I've been doing questionable things with technology since the days modems squealed.
FOSS projects I maintain may be found on GitHub;
https://github.com/S0AndS0/
Available for contract/free-lance work via Matrix;
https://matrix.to/#/@s0ands0:matrix.org
| Tech Blog | https://s0ands0.github.io/ |
Finally finished 2Bit invitation story ark in Updamm last weekend's stream, which be uploading now
And once updates finish updating I should be soon™ playing a bit more Deathloop today
At The Complex obtained ghost upgrade in that prolongs cloaking so long as one does not move, in the evening, and returned in morning to divert power to two locations during last weekend's play of Deathloop
Which now be uploading!
... And in bit may stream more
Happened across someone suffering from the AI Delusion Syndrome
... they was saying AI somehow lowers difficulty of injecting payloads
difficulty be so low as to be unworthy of measure :-|
Was talking with a recruiter, as one does on a Monday, and grasping
... sorta struggling...
they was wat sort of box to put me in
Didn't have the heart to tell 'em, with sort of things I've done, such box getting legs and personality disorder by the next time we chat is non-zero
Got password to RAK after slaying all but one enemies without being spotted in last weekend's Deathloop stream, that is uploading now
And in a few moments I may be streaming some-more today
No lie, it's becoming one of my favorite demos LLMs not being good
`qwen3-coder:30b` got closest, so close it could almost not taint performance, but, close ain't gonna cut it
Plus it still took 5 proompts to get this far, far slower than doing it old-fashioned way with hands
Update; `starcoder2:instruct` does not play nice with my hardware/Ollama version(s)
Also found incantation which does use GPU, for working models, successfully!
Also also, found and submitted _opportunity_ to enhance documentation;
As well as note for how to obtain compute version locally via CLI, because the table saying 980 should use 5.2 vs CLI asserting 5.0 consumed a few hours of me chasing my own tail x-} This has been ...
Gonna maybe try/regret local LLMs again, now that I'm using NixOS BTW™
Anyone have some good, and tested, recommendations for coding that be 30b or less in params?
Inspirational Skeletor💀