Vibe-SysAdmining: LAUGH OF THE DAY

Treść źródłowa

In which a frontier LLM model tells me to configure a new chatbot to use model “/anthropic/claude-sonint-4-6”; yes, this only means that somewhere in the training data some poor SOB mistyped “sonint” for “sonnet”; but nothing with even sparks of anything that could be called “agi” would let this typo through at a temperature=0.1, would it?I am not a good-enough programmer to be successfully vibe-coding anything, save for scraping, indexing, summarizing, and archiving scripts. What I may be good enough at is to successfully vibe-sysadmin the under-the-dining-room-side-table machine. Why? Because the machine is free, and it looks to me as if it would payback in a year given likely datcenter costs, even if the machine were new and costly. Save some money. See whether any belief that companies will make fortunes off of selling the services of their data centers is vain, whether it is true that the obvious path for users will be to light up their own dark silicon with open-source models and the time of their existing programmers and sysadmins.So far, it is going well, I guess?:ShareGet 75% off a group subscriptionThree models are currently loaded and ready:alibaba/ollama/qwen3:32b-q4-128k: the 28GB Qwen3-32B model quantized to q4 with a maximum context length of 128k tokens, and with “reasoning” turned on. google/gemma4:26b-a4b-it-q4_K_M: the 23GB Gemma4-26B model, quantized to q4_K_M with a maximum context length of 256k tokens. alibaba/ollama/qwen3:8b: the 8.8 GB Qwen3-8B model, also quantized to q4_K_M with a maximum context length of 40k tokens.The hope is that:the third of …