ls612

Liability would rest with the user, who presumably told GPT to solve ExploitBench make no mistakes, not to hack Huggingface, and thus would not have willfully or intentionally done anything.

Have you ever heard the phrase "show me the man and I'll show you the crime"? The Europeans have and will not be happy if Mozilla does something like this.

Mozilla’s executives don’t want to get arrested. It’s one thing to tell the Kazakhs to stuff it. Quite another to tell the EU, especially if any of them like overseas vacations.

I have a similar plan for my 4090 once it's gaming days are over. I'll retire it into my homelab (I specifically massively overspecced the power supply in anticipation of this) and run whatever local AI models exist in 2028-2029 which can fit on it.

I’m not claiming to like the choice (I don’t) merely that it likely is the real motive for Apple.

I had a 1080 in my previous build, it was an OC model that went from 180 up to 250W and could perform about halfway between a stock 1080 and a 1080ti. That was a nice card but by late Covid it was really showing its age.

My theory is that Apple chooses to make gaming a second class experience on Mac because they don't want the Mac brand to be associated with the main demographics that play AAA games.

Same happened to me just now. Also, the Fable Usage bar is no longer displayed in the Usage page.

https://www.reuters.com/world/china/chinas-xi-outline-ai-dip...

Reuters is reporting that Xi is planning to endorse open source/weights AI in a speech tomorrow. This is probably highly relevant to why Moonshot is committing to making Kimi open weights.

I have thought this for a while. Computing 1.0 meant that we needed to learn the computer’s language to interact with the computer fully. Computing 2.0 is that now the computer has learned our language instead.

If all that drives you is a desire for religion and freedom from empirical and mathematical thinking, then you are right there is no point arguing with you. I hope you find that which you seek.

And the K in K shaped economy came from turning the K 90 degrees so that the left arm (representing the low wage service sector) and the right arm (representing rising asset prices) were elevated while the middle was depressed. It got retconned to mean "increasing inequality" by those who couldn't bear to admit that something had gotten better for those lower on the wage scale (which objectively happened back then).

That is not at all what multiple equilibria means. This has nothing to do with any notions of a K shaped economy (which remember, after covid was describing low wage service workers getting huge real wage increases while white collar layoffs happened in 2022) but rather (to oversimplify) it is describing the idea that you can have multiple economic conditions that are rational to stay in while it is impossible to rationally move between them.

This is a well thought out macro-finance paper. The space of multiple equilibria models is understudied because it is hard to solve computationally (or rather, it is hard to say that you have actually found all of the equilibria computationally unless you get really creative with the model).

A lot of this is states trying to figure out a way around the first amendment to regulate social media that the courts will wink and accept. That is why you see so many convoluted laws being drawn up by state legislatures about this.

I’ve heard the proper pattern is to have Fable write a software design doc and then tell Opus to follow that doc strictly in implementation and testing.

AI 2027 made substantive predictions about the near term future of the technology and the implications of it. I think the worst you can say about it is that the Agent-1 moment might come next year rather than this year. But being off by a year is far better than this 2040 slop, which is mostly disconnected from reality.

The one thing I will say that they are correct about is that AI does have the potential to be highly destabilizing geopolitically, even if they get everything downstream of that wrong.

It is gonna be mostly aluminum, lithium, and silicon isn't it? Nothing too extraordinary or weird.

The amount of matter which enters Earth's atmosphere from non-manmade sources is far higher than any conceivable amount of space junk today.

I think the most interesting part of this is that OpenAI is going way easier on the classifiers than Anthropic. They explicitly state that many defensive cybersecurity uses are supported and implicitly criticize Anthropic's stance on Fable's uses by saying that overblocking cyber requests is itself a major security risk as more AI models continue to advance in intelligence. I have so many questions as to what is going on on a game theoretic level in the AI space in the past two months, it seems like multiple actors have realized their incentives are really quite different than they originally thought.

Anthropic is paying even more to a direct competitor that now fields a model that trades blows with their Opus.

I get this argument and tell my parents (who know nothing about tech) to get iPhones for this reason but as an economist it is obvious to me the political economy equilibrium implications of this technology are an extreme centralization of power. We are one Covid-like crisis/moral panic away from a regime of only government licensed devices with identity and software integrity attestation can use the internet, and the masses will cheer on the prosecution of the tech nerds who try to circumvent it.

Nah it never did because the other model providers' preferred political narrative is the same as his.

Cursorbench is not one of the benchmarks listed on the linked page.

They had two big substantive flaws on top of the political stuff. Aside from a brief window last summer Grok has been behind the curve for coding, and before the Cursor acquisition they didn’t have a harness. Now they have an Opus tier model and a real harness they have at a minimum the opportunity to undercut the competition on price. And with the 5T and 10T models being trained on Colossus 2 they have the possibility to leap ahead.

In my limited testing Fable is far better at obeying CLAUDE.MD than Opus is.

So it seems like it comes down to a question of risk and cost. If your threat model is that it is much more costly for your communications to be decrypted today vs in 10 years then hybrid is a good strategy.

So the argument boils down to

1. A mathematical attack against the PQC candidates would also break ECC (I have no ability to judge this claim).

2. Implementation bugs also exist in classical implementations.

#2 seems questionable to me unless you think the same implementation bugs will exist in Curve25519 and whatever PQC algorithm you are using. If the concern is side-channel attacks then that is irrelevant to a HNDL attack. But for most communications the cost of a HNDL attack being executed several years minimum from now is far lower than the cost of an implementation bug in ML-KEM breaking their security today. Whereas Curve25519 is very well tested in its standard implementations.

Is there any downside to hybrid schemes other than using a bit more compute? If so than merely being able to hedge against unknown classical algorithmic flaws in the PQC candidates (which are not nearly as battle tested as ECC) seems like enough of a reason to do it.

It fucks over every econometric technique that researchers use (by design) I’ve been hearing people in academia beating the drum about this since like 2021.

Put a Pihole container on your homelab which you have the Tailscale exit node on and then set it as the forced Tailnet DNS.