Abject 0.8.2 released
-1 points by mempko
-1 points by mempko
New in this release:
Quick question for the community. Why is a new release of an open-source project being considered spam? Am I tagging it wrong? Just trying to understand for future updates. This project has some really novel ideas that I'm excited about and want to share.
LLM-generated content (including significant assistance) is often flagged as spam. This was the community and moderator consensus a few weeks back.
Thanks Addison. That's unfortunate and I wasn't aware of it.
I'm someone who has programmed for 30 years now. I came to the US as a refugee. In the 90s my father got a second hand computer from a friend which I learned to program on. We didn't have internet so I went to bookstores and learned by reading, memorizing, and trying stuff at home. Eventually went to college, got a CS degree and have been building software ever since. I have written, by hand, millions of lines of code.
It's wild to me that Donald Knuth was impressed by the code LLMs generate but this community isn't.
I've built machine learning systems for 20 years and understand the math behind them. I don't have any magical thinking around them. They are essentially fancy compilers.
This community is going to be left behind because it's exclusionary. Not just to people like me (who have a strong reputation for engineering and code quality) but also for people who are getting excited about building software because of LLMs who don't know how to code.
I'm inspired by people like Alan Kay and Engelbart who believed computers can be great tools for better thinking. Augment the mind. Help us write simulations.
But programmers created a community of exclusion where normies must use their software. Pay them a lot of money for doing so. Now that LLMs are giving power back to regular people to use computers as computers are meant to be used, communities like this are clutching their pearls.
My project above (Abject) is built on concepts of object oriented programming like Data, Context, and Interaction (DCI). Using techniques like Design by Contract. Has some new ideas around how to coordinate objects and how communication between changing objects can be healed. Has a p2p backbone where people can coordinate software across machines. I believe all interesting ideas worth discussion. Oh well...
So, it's important to keep in mind that part of this disagreement is whether the technology is empowering or disempowering.
It is my personal belief (I am a moderator, but this is not site policy, I just found your post interesting enough to respond to) that the use of these tools de-trains thinking and research skills for those who have them, and makes it harder to acquire them for those who don't. That is not empowerment, it is servitude.
That's an empirical question, about the effect that using them has. It's not my only objection to the models, but you raised a particularly interesting philosophical point that I am sympathetic to apart from disagreeing on the point of fact, so I wanted to weigh in on that.
If I believed that the tools actually empowered people, I would at least find that to be a very important silver lining, and be looking for a way to salvage that benefit while removing the other serious harms they cause. I condemn elitism in all its forms and want no part of upholding it.
There is a lot of arguments about what's bad about LLMs and they are all valid. However they are empowering. I've talked to many people who have never written a program before but used claude or vercel to build a little program that solved some problem in their life.
Over the last 30 years we made building programs harder and gate kept a lot of useful stuff. We expect people to use our apps. Free software as a movement died replaced with corporate open source culture. People mostly use computers now as old media.
With my system for example, someone can write a simulation to solve a problem in their life. They can build programs that work together with minimal effort. The power improves in a compbinatorial way. I like to call it symbiogenisis, borrowing from biology.
I'm trying to channel old ideas from smalltalk, hypercard, and visual basic.
So people can use computers to compute and not just a substitute for old media.
We don't need to keep gate keeping software behind subscriptions and pay walls. Or requiring years of mastery to benefit.
The biggest problem with LLMs today is who owns them, not what they are imo.
Regarding the de-skilling you mentioned. I deeply worry about that too! That's a real concern, but that's always been a concern with any tools. What we need to make sure we don't de-value education. If you don't remember something and need to delegate to a person or LLM, then you haven't learned that thing. Memorization is undervalued.
I absolutely believe that the technology is good at making people feel good about themselves. That is not in dispute.
Have you seen Donald Knuth's paper?
https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf
I don't think it's so surprising that people are opposed to such submissions; the issue is not that there isn't anything worth discussion, but that LLMs invert the effort ratio for writing/reading. People interpret it as disrespectful to their time (which, I'm inclined to agree with; I personally want to see submissions that challenge both the author and myself).
I think a blog post about the considerations you made, difficulties encountered, things especially worth note about the software, etc. would be very welcome (granted it was written majority by you). But just submitting the software on its own, especially when it relies extensively on LLM generated code, is not as welcome. A select subset of people on the site may just outright dislike it because of the technology, but honestly, most of the reason I see hostility to LLMs is to do with the volume of content versus the (perceived) effort to engage with that content. I don't think that's unreasonable, and I definitely don't think that we're going to be "left behind" because of a policy that rewards authors for high-effort content and punishes (again, perceived) low-effort content.
I wrote about it a while ago (https://blog.mempko.com/an-abject-horror/) and submitted it here months ago. I've been posting updates like this to the project. Maybe what I will do in the future is write a blog post of what the changes are so people get it.
