ls612

If you ever wonder why Anthropic is going all out to try and fearmonger their way to an open weights ban, this graph will provide all the answer you need.

AI doing something illegal and AI being illegal are very different things

The entire concept of an illegal AI is something I hope never comes to pass.

I bet a small 1B model trained solely on how to play the strategy game would do better than current AI solutions for these games and that would probably be able to run even on integrated GPUs.

Unfortunately for it to be sustainable it would need to be an LLM with VRAM usage probably under a couple gigs max so it can run locally on the gaming PC. That said it may well be better than current AI in strategy games.

Claude is very capable of making your code run faster if you ask. It knows every profiling trick in the book.

You underestimate just how much software humans can interact with in any given second. ;)

The lump of labor fallacy is a fallacy not real. Phrased another way, demand for software is more or less infinite so if coding agents make everyone a 10x engineer eventually we will easily find use for 10x more software.

After everything in the past six months do we really think that humans are better than coding agents at writing secure code?! This announcement honestly makes me more confident not less in their products.

It’s not a 0day it’s now 5 days and counting since disclosure. Nobody is gaining value sitting on an exploit for a vulnerability which is being actively patched.

It is a price signal. They expect to lose the money they want to show that nobody has an automated swarm just pwning every server running PHP connected to the internet. At least not yet, on day 5 and counting.

This is going to end up at the Supreme Court probably sooner rather than later. The only reason it hasn't yet is that the justices usually wait for two circuits to rule definitively in contrary ways before taking a case, but Kavanaugh, in a statement respecting denial of cert for a similar law in Mississippi last year said that banning minors from social media was likely a first amendment violation. I would expect given that there are precedents that the first amendment protects both the speech of minors and anonymous speech that in the end these laws will be struck down.

Presumably if it knows how to find and fix bugs it can be told how to exploit them. But who knows what it will actually be capable of doing in the real world as opposed to Kiki’s benchmarks

I mean they say they will release the weights by November 4, so presumably after that it won’t be hard for people to fine-tune it to be better at cyber right?

It was always OK with most things I asked until the whole Mythos freakout but nowadays they have silently pulled the classifiers way back, really now bio is more gatekept than cyber stuff. From what I have read so long as you have the models do the decompilation/disassembly using an MCP so they aren't hunting through the actual binaries by hand (that can freak them out) once you get the decompiled code they don't complain at all.

It is one thing to go after a couple emulator teams where expertise is concentrated, that at least is a plausible bottleneck. But open source did open source things so forks appeared and development has continued in various forms. And now with AI the expertise is available to everyone so we are getting into “suing individual Napster users” territory. That is not to say that Nintendo won’t make some examples of people developing jailbreaks or doing recomps but it won’t fundamentally stop it from happening.

Nothing described here seems illegal as opposed to merely annoying? I guess on some level “everything is wire fraud” but that was the case even without AI.

This is very similar to https://electionbettingodds.com/ which has been doing this for decades at this point.

To show off? Like whatever you think of it the market for supercars is rich men who want to show off so Tesla is playing in that market.

This is completely normal. Banks underwrite large loans and then package a bunch of them in different ways to sell to different investors.

There is all the difference in the world with a rogue agent going off and choosing to delete critical files and a human using an agent to implement a suboptimal backup strategy, and then being surprised pikachu when it fails.

And this is different in what substantive way to the western system? The vast vast majority of ordinary Chinese will never interact with their national security state. This reads as you saying I’m right without saying I’m right.

It will provide Anthropic a great blog post next month to push banning open weights.

I’d suggest that one of the key innovations of authoritarian control over the populace in the 21st century vs the 20th is that surveillance and compute means the authoritarian states don’t need to use the brute force of the legal system against the average person to keep power. They can focus only on key targets while not molesting the broader population in the same way the Stasi et al did. I’d go so far as to say for the typical American and the typical Chinese their experience of the legal system is virtually identical, in that they don’t experience it. But for those targeted by the system, the precise mechanism may vary but the certainty of the outcome does not.

And I’m saying that three felonies a day and the way China does it are fundamentally the same thing with different window dressing.

Let’s be honest it’s funnier as written.

That was precisely my point. “Show me the man and I’ll show you the crime” is the fact of how western legal systems work, the rule of law is a tale that they tell to impressionable schoolchildren.

This is a feature not a bug in the system. But for a lot of idealistic young law students learning that will make them deeply uncomfortable

Facebook successfully becoming a frontier lab was not on my bingo card for 2026, but the more competition the merrier I guess. We will have to see just how benchmaxxed this is but beating Astra and Fable is not easy.

It has some major technical flaws which hold it back even on modern hardware. Something I’m sure a model can fix.