in yak shavings.

There's something funny about complaining about cheating in a hacking competition.

Well actually I get it. In cycling motor doping, putting a hidden engine into the bike, seems more offensive than regular doping. I think this is because there is a continuum from eating well to taking supplements to injecting stuff, but having a engine breaks a fundamental idea about cycling. Similar hacking is about cleverly abusing the rules.

“Malawi is poor because its agricultural productivity is low” is closer to a tautology than an answer.

I used to make the same error. Thing is in the natural sciences this looks like circular reasoning, but in the humanities quite commonly things just hang in thin air. Case in point, the banker looks at the poor farmer and denies credit because the guy doesn't have capital, and the farmer doesn't have capital because he can't get credit. Thing is, both sides understand that.

Pull left-pad as dependency presumably.

dig manages to dig out ips for heise.de and tagesschau.de but not spiegel.de amazon.de and google.de However, dig @8.8.8.8 has still amazon.de cached, unlike 1.1.1.1 so perhaps Google to the rescue?

[Edit] After playing around with it, google seems to have at least some pages cached. After setting dns to 8.8.8.8 amazon.de and spiegel.de work again, my blog does not.

They own eBay + GME + some financial alchemy. If you aren't a financial wizard you should assume that the value of the financial alchemy is negative. (Because 99% of the time it is.) Now, what are the synergies of eBay + GME that outweighs the chaos caused by the merger and the finance stuff?

Senior figures (Smolin, Woit, Hossenfelder, Penrose)

Well one of the three is not like the other, three are very accomplished physicists, one is a youtuber who lies about the game to get clicks. (And we know she lies because she used to play the game quite competently.)

And sure enough they start talking about interpretations of QM.

No, but actually yes. Guardrails usually refers to a step in the inference pipeline where you check that it is consistent with policy while open weight models don't come with such a multistep pipeline. However open weight models are aligned during RLHF step, which means they will refuse to discuss overly sensitive topics. There are techniques to remove those, if you look for uncensored models on huggingface.

Remember folks, you are only allowed to laugh at their misfortune if you tested this month wether you can restore your backups.

My theory is, that a lot of security bugs are low hanging fruit for LLMs in the sense that it is a bit tedious but not that hard pattern matching. (Let's see the free occurs in foo(), so if I trigger bar() after foo() then I have a use after free, that should be possible if I trigger an exception in baz::init().)

If we go with that 15% of household budget, that would be something like 10k and people used to buy a lot less clothing back then than they do today, probably less than 10 items a year. Now, if you take 1k you can probably find a tailor who makes you well tailored trousers, and another 1k for a jacket. This is not how we shop for clothing today, the productivity gains go into fast fashion, being able to buy trousers each month.

I've seen an interview with a food stylist and she pointed out that when putting pins and needles into a burger, then you have to pay real attention to that burger because you have a really great looking burger, full with pins and needles.

Yes, and it took our ancestors 200k years. I would like to give our descendants an head start.

Kinda on topic, I thought it is a quite interesting article that gets better if you overlook the sensationalist framing.

So genAi ads can now be A/B tested by autonomous systems, to be shown on an social network for agents to be appreciated by ai agents.

On one hand, yay automatization, on the other hand, I feel weirdly left out.

Kinda like learning bash. The most annoying time was when I figured out how to send myself SMS via bash script.

Hey, it's the semantic web, but with ~~XML~~, ~~AJAX~~, ~~Blockchain~~, Ai!

Well, it has precisely the problem of the semantic web, it asks the website to declare in a machine readable format what the website does. Now, llms are kinda the tool to interface to everybody using a somewhat different standard, and this doesn't need everybody to hop on the bandwagon, so perhaps this is the time where it is different.

From the public comments over the last few days, my guess is they want a militarized version of Claude. Starting with a box they want to put in the basement of the Pentagon where Antropic can't just switch off the ai. Then some guardrails are probably quite bothersome for the military and they want them removed. Concretely if you try to vibe-target your ICBMs Claude is hopefully telling you that that's a bad idea.

Now, my guess is in the ensuing lawsuit Antropic's defense will be that that is just not a product they offer, somewhat akin to ordering Ford to build a tank variant of the F150.

Bitcoin did two things to this paper, first it demonstrates that Byzantine fault tolerance has practical applications, and second it demonstrates that anytime you have to deal with Byzantine fault tolerance the question is not "How do I verify this message?" but "Why am I trying to deal with those assholes?"

Probably. In previous years there were streamdumps available immediately somewhere (though figuring out where took me usually a day or so) and a re-live version on media.ccc.de a bit later. (Usually hours, but from time to time a day or so.)

Pretty cool, or more probable hot. Though I highly doubt it is something resembling a planet up close, it is more likely some kind of remnant from forming the neutron star that just happened to have the right size and ended up in the right orbit to show up in exoplanet surveys.

It's really hard for start ups to compete for VC money: "Hello Mr. VC I'm going to burn your money for buzzword" just no longer works, first openAi has an industrial buzz word generator, second their money burning plan scales much better than my money burning plan.

I already have 0 videos on youtube home screen, some combination of not being logged in, firefox privacy settings and ad blocker causes youtube to post a passive aggressive message and a search bar. I kinda like that Ui.

No, but it is amazing first sentence. Everybody goes, this story is specifically for me, I'm very special.

Maemo was an actual GNU/Linux, not just kernel with custom userland. Logging into a cluster from my N900 and having plots just appear on screen thanks to X network transparency is still one of the most futuristic things I have ever seen a computer do.

Pretty sure bullshitting is pretty universal. In particular the pattern were I, the mighty expert, insist that only I, the mighty expert, can decode the meaning of those deceptive others, and therefore you, the gullible rube, has to give me all your money so that I, the mighty expert, can keep you, the gullible rube, save from those deceptive foreigners.

Sounds really like a development environment problem, I mean if you can't handle that your language suddenly changes it's name in a not backward compatible fashion, how do you ever stand a chance to handle leap seconds correctly?