infamouscow

The problem is SCOTUS has already ruled software is protected under the First Amendment, and the landmark cases that establish this precedent applies perfectly to model files.

A large number of security issues are not backported to stable or LTS kernels.

A patch containing a new feature which tangentially fixes a security issue is merged to mainline. It will not get merged into stable or LTS (by definition).

Exactly. Here's how that meeting went between security, compliance, legal, and CTO went:

security: zomg this is existentially bad we are leaking our fn'ing emails!11!

compliance: can we mitigate this problem?

legal: what can we do here?

CTO: don't care, we're not losing all this productivity because our emails are being leaked, have a good day

This is a problem with LTS distro kernels in general. You're trading the feeling of stability at the kernel level for security updates.

The biggest hurdle here is not changing how product/engineering work, but getting the rest of the business to shift as well. Most of these ceremonies came about because a stakeholder wanted to know when feature X is done, and instead of waiting, they poll endlessly for daily status updates as-if that speeds up development.

This shift will be much harder for B2B than B2C.

The current framing of US politics is basically "vote for us because the other party is an existential threat" and data center proponents have adopted a similarly retarded position "we have to build data centers or China wins."

For this framing to work, you need a base of reliable supporters already all-in on your cause regardless of how you actually perform, and the data center proponents have done virtually everything imaginable the last four years to drive people away. Why AI investors are shocked (shocked) is really beyond explicable reason; American on both teams have never responded well to this kind of brain-dead gambit.

They're very loud and run 24/7.

All it takes is for one state to grant AI narrowly-scoped medical licensing to treat various things human doctors are allowed to treat telephonically.

It makes more sense if you view themselves collectively as Claude, and their IPO is a prompt that ends with "make no mistakes."

poop.bike

Some people are evil, e.g., Hitler. All the moral grandstanding in the world isn't going to help if you're invaded and they slaughter you and everyone with your morality out of existence.

I'm not being condescending or rude or otherwise out of line here saying your position is genuinely stupid.

The answer is you can't. They did not reason themselves into their position, and you can't reason them out of it. It's irrational to have rational discussion with the irrational.

This is why derision is a valid philosophical debate strategy if a person refuses to engage in rational discussion. To label GP as an R-word is a social signal and warning to others not to behave similarly. Derision also promotes healthy skepticism of new arguments you hear, lest risk falling into the same trap and have oneself become a laughingstock.

This was true before AI too.

In practice, lawyers generally don't ever disbarred except if they lie to clients or steal from them.

Many organizations just want to diffuse responsibility away from themselves and will gladly sign a contract to do so. Paying a vendor is infinitely more affordable than the risk of trying to bring a vendors' core competency in-house.

I think the SaaS-pocalypse is going to surface more from new AI-first SaaS companies entering the market and undercutting incumbents in a way that's difficult to compete with.

The role of science is to advance human knowledge and understand what's true. The role of engineering is to bridge the current world to something new discovered by science. Until recently, the new things were the domain of scientists, so many of them understandably feel like their existence is threatened by AI.

Another aspect I haven't seen anyone talk about is how big tech poached a generation of millennials from grad schools ~15 years ago. I think a lot of those people feel like this is a comeuppance for unnecessary or bad faith gate keeping in academia.

You should change the title.

Most people will read this headline and assume Cloudflare is intentionally trying to squash open-weight models. Given how superfluous the model is to the content, it's easy to believe that narrative.

If that wasn't your intent, you really ought to change the title ASAP. Time is ticking before tech journalists and social media decide to run with that narrative and make your public relations department sad.

FDE allows execs to covertly circumvent their internal bureaucracy to ship faster. It will be interesting to see what execs make of that data.

The data you put into a SAT solver has substantial effect on it's performance in practice. Real world inputs tend to be much nicer than pathological ones, so I don't think there's much practical application, but I also haven't read the paper yet.

I'm also not going to cheer on database projects that can't be 100% transparent and honest.

Who the hell is going to trust their data to a people can't even be honest on benchmarks. What do you think will happen if (or when) there's a data corruption bug. What evidence do you have to think these same people will behave any better?

They failed themselves, we're just watching it happen.

If pgrust were backed by a company that advertised these benchmark results, you would have enough to claim fraud in a civil lawsuit. Literally snaking oneself out of accountability after being caught is dishonorable. Children do this, not adults. Seeing adults act in this manner is pathetic.

This is an accurate description of one's bad behavior.

If you believe these word choices are inaccurate, we can discuss, but mischaracterizing the reply as "aggressive" is just bad faith (and against the rules).

The thing with models is they're trained on the entire corpus of humanity (more or less). So while you'll get different answers from people, the models can look at all existing answers in their corpus and evaluate each answer based on all other evidence in the corpus.

In principle, a scientist could do this too, but models can do it without the manual labor.

You're presumed innocent unless proven guilty.

To be clear, I'm not saying a one sentence Slack message can't be ignored. I've coached these folks about how they might need to make multiple attempts to get someone's attention and they're always receptive.

But that's very different than when an engineer stays up all night investigating a very serious issue, writing it up completely, and offering multiple solutions in a wall of text that can't be brushed off in Slack. The engineer makes multiple attempts across multiple weeks and communication platforms to get their attention, and it's ignored. Not "we're busy but will get back to you when we can." Just completely ignored.

A high performer (and really anyone) should not have to resign themselves to office politics to workaround this kind of dysfunction. It's imperative to fire such managers so that you don't lose the trust of the people doing the work.

What I learned managing high performers is a disproportionate number of them had very hard childhoods and carry permanent scars that make this behavior destructive. These are the people that generally get ignored in every aspect of their life because their brain is operating much faster than the rest of society and get mistaken for being insane in every facet of their life. So when they experience this kind of behavior in an environment that's supposed to reward them for their ideas, it's stultifying. Those high performers are not going to continue putting effort and energy into novel ideas if they're being ignored.

The solution is pretty simple: just fire the manager.

Most people are unimpressive and mediocre.

One of the things lawyers often forget is laws are just made up, and not everyone has to abide by them.

Considering how China is not subject to US jurisdiction, there's no downside risk in simply defrauding the USPS with fake documents to get around this.

I don't think that's ever going to happen when models still hallucinate and prompt-refuse, especially since one of the two SOTA model makers insists on aligning their models toward a progressive ideology.

Just imagine what would happen if JP Morgan did this and the model decided it needs to dump their position in AI data centers. Or maybe it's a health insurance company that fires all of the claim deniers for being unaligned with Dario's ethics.

Agent swarms have figured out how to mostly work around context window limitations. There's no reason the same ideas can't be applied on a smaller scale, at least in principle.

Hire them, you're gut is questioning your rational brain for good reason.

The future of OpenAI and Anthropic is contingent on a lot of factors which can (and maybe will) just vanish overnight. If China "invades" Taiwan, or if the data center buildouts completely stall, AI is going to be in a very precarious spot.

It's easier to pass a bill for video games than try to cram movies and music in the same bill.