• 0 Posts
  • 302 Comments
Joined 1 year ago
cake
Cake day: June 29th, 2023

help-circle
  • The body. It’s feeding you vast amounts of information every moment, it’s the one making decisions, you’re the AI assistant providing analysis and advice

    If you clone a tree, you get a similar tree. The branches aren’t in the same place. If you clone a human, why would the nerves be laid out the same way? Even if it’s wired up correctly, without a lifetime of cooperation why would your body take your advice?

    Imagine you wake up. Red looks blue. Everything feels numb. The doctor says “everything looks good, why don’t you try to stand up?”. You want to cooperate with the doctor, but you don’t stand up. You could move, but you don’t. Rationalizing your choices, you tell the doctor you don’t feel like it. You feel your toes, you shift to get away from the prodding of your doctor, but you just can’t muster the will to stand

    Imagine you wake up. Your sight is crystal clear, you feel your body like never before. The doctor says “don’t move yet”. With the self control of a child, you rip out the itchy IV to get the tape off of you. The doctor says something in a stem tone, and you’re filled with rage. You pummel the doctor, then are filled with regret and start to cry

    Emerging science suggests this kind of situation could lead to brand new forms of existential horror




  • That’s because we’re using it wrong. It’s not a genie you go to for answers to your problems, it’s mighty putty. You could build a house out of it, but it’s wildly expensive and not at all worth it. But if you want to stick a glass bottle to a tree, or fix a broken plastic shell back together, it’s great

    For example, you can have it do a web search, read through the results to see if it actually contains what you’re looking for, then summarize what it found and let you jump right there to evaluate yourself. You could have it listen to your podcasts and tag them by topic. You could write a normal program to generate a name and traits of a game character, then have the AI write flavor text and dialog trees for quest chains

    Those are some projects I’ve used AI for - specifically, local AI running on my old computer. I’m looking to build a new one

    I also use chat gpt to write simple but tedious code on a weekly basis for my normal job - things like “build a class to represent this db object”. I don’t trust it to do anything that’s not straightforward - I don’t trust myself to do anything tedious

    The AI is not an expert, I am. The AI is happy to do busy work, every second of it increases my stress level. AI is tireless, it can work while I sleep. AI is not efficient, but it’s flexible. My code is efficient, but it is not flexible

    As a part of a system, AI is the link between unstructured data and code, which needs structure. It let’s you do things that would have required a 24/7 team of dozens of employees. It also is unable to replace a single human - just like a computer

    That’s my philosophy at least, after approaching LLMs as a new type of tool and studying them as a developer. Like anything else, I ran it on my own computer and poked and prodded it until I saw the patterns. I learned what it could do, and what it struggled to do. I learned how to use it, I developed methodologies. I learned how to detect and undo “rampancy”, a number of different failure states where it degrades into nonsense. And I learned how to use it as another tool in my toolbox, and I pride myself on using the right tool for the job

    This is a useful tool - I repeatedly have used it to do things I couldn’t have done without it. This is a new tool - artisans don’t know how to use it yet. I can build incredible things with this tool with what I know now, and other people are developing their own techniques to great effect. We will learn how to use this tool, even in its current state. It will take time, its use may not be obvious, but this is a very useful tool



  • Oh, the metaphor goes further. We’re not the pilot of the suit, we’re not the hardware, we’re not the OS of the suit, we’re the AI assistant

    We speak for it with the other AIs, we get called up to handle things we don’t have learned behaviors for, we analyze and provide feedback - we give advice and it feels like we’re making decisions, but we’re not



  • That’s not what I’m mad about. I’m mad that it won’t ever work - Ubisoft isn’t trying to figure out why their games are failing, they’re trying to figure out how to keep the stock price projections up

    Hence this article, which is signaling to wall Street “we’re going to make layoffs and hire cheaper, less experienced people”. They’ll probably do it by closing studios and buying up new ones - that’s pretty much their standard operating procedure. They buy up a studio, take their IP to add to the pile, then turn it into a formula and churn out games until the players lose interest in the IP

    What’s the problem? They’re too damn big. What’s the solution? Block them from acquiring more studios and they’ll die without leaving a swath of destruction on the way down. Ideally split them up. Do the same with Microsoft and EA, and we could save the gaming industry overnight (granted, more like over the course of a few years)

    Voting with your wallet doesn’t work because to the leadership of a Corp, sales aren’t what matters. Stock price matters, which is only tentatively linked to how profitable the company is, which is only tentatively linked to the quality of their products








  • Why do you think C is the one true language? It’s a tool.

    There’s a single very simple answer to “what tool should I use?”. Use the best tool for the job

    The job is the objective - what are you trying to accomplish? What are your priorities? What compromise is best between time, cost, and quality? What are your abilities? What’s in your toolbox right now, and what could you obtain within the time frame?

    For you, the best tool might always be C. I don’t know how you’ve specialized or what you do, but C is powerful. Maybe you have an orderly thought process code meticulously, maybe you struggle to learn new languages. Maybe there’s just no better option for the jobs you take on

    For me, C is rarely the answer. Not never, but outside of school I can count on one hand how many times I’ve chosen it. I code intuitively and feel how the code fits together, I can pick up languages on the spot and switch even more easily. But I’m not meticulous, it’s against my nature. I make mistakes frequently - but I learn by doing, and I don’t need to understand to start doing

    All that said, why do we keep making languages and frameworks? Because as programmers, we build the tools. We can also share them without losing them. The perfect tool for one job won’t be the same for any other job, but a pretty good tool for many jobs is a valuable tool

    The trade-off with our tools is between power, versatility, and cost (generally being time). We all want powerful and versatile tools - but our time is limited, and so we can’t afford the cost

    Ultimately, I think you’ve correctly spotted a recurring problem but misidentified the cause. The cause isn’t the tools, it’s the fact that the cost is someone else’s time. And the fact we have no way to translate money into their time

    A corporation can fund a team to continuously develop a tool they rely on. An individual can’t - we could chip in a few bucks here and there, but we use a lot of tools. We don’t know good tools from bad ones until we use them, we don’t know what tools are used to build the ones we need either.

    So everyone and their mom wants to build a service to fund work on their tools. I hate services, I don’t want to give them my data or my money - I want tools that will work on my devices, not because I don’t want to deny them pay for their work, but because I pick up, drop, and modify tools all the time

    That’s the real problem - if I could donate x dollars a month to support the tools I use, I would. If I could choose for us all to pay more taxes to support the tools we all use, I would take that deal. Hell, I’d go through the effort to generalize my personal tools

    Instead, the only real profit to be had in OSS comes from companies, because they can afford to fund them directly, or services, which individuals tend to hate but companies barely notice. The tools aren’t the problem - the economics are the problem


  • It’s actually really fascinating - cats seem to rapidly learn culture while they’re weaning

    Cats in Japan are very friendly and trusting of humans, cats in America are more cautious and wary

    Japan has folklore about multiple variations of cat yokai that range from fickle trickers to malevolent supernatural ones. Cats are considered good luck, killing them invites bad luck. They have euphemisms like being in no position to refuse even a cats help, and their presence being a good omen

    America has folklore about cats being bad luck, and tied to witchery. We still use euphemisms about skinning cats, letting them out of bags, swinging them, etc. Killing cats wasn’t abnormal behavior even a century ago

    And apparently, if you bring a female Japanese cat to America, it’ll take several generations for the descendents to localize to the culture. They even meow differently