I do like the Potemkin Village metaphor for things like the website in question. Given how little effort something like this takes, I imagine there are a lot more of them than I've considered before, and given the way search engines operate and index, and LLMs scrape relatively indiscriminately, it seems to have implications that go beyond Google's stupid summarized answers at the top of the results page.
As another American, you understand that reasonable taxes applied in a progressive manner (with those who have benefitted more from society paying more) are what allow individuals to enjoy the freedoms that you want, right? Because if we don't all pay in according to how much we have, then you're free to make your own decisions until all of the people below you rise up and take back what they have been denied. If you don't believe me, look at essentially any part of history.
Look, our culture does not reward excellence, so those people are doing the least possible to not get fired. That's kind of the whole American thing in general at this point. Nobody is doing great outside of technology workers, and even then, only some of them. You're not going to have a great experience in general from people here. There are always outliers, but those are individuals who have found some internal motivation to go above and beyond, and it's not a systemic thing.
How difficult would it have been to isolate that problem if you didn't already know the SQLite subsystem it was in? This feels, to someone relatively ignorant of the SQLite / Tailscale / Antithesis architectures as a "hindsight is 20/20" kind of thing, but I'm open to learning more.
It was coined by Apple engineers who would agree to whatever Steve wanted them to do while Steve was at their desk, and then when he left they'd be distraught over what they'd agreed to.
Unless you're executing these commands in a loop over a large number of items, or the item itself is gargantuan, it's almost always harmless.
Personally, when I'm exploring, I build a command line iteratively. Cat the file to see the content, pipe to grep to get the lines I want, sed/awk/cut/etc to finagle from there.
It would be nice if inference could somehow perform token generation using "contractions" of "fluffy" tokens, where combining those tokens doesn't decrease nuance but provides additional efficiency. That may already be happening - I haven't looked at the most modern methods of inference in a long long time.
I don't know if it's fair to say that a tokenizer is being less efficient if it generates more tokens per text. I think it's more fair to say the tokenizer is more nuanced. The question is whether the additional nuance permits better model output, which could justify the additional token cost in inference.
Oh yeah, well that may be. Once I got X working in Slack, I did boot into that, but then again, back then I didn't know how to recompile my kernel and the machine did this interesting loop where it would try a few different drivers until it got to the one that worked with my matsushita CD drive.
It doesn't take away joy. It does cause me to be slapped in the face with the "full" feeling while I am in the middle of eating, which is like 15 minutes earlier than I used to get it. I don't overeat anymore because I would be literally miserable if I did. It would be like force-feeding myself. But before I get that full feeling, food is excellent.
Oh hi, I'm a heart attack survivor who had GLP-1 prescribed by my cardiologist for its heart-protecting benefits. I told my cardiologist that I wasn't losing weight as fast as I thought I would, and he said, quote, "I don't care about you losing weight. That's not why I prescribed you this."
So I'm not sure your first sentence is universally accurate.
I learned this from watching House. I am certain that it's true, but I also kind of wonder how much backpressure there is against this kind of thing because the scanners are also insanely expensive to run per minute.
>I suppose one solution would be to completely vet the training data such that nothing deemed "dangerous" exists in the data, which would be a huge effort.
I can see how this is tempting, but I suspect it would yield a naive model. I think the only way to improve this is to use a model that is legitimately advanced to support the concept of empathy, which may allow it to recognize others as being separate from itself, similar to how toddlers develop this sense (https://blog.lovevery.com/skills-stages/empathy/)
I think it's a side effect of the Transformer architecture. The worldview where all input is equally trusted, and there's no concept of "the other", makes it hard to build effective guardrails where some input is trusted and other input is not trusted.
It seemed obvious to me that this was bearing stiction and that manually rotating it during the start allowed the fan to spin on its own after that, but I could be wrong and maybe the fan was dead entirely?
That's not necessarily true. Even spaceships in LEO will perform temperature-driven rolls so as to distribute heat and radiation. I have to assume that long-term ships like interplanetary transport will do the same.