Yet somehow, Oprah returned.

Shit, I remember it being 50% of people are above average intelligence. Guess I got split into the stupid reality :(

Also a great way to learn Dvorak. Memorize the key combo to switch between the two depending on how detailed you need to be in telling them they are wrong, but as long as you keep making yourself spend a little more time on the less familiar layout, you’ll eventually become fluent and won’t have to contort your fingers as much regularly to type quickly.

Though typing games can help, too.

For when you need to do an assignment due the next day but your roommate keeps yelling at you to shut the fuck up already because they are trying to sleep while you slowly dictate the introduction to your 5 page essay, which then gets you kicked out of your class because you missed removing a few of the "SHUT THE FUCK UP!"s that your voice to text helpfully added for you.

Which just shows how fucking stupid this current LLM-based AI approach is. There isn’t a way to differentiate between data and meta data or instructions. It all just gets shoved into a prompt that might end up the length of a short novel by the time all the context has been added and read operations have finished. A tool so sensitive to its input that adding a period at the end of an instruction could completely change the output it generates, even with temperature (randomness) set to 0.

I’m not even sure this can be fixed. Like, even if they they try separating the instruction input from the supporting data input, LLMs don’t follow instructions in the first place, they just predict text and having instructions in the context can strongly affect the output it generates. Meaning there are no instructions to separate from the data; it’s ALL just data and platforms like Claude Code just give it the ability to do things with that predicted text that hopefully follows your instructions and uses your data rather than the other way around.

I think we’re stuck in a local minimum of an optimization problem for AI because an LLM is much easier to make than a more reliable form of AI. You mainly need to throw a lot of text at it to train. There’s probably other tweaking that goes into it, like a way to do more training using user thumbs up/down feedback, but it’s just the big data approach of soaking up all the data they can find and just throwing it at a blank statistical model and see what it spits out.

If we want something like the Star Trek computer, I’m pretty convinced at this point that it’s going to take a completely different foundation, but the industry is currently stuck on improving LLMs.

I’ve had some great “conversations” with LLMs about their weaknesses. Sometimes they bring up angles I haven’t considered or give more detailed specifics on something I only suspect or just have a high level understanding about. They are often trained on conversations about themselves (the LLM you’re interacting with specifically as well as LLMs in general).

Though be aware that if you are talking with an LLM about how it works, that whatever discussions it uses to predict its response had to have happened before that particular model was released. So unless its weaknesses are placed into a hidden prompt, you can’t discuss a model’s specific weaknesses with itself. It can only discuss previous versions of itself or general LLM weaknesses from its training data, though it could do a web search about it for you.

Calendar stores have whole aisles that also beg to differ.

School playgrounds, too, but they wouldn’t know what the question means and would start talking about pokemon before returning to their game of tag.

So what I’m seeing is a chance for men to confuse the data by installing a period tracking app, tracking fake periods, then skipping a few months and resuming, which might make some asshole cops doing a particularly asshole investigation waste time and resources. A parricularly ambitious man could even set up multiple accounts to pretend to be a whole slew of briefly pregnant women.

Or understand the difference between software that mines data and software that just does what it says. If it advertises cloud features, then it’s probably a data mining app. There’s firewall apps that make any other app need permission to use the internet at all that you can use as a gatekeeper, but then you need to adopt the mindset of wondering why an app wants internet access instead of just clicking “allow” so that things work asap.

But yeah, pen and paper or even a spreadsheet not dedicated to tracking periods are good options if you want to avoid worrying about all that. Only thing I’ll add is that it applies to a lot more than just period tracking apps and IMO is as useful these days as knowing how to do basic car or home maintenance.

Even if it uses digital over the USB connection, I prefer that toslink is only one-way so that my soundbar and PC each have their own volume controls instead of sharing a single volume.

Set up a giant hamster wheel connected to a battery with a meter. Ask for a few volunteers to come on stage to run it. Tell them they have about 7 seconds between stopping the wheel and the battery running out, so that’s how fast transitions need to happen for the show to keep playing.

Some Egyptians think their massive spoilers help with handling their chariot but they just introduce drag for no real benefit because the power comes from the horses in front of the chariots and allowing your chariot to slide to the side as your horse pulls it through a tight corner makes it easier for the horse to maintain speed (or horses if you’re a fucking show off).

You know how many snapped hitches I’ve seen on chariots with spoilers vs how many without? 3 and 1, which should show a clear trend. And that one without a spoiler happened because she was trying to go way too fast over cobblestones for some reason.

First rule of kinks: always assume that everyone around you is fully into it.

Ah shit, I forgot the first rule and ruined your shame kink, didn’t I? I mean… It’s all your fault for not saying anything, you should be ashamed!

Oh sorry, I was trying to help get you off by acting irritated, but I guess you don’t have that kink yourself.

There is feedback that makes it back to their servers: the next prompt.

Though from my understanding, a lot of the agentic tools do a lot of prompt generation, so that would be a weakness of just logging everything and using subsequent prompts to evaluate earlier ones. They’d have no idea whether each prompt is coming from an actual user or another LLM.

Sounds like a bunch of timing attacks could be rendered useless if access to an accurate timer required special permission. And without the permission, it either limited the resolution or added random jitter to any timer APIs.

Irritant play? Like this message you just posted?

The incest issue is actually an artifact of the religion initially not being monotheistic. Each region had their own religion and they were all treated as equally valid (at least by early Jewdaism), so the creation story was basically “God created a paradise (and everything else) and two people in that paradise, but then they got kicked out and had to go live with all the other people made by all the other gods”. Adam and Eve had Cain and Abel, Cain killed Abel, so it was pretty much just Cain that carried on that legacy.

Then, eventually they went monotheistic but didn’t/couldn’t resolve the issues that introduced with the earlier stories (Noah’s story also has similar issues, though I’m not sure it also once had a resolution of “other gods made more people” or if the resolution involved smacking anyone who brought up the question).

Which is exactly why there’s also an effort to kill home computer ownership. Capitalists hate all the pesky competition they have to compete with that makes their enshitified products look like shit.