burden

at the end of my rope, folks if anybody is hiring software engineers/web developers/whatever hit me up 7+ years experience across the stack, mostly js/typescript/react web things, sql/nosql dbs, python and php scripting, etc negative bank acc rn, rent overdue can start asap

life used to be so simple I could wake up and post this with some shit like “who up getting liquidated” and feel alright in the world

Bro my tweets are going to go so hard after I die

yv was a genuine individual and a true homie since day one. rest easy, my friend. this place will never be the same without you.

trying out the whole swarm of agents working on a feature thing and I’m not exactly sure it’s any more effective for what I’ve used it for, but it definitely feels cool and chews through tokens crazy style

Need chair recs to match this desk pls

@henlojseam they’re just like me

@0xburden The trick for the smaller models is to steer it a couple of times, like if it fucks up you tell it “you sure bout dat dawg?” And it’ll somehow figure the right path forward

running qwen3.5-30B at 8bit quant on a cloud gpu just to try out and it's working better than expected tbh, will be interesting to see how much code quality we can push out for cheap as open models get better and take up less resources

@henlojseam Yeah it’s gotten lost while trying to do some shit for me more than once, definitely needs more explicit prompting but still p usable for a model that’s living entirely on a 48gb gpu

@0xburden It’s bad at more complex reasoning tasks so I use it as a subagent

also tried devstral 2 small recently and was massively disappointed, still keeping an eye on mistral though

｢こいつ、おやつ持ってる時しか寄ってこないんだけど…｣

💛💜💙🧡

in las vegas at a bubble craps machine whole net worth on the dont pass line ($30)

@0xKoda interesting, I’ll have to try soon

@0xburden gud model much better edit/patch quality over 4.7, not opus 4.6 level yet but they’re def doing things

So what’s the verdict on GLM-5?

@0xKoda Could use pi or opencode as your agent tooling and run whatever model under the hood

ok codex 5.3 is much better than the new opus, but i much prefer claude code than the codex app. Terminal orchestrator would make alot of sense here.

currently finding out how well kimi k2.5 can handle building out an nvim plugin

got a hold of a cheap thinkpad off of ebay again recently, feels good having a dedicated linux machine back in the house

不釣合

still testing but p workable atm had a clanker port over my nvim config with vim-plug to a more modern lua config with lazy.nvim made a repo and some scripts to get up and running quickly e.g. need to install on some other machine (vm/dev container etc) github.com/0xburden/son-o…

If you see this, lmk what model(s) you’re daily driving for agentic code and what agent tooling you’re running it in most the time recently I’ve been switching between minimax m2.1 and opus 4.5 in opencode