selectedambient

for sure; you can contribute to things you already use, or build things you will genuinely use every day that solves problems you struggle with. 90% of my builds are just things i wanted. i'm sure there are others that share this sentiment. just my 2c.

100% feel that, finishing my mscs in august. keep your head down and keep working on things that are genuinely useful to you and affect your daily life. often, if you're having a problem, others are having the same problem. it is rough out there though, that is undeniable.

nice i’ve been using a program i built to listen to ytmusic for free from the terminal without a sub. i find so many channels now are just long form ads or the same 5 takes on “the new thing.” would be interested in what you found though in terms of channels. i do find this one dude, “saveitforparts,” entertaining.

would be interested in checking out some of the rss feeds

i 100% agree with what you’re saying, i’m just using a catch all term (news, entertainment ykwim). yes though, demonetize all reaction videos.

that would imply i just let it go to town with no understanding.. so no.

here are a few i have put time and effort into. these are not “vibe coded”, but an agent was utilized at points to save copious amounts of time implementing my architectural decisions; my schedule is pretty slammed as is.

https://mithraeum.studio - local first agent and editor in C, also a few models on HF (mainly jsut qwen wrapped atm but working on from scratch) https://fieldopt.dev - SaaS for dispatching jobs to the field (technicians, trades, delivery, etc.) https://github.com/zblauser/ytcli - youtube music from the terminal in zig (ps it’s free, no sub needed)

accurate and they probably had access already.

accurate, but i’ll keep building anyway

okay i watched his video on it, i can definitely abide.

did you not read the entire post? also correct the one file IS the entire thing. compiling one file on iSh on an iPhone actually works well.

sure i can share it. it started as a text editor that i incorporated an agent into then moved to a standalone agent. now i’ve got the editor, agent, and have begun work on actual from scratch models (these aren’t available yet), but i’ve been messing with fine tuning ollama coder models as something in the meantime. the goal is embedded models accessible to everyone offline. too many tasks require many less parameters than people are using on them and don’t need to be routed through data centers. mainly just built it for myself but figured others might find it neat and/or useful. it’s all C, no packages or third party libraries. also to be clear i’m not against ai, im against it making 100% of the decisions for you. outside of its own models it’s capable of OAuth with claude and gh. it can also api into many others. tooling is always ongoing maintenance and i’ve got the next version of basically all three of them on the back burner waiting to be pushed. so anyway, here’s my little setup, here’s to not getting killed in the comments. contributions also welcome. https://github.com/mithraeums/hako-code

"gpt can you do these PRs for me"

i write an agent in C, create my own models. crickets. pewdiepie gets into programming on a whim creates slop, 3k stars.

i feel like regardless what you're doing, consistency is key, aside from actually learning right? you mentioned people running three sessions at once on projects they have no hope of maintaining. very fair point, it's just gambling at that point. however, working on the same project or few projects, you DO hope to maintain (even with ai) for 8 months to a year straight is an entirely different experience than trying to powerhouse anything and everything just to have it? or something, i'm not really sure what the point in this would be. it isn't applicable on a resume or impressive to anyone with any real technical experience. at least if you're staying consistent you're learning something about the process, how to improve it, everything it does, etc. i've seen it time and time again, previously nontechnical or barely technical people "getting into coding" (i.e. using ai), creating something that would've taken time 10 years ago and marveling at it like they've done something. meanwhile, without thinking.. "if i had no prior experience and was able to quickly throw something together with AI, how valuable is the thing i threw together really?" to be clear i'm not saying you're doing this, but this is certainly what a LOT of the people you described are doing. this isn't even delving into the bugs and security flaws their programs are most likely full of. never mind they're learning practically nothing. anyway, i generally agree with your sentiment.

if you’re not about to pay wp to read a billionaire’s opinion > http://archive.today/qzr0f

this seems more like the very tip of the iceberg here v a complete database, anyhow very neat

a king among men.

lmao this analogy

that's great and all, but i wonder how many of them are getting 50 year mortgages?

too real

i’d absolutely take them off peoples hands.

this is why offline embedded and local models is the best path forward continuing with AI.

sooo tldr; curl is safe.

vim will never be overridden.

it is good to hear that they at least admitted to it and complied.

that's just obnoxious, do you have a link to the article they created about it?

i can't :q you vim

therein lies the point, the secret to ai's success is human reasoning.

then you see people like karpathy out there telling people to just vibe it out, ignoring the fact that he has like two decades of experience to wrap his head around failed/correct output