• 2 Posts
  • 50 Comments
Joined 1 year ago
cake
Cake day: July 3rd, 2023

help-circle

  • Uli@sopuli.xyztoTechnology@lemmy.worldThe most popular GenAI Tools
    link
    fedilink
    English
    arrow-up
    5
    arrow-down
    4
    ·
    21 days ago

    My feelings are mixed. Everything you are saying is true. LLMs, right now at least, are a huge waste of resources. It’s triggering us to move closer to fossil fuels when we should be moving away. Every time I step outside to a nice balmy day, I think, am I going to miss this in a few years’ time? In a few decades, am I going to envy my current self who can do dishes without worrying too much about how much water goes down the drain? Are the generations to come going to look at my occasional can of tuna with contempt and jealousy? Or will they even have the luxury of retrospection?

    I understand what we have to lose and how little we are doing about it. But I have also grown up being subjugated inside a capitalist hellscape. And I’ve spent the past few days having ChatGPT help me set up a CI/CD pipeline and start coding some games I’ve wanted to make for years. It’s allowed me to take a few hours of free time and make progress that I expected would have taken a week. It doesn’t have that effect on every task, but when learning new software, it really feels like having someone knowledgeable sitting next to me to answer my questions and point me in the right direction.

    GPT 3 was kind of a neat party trick - sounds kind of like a person, but a pretty dumb person. GPT 4 sounded smarter, but still couldn’t code for shit. The o1 model still makes mistakes, but it retains the thread of our conversation weeks after the fact and has put together some code that I would have struggled to do myself. Even if it loses more money than it makes right now, I can see the value in progressing development until we achieve AGI.

    People have expressed hopes that AGI will solve a lot of the world’s problems. That it will know just what to do about climate change. That it will crack codes in our DNA and give us endless healthy life. I am doubtful that these dreams will come to fruition. At least not in the way people think. It might have the intelligence to tell us things that we should have already known. Like that we can’t get much better yields in scrubbing carbon from the air than nature itself and we should have reforested far more land than we currently are. And that immortality will take huge amounts of resources and will come at the expense of the health of the masses. More gain for the rich. More suffering for the poor. Business as usual.

    But I think there is a window of time where we can be hopeful about what AI has to offer. And we may even be able to leverage it to solve a big piece of the income inequality puzzle.

    If we make a social media app that is not designed for profit but instead for the good of the people, there are a lot of problems such an app could solve.

    We could design it to seek out real (non-bot) contributors. It will always be an arms race trying to sort real humans from bots but that is no reason to give up. It is a reason to get as far ahead in the race as we possibly can. We should build an app that both recognizes when someone is very likely to be real and when they have also contributed to a cause.

    Imagine an application that tracks creative innovation, such as the creation of a funny video or a new meme format. When someone makes an idea and it is popular, the AI model would determine how much of a given experience is improved by their idea and give them profit residuals based on their contribution. And the more ideas that get built on top of the original idea, the more the newer contributors are rewarded for their contributions.

    Think about if people could design a farm from the ground up using a socialized app for collaboration. Someone could design a camera system to keep track of livestock wellbeing and to head off diseases. They could make AI-empowered systems to track livestock happiness and find ways of increasing quality of life. And creating more humane automated methods of turning crops and livestock into food ready to transport. Some people would focus on creating ideal distribution methods. Others would create stores or restaurants. Others might work on the people themselves, encouraging them to give new more climate friendly meal options a try. Investors would be paid their dues, but there would be no CEO or board of executives. The means of production would belong to the people.

    When people talk about the potential of AI, that’s what I envision. If I can make some passive income with my games and apps, that’s the next project I’ll be diverting my time towards. Because this is a narrow window we have to make this happen. The technology is here, but barriers from climate change and income inequality are only going up. We can lament the fact that AI is currently not profitable and hurting the planet, or we can put more of that energy to use by taking the tools humanity has made and using them to dismantle the systems which made this timeline so intolerable to begin with. The only way to take the current system apart is to make a new one that outcompetes our old ways of life in every measurable way.







  • Read through the Readme and it’s definitely a good tool to know about. It doesn’t fit the needs of my current problem, but I’m certain I’ll use it in the future for context sensitive searching, since grep/awk/sed/tr have definitely fallen flat for me in the past. I might also be able to study how they utilized tree-sitter CLI when I explore my own implementation.

    For my purposes, I want to take a group of similar-yet-different YAML file sets (though file type should be arbitrary), and feed them through a tool that will spit out a YAML template containing everything that is shared between multiple sets.

