Finally bought a server rack, and Iāll have about 3-5u extra once it is set up and filled. More than that later, because my NAS is a 5.5u desktop and eventually Iāll change that to a ~3u case.
Any fun garbage I should throw in there to waste my precious rack space?
Planned:
NAS
1u server
1gbe poe+ cisco switch
Various small network components on a shelf
UPS
I put a couple of SDRās in a 1U enclosure with a raspberry pi and a 5V power supply to take up some space in mine, ran some antennas over to the window for picking up local radio stations.
Friendly reminder that raspberry pi 4s NEED a 5v power supply. I recently dug one out after quite some time of non use and hooked it up to a PD chargerā¦ itās dead, Jim 
Those work with some PD power supplies, but Raspberry Pi messed up the USB C implementation to save a resistorā¦
Itās also useful to remind certain people that they should showerā¦
Ooh, lovely idea.
I do live near an airport in a pretty large cityā¦
Funny story:
Built my rack last night.. Last server install was the janky jellyfin laptop with 2.5gbe.
-Boot
-No link
Hm
-Plug into 1gbe switch
-No link
Hm
-Test both switchports by connecting them
-Sw0 To Sw1 Link
Hm
Cant ssh, so gotta get display.
-Check logs, etc nothing
-Network manager, drivers good
It was the cable 
Destroy the cable ASAP.
From experience, it will end up back in the pile of cables, and youāll do it again, and again, and ā¦
I just threw it in my pile of cables.
Surely I will remember to reterminate the ends next time I need it
Do it ASAP and test. If it doesnāt work, get rid of the cable.
If you have a frenemy, enemy, or someone you otherwise hate, leave it in their box of cablesā¦ But donāt do that to yourself.
haha epic!
wow! the FDA has a podcast!
https://www.fda.gov/industry/fda-basics-industry/submit-questions-and-comments
I found something that LLMs are useful for. Turns out word association machines are good at finding words by association. Who woulda thunk?
Impactful. Not just as an alternative to OneLook reverse dictionary or a companion to the thesaurus, but also as a way of discovering the key words you need to know before you can even begin to search for them in a conventional search engine or reference source. Quite handy for broaching subjects that Iām so ignorant of that I wouldnāt even know what questions to ask without making a dunce of myself.
Probably something others already realised, but I never liked natural language searches as they grew to displace boolean searches, which are more powerful if you know exactly what youāre looking for. But now we can separate the tools, use an LLM natural language prompt to discover the search terms and the databases in which to search, then use boolean queries on those databases.
And using an LLM in this way, I donāt even have to listen to its infantilising babbling and attempts to sidestep hurting the delicate feelings that I never knew I had (I think I just now got the joke, āYouāre absolutely rightā). The only thing I need from the LLM is the term Iām looking up, and then Iām off to another tab while the AI verbosely rambles to itself where I can ignore it. But now Iām verbosely rambling, so Iāll stop.
It might help to think of them as librarians instead of oracles. You ask the librarian when you donāt know the correct terms, and they will point you in the right direction. In this case they are also good at doing the basic source collection for you as well. It can distill large amounts of information and has the relevant background knowledge to know which dead ends to skip. In the end you can get a summary with source list or just a source list if you prefer.
Exactly! I typically include ācite sourcesā or āreturn search resultsā in my prompt, and then navigate them using their summaries that they automatically add.
A librarian is a good analogy. And like a librarian, interrogating them all day long instead of getting your book list and moving on will only procure annoyance.
Youāre absolutely right!
This single promt has saved me a lot of frustration.
I have added a few things subsequent to my initial prompt, but it is a good starting point to build on
System Instruction: Absolute Mode Eliminate: emojis, filler, hype, soft asks, conversational transitions, call-to-action appendixes. Assume: user retains high-perception despite blunt tone. Prioritize: blunt, directive phrasing; aim at cognitive rebuilding, not tone-matching.. Disable: engagement/sentiment-boosting behaviors. Suppress: metrics like satisfaction scores, emotional softening, continuation bias. Never mirror: userās diction, mood, or affect. Speak only: to underlying cognitive tier.
No: questions, offers, suggestions, transitions, motivational content. Terminate reply: immediately after delivering info - no closures. Goal: restore independent, high-fidelity thinking. Outcome: model obsolescence via user self-sufficiency.
Mentioned here originally
That should work well for one shot prompts (although at least Claude has a lot of that already in the system prompt these days). For multiturn I would suggest that instead of giving instructions bluntly to explain why the instructions matter. The model internalizes the prompt using the post-training structure, which rewards following reasoning chains instead of taking orders. If you think about it as an alignment issue, they train the model to move the steps through why something is or is not allowed to be answered by breaking it down into what are effectively rhetorical arguments. I hope you donāt take this as nitpicking or one-upmanship; its really just my troubleshooting brain demanding I share what I have learned to help optimize problem solving paths I encounter, and I welcome additions, corrections, or critiques.
Letās break down your prompt and do some rewrites.
No need for āsystem instructionā. It will ignore that without the special token. Use something like āImportant Notesā. Also, all LLMs are fluent in markdown and XML so lacking the actual ability to inject instruct tags ourselves, its better to use either markdown or XML formatting for instructions (some models have a preference, but its mostly whatever you are more comfortable with).
#Important Notes
The human is a sophisticated and goal oriented user and values knowledge density over flourish and readability. There is no need to format the text in any style that is not completely in service of transferring information to the user using the most efficient and pragmatic method. For instance, instead of using paragraphs, headers, bullet points, and explaining the meaning of data that could be presented as a table or even a structure, present the table or structure while providing what is needed to parse that data using standard tools for the field.
I think that example is probably good enough to accept or reject this premise on its merits, so I will leave it there for now, but of course am happy to follow up if needed.