Notices by Cassandra Granade 🏳️‍⚧️ (xgranade@wandering.shop)

I've talked a bunch about exit strategies on here, but mostly in the form of knowing where the exits from centralized corporate platforms like Bluesky are.

It's also important, I believe, to know what possible exits you have from the corporate OS duopoly of macOS and Windows. That's not to say "go use Linux!1!!1," it's to note that Cook gave a massive donation to Trump and that Windows is quickly getting more and more AI-laden. At some point, you may not have a choice… planning ahead can help.

I have come to believe that once that happens, it's catastrophic and almost unrecoverable... if organizational decision-making is impaired, everything becomes suspect and dysfunctional.

If you're running a company, nonprofit, government agency, whatever it is, your decision-making process is *the* thing that allows you to function. Once that's gone, you're cooked.

I read this news in mind of my belief that organizational structures that promote people on the basis of them not understanding the limits of a currently in-vogue technology will suffer an inversion of expertise.

That's not even Peter Principle, it's that a lack of technical competence concentrates in more senior positions and in positions where critical decisions are made.

https://mastodon.social/@mhoye/114473063306352182

I just cannot put myself in the headspace of a programmer who hits a button that pipes word mcnuggets directly into the terminal and runs them.

We've known that `curl | sh` is bad for a long time, how does that get better if the thing you curl is a badly designed RNG?

(And yes, I've used `curl | sh` and even mentioned it positively in my gemlog yesterday. It's bad, but on some occasions the best of a bunch of bad options. That's not the case here, though.)

The whole thing flies in the face of the kind of engineering discipline that leads to formal specifications, IDLs and APIs, reproducible builds, linters and fuzzers and memory-safe languages.

To throw that all out in favor of sparkling `cat /dev/urandom | sh` just breaks my brain. Why do people think that's an OK way to make software?

Search engines have gotten so much worse. I keep trying to search for one simple thing in both DDG and, as a fallback, on Google, but I keep getting the exact opposite of what I'm searching for.

I've been doing this almost as long as the public internet has existed, I know I don't suck at finding things. It's just that search engines have gotten really, really bad.

This might be what finally pushes me hold my nose and try out Kagi. I hate their AI marketing so very much, and I hate giving money to an AI company, but they're *all* AI companies now, and I don't know how to *not* give money to AI companies without moving to a remote harbor island, renouncing my internet connection, sealing all my computers in concrete, and illegally dumping them offshore.

I have said my piece about Sabine Hossenfelder, and I am not about to ever take it back, nor defend someone who puts that much vile transphobia out in to the world.

It is also true, however, that this line of argumentation against her gives me significant pause.

Academia is not science, and the enemy of my enemy is not my friend.

https://mstdn.social/@david_colquhoun/114449610404687235

There's not a lot of daylight between this kind of religious ecstatic fervor for AIs and the NYT article that concluded that, because ChatGPT output a string containing "I am alive," ChatGPT must be sentient.

https://merveilles.town/@lrhodes/114450638623375146

@whitequark To be sure, that's why I emphasized from the second toot that I wished there were more resources available — I understand that reading, and I definitely don't intend to encourage more unpaid labor.

It's absolutely fucked up that the companies that do A/B testing don't give a shit about the ethical consequences of such, and the people who *do* give a shit don't have the resources to back that care up with aid from specialized expertise in ethical experiment design.

@whitequark And like, if an IRB is just acting as a gatekeeper? That's not helping anyone do things ethically, and it's not helping any potential or actual users. It's concentrating power, and that's it.

I don't know what the right structure is for providing expertise to people trying to do their best while also telling corporations running completely unethical uncontrolled and nonconsentual experiments to fuck off.

@whitequark But absent that structure, I absolutely recognize that people are doing experiments on actual humans, both to maximize the profits of unscrupulous corporations, and to learn how to better benefit other people — I encourage that that experimentation be done with all the empathetic concern and attention to ethical conduct that implies.

@whitequark I mean, fair; I'm not sure I agree (indeed, or else I wouldn't have said what I did), but I definitely take the critique.

@whitequark I do hope that's something that you don't find deranged, as you say; I value your opinion and expertise in general and on this stuff in particular. Whether you do find it deranged or not, though, that's more or less where I'm coming from?

I think tech as an industry (and yes, that includes my own time in such) has normalized a degree of nonconsentual experimentation that's now causing huge problems.

@whitequark To be clear about my intentions as well, no, I'm not trying to be inflammatory nor is any of the above intended to incite anything.

Rather, I am incensed that AI chatbots and the like are the logical extrema of the kinds of experimentation that have been normalized in the tech industry, and that that extreme has caused significant harm. I posit that being clear about the roots of those ideas is helpful in understanding the extremes.

@whitequark To wit, I don't think there's an inherent conflict between collaboration and experiment. Indeed, a participant in an ethically designed and conducted experiment is there on purpose, at least nominally because they agree with the goals of the experiment. That strikes me as quite collaborative, which is awesome in my book.

But also, experimentation is something that can go wrong when done without proper knowledge; I would love for there to be more resources to help.

@whitequark Maybe put differently, the same kinds of arguments and excuses used to justify A/B testing are now used to justify tweaking LLMs "based on user feedback." OpenAI has even admitted to that recent versions of 4o encouraged people to think of themselves as religious prophets because of how they interpreted user feedback.

I posit that understanding why A/B testing can both be useful and harmful is itself useful in deflating OpenAI's arguments.

@whitequark I mean, fair, you can think me deranged if you like... there's not much I can really do about that?

To be very clear on my part, though: none of that is to say that telemetry and the like are wrong. Rather the opposite, in that I wish there were better resources made available to help use it ethically. As you say, it's quite useful and can help make informed and useful decisions.

@whitequark There's a lot we understand about human behavior, both at macroscopic and microscopic levels, that comes from experimentation. One-way mirror studies, A/B tests, and pretty much any kind of user testing fall under that, in that you're providing stimuli to a human being and seeing how they respond.

The understanding we glean as a society as a result of those experiments is immensely valuable. What I'm calling out is doing those kinds of experiments by accident.