I would counter the writing/reading effort ratio with that people don't read anyway. People especially don't read code. Hand written or LLM generated, it doesn't matter. Reading code is a skill that needs to be trained and most are not trained on it but are trained to write it.
Reading program code doesn't even use the language portion of the mind. There is a deep mathematical connection with software code and geometry using symmetry and symmetry breaking. A recent neural science paper proved this where reading code doesn't even elicit the language portion of the brain in the MD system.
https://elifesciences.org/articles/58906
I would prefer someone just downloads the program and tries it. You won't get it just by reading. Programs have an interface, geometry. The code is the stuff that makes it all work.
For text I write by hand because I have a voice and my own perspective. LLMs don't have a perspective. People need to think of them as fancy compilers.
Regarding the perceived effort, I doubt anyone trying Abject would think it's a low quality effort project. But then again, if nobody tries it, it's all perception.
Is there anything on the github page or website I can do to increase the perception of quality?
Is there anything on the github page or website I can do to increase the perception of quality?
It's a matter of presentation. This submission is to a repository and announces a release. That doesn't say much nor offer any information, other than pointing to a tool and adding a description (which, itself, doesn't give much context to the tool).
I would counter the writing/reading effort ratio with that people don't read anyway. People especially don't read code.
See, but I do read. I read things that tell me something new. You're right, I'm far less likely to read code because it rarely does say anything new on its own, so I don't usually engage with submissions like these at all. So if you and I agree that we aren't going to read it: why did you submit it? If you're proud of it or something about it was interesting, tell me why (maybe in a blog post! :D). Otherwise, it's just a link to a codebase that I have no real connection to or reason to engage with, so I'm not going to spend time on it. If anything, using lobsters to announce a release is spam; the submission needs to carry something with it other than "there exists more code, here is the changelog".
That's fair. I do see a lot of releases here but you are right. I'll do a blog post instead of relying on someone to explore.
That said your original comment about AI is still relevant and maybe Lobsters isn't the community that would appreciate a project like this anyway. I'll put my effort elsewhere.
That said your original comment about AI is still relevant and maybe Lobsters isn't the community that would appreciate a project like this anyway.
Years ago I did some work for a pair of founders. One of the founders would go out, meet investors, raise capital, etc. The other founder liked to use the company blog for political screeds and literal drunken stories about his evening. That company fell apart because no matter how good the one founder was at raising capital, its image was tarnished as soon as anyone researched it and they couldn't get into the business they wanted.
I would simply suggest that you consider how this project would look to someone outside of it. You openly call it an abject horror, insinuate it is a living organism that can self-heal, and refer to using it as "the descent." The messaging protocol basically requires an LLM to parse every message and you paint a picture of a thousand "living programs" that find each other in the darkness.
In the already polarized space of AI discourse, you might want to consider if rebranding your ideas would have them be taken more seriously, as opposed to playing directly into the argument that AI is damaging to the planet and a descent into darkness.
I hope this feedback is helpful.
I expect there are hundreds of cool AI coded projects getting proposed here every day. To the developer doing it, they all are the one cool AI project that cares about code quality, has that one cool and unique cool idea, ... . To the rest its just spam. Sorry.
This is of unfortunately also true for my super cool vibe-coded project:-) I can tell you about it for a bit if you want.
I understand your point, that everyone and their mom thinks their project is cool when it's not. What is unique about lobsters is it's an invite only community so the bar for cool should be much higher here. It's not like posting content here is open to everyone.
Seems people forget that.
This community is going to be left behind because it's exclusionary. Not just to people like me (who have a strong reputation for engineering and code quality) but also for people who are getting excited about building software because of LLMs who don't know how to code.
I would encourage my fellow Lobsters? Lobster-ers? Lobsterians? to cultivate open minds. Obviously not every idea is worthy of equal time and energy, and sometimes we don't even have the time and energy, and that's fine. But I don't want to see us become too dogmatic, because that doesn't lead to interesting conversations.
Simultaneously, and I'm saying this with as much love and as little animosity as I can muster: I would encourage you @mempko not to waste your energy trying to convince the Lobsters community that they are "going to be left behind". Even if it's true, it's not an effective way to convince people! In fact, it's a great way to do the opposite of what you want.
I personally would love to see a human-written blog post discussing what you've built and how it solves a problem you personally have. Obviously you're excited about it! Good communication is about getting other people excited as well.
That's valid criticism. It's not my first time writing about it and my initial submission on Lobsters got some good engagement (https://lobste.rs/s/yk91rd/i_may_have_solved_long_standing_problem)
I'm always wiring. Next time I'll pair the release with a blog post explaining it. I'm fine with people down voting which is why I asked why it was marked as spam to learn. I don't want to spam here!