Since fable 5 went down just some time ago(although this is no longer the case), a lot of fable or Mythos related content including models and datasets were completely killed from HF, it seemed like such a blow to the local development community to lose such important files, this is what inspired me to work on my project, it’s a p2p decentralized distribution network (aka a torrent site) that pulls files from HF and verifies them through sha256, I’ve been making slow and steady progress but so far I haven’t made a single public post about it. I’m genuinely scared of a number of things happening, that the project won’t gain traction, ridicule from more experienced developers, I’m also worried about security vulnerabilities which I’m sure I haven’t fully patched yet…
How do I get over this fear of talking about my project? How do I gain stars on GitHub, and community engagement? After all, a decentralized distribution network is nothing without its peers…
Dont worry about the people that hate have faith if your project is usefull ot will find its userbase
Good well balanced and mentally stable developers should not ridicule you.
Anyone who does, block their arses.
>so far I haven’t made a single public post about it
Well, now you made one. Congratulations.
Maybe you need to look at the reasons you feel unsure about promoting the project. You seem very eager to get stars on Github (and let’s not get into the GH situation, nor “AI”, nor the fact that your open source project relies heavy on proprietary models).
I’d suggest you ignore the scoring system for the moment, and just write from your original enthusiasm about the project. Leave marketing to corporations that can afford marketers.
Stars and upvotes and likes are addictive metrics, and can have a deleterious effect when they don’t show. And yes, stars decide how high you rank on Github (see parenthesis above), but that’s for a later stage. It begins with writing the first post — and for that, again, congratulations 🙂
And you should care what people you don’t know on the internet think because… ?
social media is not a reliable source. it can expose you to new information, but it is not reliable .
Feedback is how you learn. how you find out what works and just as importantly what don’t work.
i think there was a study that showed that people that take a course on decision making have lower anxiety. it makes sense, once you do a risk-reward analysis and consider alternatives it makes you accept the risk and act despite it. realizing the anxiety is unhelpful or ineffective repeatedly leads to Desensitization and eventually reduces it.
with that said, there might be some underlining issue and you need to work on self regulation skills. using stuff like CBT/ACT/NSDR/meditation/stoicism. there is a open source test for the big five personality test. if you score high on neuroticism, it might indicate you should work on it.
I mean I support this in terms of decentralization. I think lemmy wont be as receptive as they’re allergic to AI.
There’s a difference between being allergic to AI and being allergic to LLMs shoved everywhere they don’t belong. There’s good use cases for AI and LLMs in particular, but I have rarely seen it being used well. Most people who unconditionally support AI usage are incompetent idiots who couldn’t do anything without AI.
Fully agree with the last bit. And let’s be honest when we admit that there are “good use cases for AI and LLMs”. None of us here are doing protein folding or crunching other ginormous datasets for science.
Using Claude to write code you don’t have the chops to quality check… doesn’t really qualify. That’s just choosing convenience over safety. “Move fast and break things” as a service.
Using Claude to write code you don’t have the chops to quality check… doesn’t really qualify.
What if I do have the chops to quality check and actually do treat Claude like a junior coder?
I often use it for annoying tasks like “transform this data structure into the one which is specified in the new major version of the upstream API” and write a test before.
It basically saves me typing things and I would need to test my own code anyway. That’s one of the good use cases: glorified autocomplete 😅
What if I do have the chops to quality check and actually do treat Claude like a junior coder?
Says every Dunning-Kruger sufferer who don’t and can’t 😉 I don’t need to doubt your skills and dilligence, just those of every lazy hack who whack out slop projects without a second thought.
Sorry, but I think those outnumber capable, security-minded developers by a large factor.
Says every Dunning-Kruger sufferer who don’t and can’t 😉
Good point, still, I wouldn’t ever vibe code because I’m faster doing it myself than screaming at an AI to do it wrong three times over and tell me “oh you’re right actually, I’m sorry”
Sorry, but I think those outnumber capable, security-minded developers by a large factor.
Yeah, they’d be shit without AI aswell though, they just have way more output now that the shit they churn out actually compiles after screaming at a chat prompt for hours.
What if I do have the chops to quality check and actually do treat Claude like a junior coder?
When you have an actual junior coder, they learn over time, including from their mistakes, until they’re a senior coder that can do the same for others. When you’re “treating Claude like a junior coder”, you’re just cleaning up after a mostly unchanging system. See also this article by a Zig contributor.
And having AI do annoying tasks (which you then need to double-check if correct anyway) you’re missing out on the opportunity to get better at problem solving by skipping the part where you find a programmatic solution, even if it’s a one-time thing. (Though when you work on an open source project, it’s nice if others can verify how you got to your solution in the first place.)
When you’re “treating Claude like a junior coder”, you’re just cleaning up after a mostly unchanging system.
Good point! As soon as I feel like the LLM isn’t helping me but instead costs me more time than it saved, I try not to fall into the sunk cost fallacy.
I found it okay-ish to write documentation and fix smaller things which are just annoying chores. It can “learn” to solve the same issue in slightly different contexts if you write proper documentation. Which has the additional benefit that other humans can actually learn from said documentation aswell.
you’re missing out on the opportunity to get better at problem solving by skipping the part where you find a programmatic solution
I’m already pretty good at problem solving. I know how to solve the things I let AI do. I don’t need to solve the hundredth similar issue in a slightly different way just to practice, because you don’t learn from success, only through trial and error.
As soon as LLMs generate code which I find to be subpar, too fancy, opaque or complicated, I take it as inspiration/challenge and write a better solution. I don’t blindly merge LLM generated code. Just like with a junior coder, I don’t babysit LLMs, but at some point I’ll just do it myself.
As I said, LLMs are basically glorified autocompletion, and anyone thinking they should have them solve problems is on a path to idiocy and incompetence. You’re completely correct in stating that one shouldn’t become lazy so they don’t forget how to actually do their job. That’s when you can be replaced by a shell script.
Sorry but you’re screwed. Lemmy is filled with an Anti-AI army of users ready to downvote every post you’ll make.
2 options: tough it up, or gang up with other AI devs and create an AI friendly community for FOSS / selfhosted apps.
Personally, I don’t care if it was AI assisted. I only care if it’s good or not. And by “good” I mean a lot of things. Including but not limited to:
- Maintenance history
- Contributors
- Performance
- Code quality / readability
- Documentation
- Dependancies
- Security
- Privacy
- UX/UI
- and more
Sorry, I don’t speak AI zealot speak, what is HF (huggingface.co?) and why do you need to host proprietary models there?
No proprietary models are hosted, you can only access models hosted by the community or ones with open source weights,
To answer your question (I’m assuming you are talking about me hosting it, not HF), I do actually pull LLMs from HF, but also a lot of other files, including audio models, datasets, and AI adjacent systems such as retrieval systems, etc.
I think my project has primary use cases beyond LLMs.
No proprietary models are hosted, you can only access models hosted by the community or ones with open source weights,
why mention Fable and Mythos then, which are proprietary models? Community hosted models can also be proprietary. Proprietary in contrast to Open Source just means owned and operated by a corporate entity, which can be part of a community as long as the license is not Free.