    Then, I want it to create a file for each YAML which defines which parts to pull from the template file and a list of variables to be inserted into holes in the templates. Basically creating a madlib that can recreate any file in the original group given the right list of variables to insert.

    For example, if I have a hundred YAML files that are mostly similar but contain different project names, have different server types provisioned, and are pulling different product versions, I would want this script to parse all hundred files and spit out a template that could be used as the basis to build any of the hundred files. The template would be combined with a hundred variable trees that would insert each unique part of each file into the right place.

    In effect, I could have a small variables file that gives only the unique portions of the equivalent YAML - in this case, it would contain only the project name, the server type, the product version. Then, these small files could be combined with the universal template to recreate the original hundred YAML files. But unlike using a simple override mechanism, I would be able to change elements of the template YAML including broad structural changes, and after some processing, the change would affect all one hundred output YAMLs.

    One could track things like environment variables that are specific to a certain project version and require that whenever a project version has a particular value to insert a particular environment variable into the output YAML. Or a centralized file could be made specifying which product versions correspond to which projects, allowing the engineer to change all product versions for a given set of projects in one go. Or one could create a universal template of IaC code that’s applicable to a broad swath of use cases and quickly build out a full set of YAML manifests and Terraform files using a small file that specifies what components will be needed and where to authenticate to the server.

    I’m not aware of any tool that does this, but I think tree-sitter gets me much of the way there. If I can use it to parse any given file into a context aware tree, I would then need to make a script that combines the shared features of many context trees and splits the unique features out into small variable files. Then a script to merge them back together as needed. And something to manage file system structure, such as whether to parse every file individually or to strategically merge some sets so you have one variable file that produces multiple output YAML.

    Sorry I’m brainstorming at you, just trying to figure out if the tool I’m envisioning is even feasible. Seems like it is, but I’ll have to figure out how to use tree-sitter CLI before I begin.




  • This is super cool. Watched the talks from Max Brunsfeld, surprised this has been around since 2018 and I haven’t heard of it.

    I actually tried some complex parsing myself lately. I had a bunch of YAML I needed to maintain for various deployments in a CI/CD system. I really wanted to have one YAML template to generate the files, plus a file for each project with unique elements to be injected into that project’s generated YAML.

    Probably was more of an indication that we needed to clean up the overrides we were putting on top of our Helm charts, but I wanted a way to generate our lengthy override files without having to manually keep track of where the differences were between projects. And maybe even stage changes to deployment files for when new product versions are released.

    This is exciting. I’m going to look into Tree Sitter more and maybe try to contact the dev. It seems like it does everything I’m looking for, just for an entirely different use case.




  • I used to work for a company that did various kinds of biometric recognition. I unfortunately was paraded past these cameras many times for testing purposes, so my face was compromised many moons ago.

    We had two kinds of products we installed in airports. When looking at large crowds most airports wanted cameras that would monitor the flow of traffic, determining if there were any bottlenecks causing people to arrive at their gate (or baggage claim) after their luggage.

    The other product was facial recognition for identification purposes. These are the machines you have to stand right next to. There are various legal reasons airports did not want to use any crowd-level cameras for identification. They hadn’t obtained consent, but also, the low resolution per face would lead to many more false positives. It was also too costly.

    But we did have high def cameras installed in strategic locations at large music halls. These private companies were less concerned with privacy and more concerned with keeping banned individuals out of their property. In those cases, we registered faces of people who were kicked out for various reasons and ignored all other faces.

    My point I guess is twofold: first, you might not be facially tracked in as many places as you think you are. Second, eventually you will be and there’s not a whole lot we can do to stop it. For many years, Target has identified people with their payment card, used facial recognition to detect when they return to the store, and used crowd tracking to see where in the store you go (and sometimes they have even changed ad displays based on the demographics of people standing nearby).

    Mostly, you will be identified and tracked when there is financial incentive to do so.



  • I finally got fed up with my Windows machine and upon seeing symptoms of motherboard failure, I’ve ordered all the parts for a new rig and intend on installing Linux as my primary OS.

    Haven’t decided on a distro yet. I’m a DevOps engineer with a few passion projects, so I plan on setting up a couple of kubernetes clusters where I can play. I do all the usual things (word processing, gaming, web browsing, multimedia, etc), plus some AI stuff (stable diffusion, local LLMs, OpenCV). Ideally don’t want to have to fuss with drivers too much, but I don’t mind getting my hands dirty every now and then.

    Is Chimera the kind of distro I should be looking at, or should I pick something else for my first go at full-time Linux?