thx! i implemented all the ideas like the ones i presented, including free llm inferences based on basic tagging, a transparent feed algorithm instead of the usual engagement-based one, and hidden dislike counts. you can see it here: https://github.com/karpathy/KarpathyTalk/pull/7
i haven't run it locally yet, but if you're open to these ideas, just give me the word and i'll get it running.
looks like some good content is coming our way
tbh the real advantage here isn't even the cost, it's the customizability. as models keep getting better, that's what's gonna kill the prosumer market. it’s gonna be tough for you and the other guys making niche products to find buyers in the long run. seems like building for the masses is the way to go no matter what.
share on github, will add this into my reading list
Before responding, take time to think through this carefully and thoroughly. Consider multiple angles, potential edge cases, and the full depth of the question. Prioritize accuracy and genuine usefulness over speed.
Give the best answer you possibly can, one that is complete, well-reasoned, and directly actionable. The quality of this response matters significantly to the person asking.
i don't think we have a huge audience that'll actually care about us. i feel like they're just chasing the masses. even a small, healthy tech community here would be more than enough on its own. right?
🤫🤫🤫
tbh it’s no exaggeration to say the WHOLE OPENCLAW ECO is just held together by duct tape. the community was really blowing up about it: https://www.reddit.com/r/openclaw/comments/1s4ixul/its_time_to_be_real_here/
i tried browserbase for a bit, but i couldn't figure out how to manually log in and save my cookies like i do in my own browser. i ended up just building session persistence into agent-browser and set an alias in my zshrc so i could just run 'ab' for my agents.
i was using the playwright chrome extension for a while too, but that thing was super fragile. it needs custom mcp config, but it might be worth your attention.
everyone thinks their own ideas on agent prompting loop are the most valuable, which is actually kind of cool. seeing all these random, creative ways people are using AI in the community has been pretty motivating tbh but before i jump in and start testing my own stuff, i look at what’s already getting positive feedback (i usually prioritize real user comments on reddit over github stars) and check out the person's background. if both seem solid, i grab a coffee, go through the files one by one, and try to figure out what kind of loop they’re trying to build.
would appreciate it if you could consider adding an english version of README?
it would be cool if the community could label AI-generated posts early on (twitter is kinda doing this with dislikes now). if we're all gonna agree on some kind of manifesto, the community should be able to quickly report stuff that feels AI-written and doesn't fit that vibe, maybe with a warning system in place. otherwise, it's just gonna turn into a "molt book" twitter clone, and i don't think anybody wants that.