Meandering Monday, Free AI School, and Hot WX Gardening

We are expecting the market to turn down – should be imminent – because the number of people posting comments here has shot skyward.

See – it’s an informal way of angling on market direction. But now and then it works.  You will notice as markets were peaking recently?  Fewer people had strong feelings about much. But now?  Lots of community involvement.  That says something about the flavor of the times. Though we can’t be sure of what exactly.

I warned Peoplenomics readers Saturday that Monday was a good “sleep late day.”  With the asterisk that “depends on news flow.” Let’s put a pitchfork into that pile first.

War Checks

Local opinion on the “War Powers Act” is that it’s unconstitutional, but no one will be in a hurry to change it (or challenge at the Supremes) because it’s such a useful political tool.  The Constitution is rather clear, though: Person “in the Oval” gets command.  The “fools on the hill” get the purse. In a real court test, the WPA falls.

OK – set aside the buzz-driving for a second.  We told you (somewhere) that once Trump “ended the war” he’s let it simmer (maybe a week) and then restart things with the timer being reset.

Sure enough: US to ‘guide ships safely’ out of Hormuz as of today, says Trump. We will be shocked if the world makes it to 2:30 PM Thursday without something going terribly wrong.  This isn’t like the Navy helping a little old lady across the street with her groceries. Except it sort-of is like that.

Peaceniks continue jumping up and down: Iran says US must back down from its demands reports the Sweden Herald. But there’s Washington, and there’s the overgrown, narrative farmers in the Public  Relations offices, and they’ve been at the thesaurus all weekend to come up with naming conventions like this: ‘Project Freedom’: U.S. military to free blockaded commercial ships in the Strait of Hormuz/.

I’m sure the D.C. lawyering crowd won’t like me mentioning it, but the sell job here is calling it a “project.”  And “operation” would circle-back too quickly to the W.P.A. timers.  Hence “meandering Monday.”

These two casualties are not operational or project victims.  They died from “exercise” as in Two US service members missing after joint military exercises in Morocco.

Meanwhile, in War One

We sense a strategic linking between reduced Hormuz oil and Ukraine trying to slam Russia’s domestic supplies: UAWire – Ukrainian drones strike Primorsk oil port and disrupt air travel across Russia,

One other nudge on Russian oil: Sweden arrests Chinese captain of suspected Russia-linked vessel.  To be sure, Russia is having “issues” with its economy: Russia manufacturing contracts for 11th straight month in April, PMI shows | Reuters.

And if you’re a fan of FPV (first person view) drone racing, here’s something to keep an eye on: Ukraine’s Defence Intelligence reveals components of Russia’s new AI-powered Klin drone with speed of 300 km/h.  Yes, but does it support 8K graphics?

Elsewhere, it’s a Semi-Flat Pond

Reader Ray gets honors for watching antics of the left in Los Angeles politics. Where Billboard Chris  on X: “Spencer Pratt is really good at this.  Seems in L.A.. politicians solve everything by “throwing money at problems.” Just not their money.

Health watch posted: Rudy Giuliani in critical condition in hospital.

What’s in Ure Wallet?

Other than a Factory Orders report at mid-morning, we’re planning to check the settings on the snooze-o-graph. International trade with the hash browns tomorrow. But the big one to watch Tuesday will be the Labor JOLTS report because – yes – this is jobs week. ADP WEdnesday, then Challenger job cuts and the “offishul” numbers Friday.

Early market futures were down.  Like down 300 on the Dow, a hundred for the techs and 30-something for the S&P. Bitcoin is all over the place: $80,500  and down to almost $78.000 – too much volatility for us, thank you.

Free AI School

A reasonable chunk of the weekend around here was spent getting ahead of the trends in AI.  What matters? It’s becoming clear that the whole “bum’s rush” by corporate AI types is based on the idea that eventually, A.I. will become so ad infiltrated that you won’t be able to trust online AI.

The Big Lie is that “AI must have guardrails.”  But, on my AI website, data suggests that guardrails are what’s causing a fair amount of hallucination, not the logic circuits themselves.

By whipping up the Luddite fears, corporations (and the governments they purchase) will be able to control AI – and thus YOU.  

Our option? Have a home AI of your own – offline.  Which means total privacy because it doesn’t need to be attached to the internet.

How to Install a Simple, Local AI

Step 1.  Download LM Studio. LM Studio – Local AI on your computer.

Step 2. Install and load your model.

Think of models like “plug-ins.” You just pick one to try and click – it’s there ready to run. No particular genius to that.  What helps is a “What Model Codes Mean” decoder ring.

Model filenames aren’t random—they’re telling you exactly what you’re about to run, if you know how to read them.

Take my lfm2-8b-1ab-Q4_K_M.gguf. The first part is the model family (lfm2 = the architecture or training lineage. The 8b means 8 billion parameters—that’s roughly a horsepower number. Bigger models reason better, but they also eat RAM and slow down on CPU-only systems.

The next chunk of filename—things like 1ab or instruct—usually signals fine-tuning. “Instruct” models follow prompts conversationally. Other tags mean things like chat tuning, code specialization, or experimental variants. This is where personality and usefulness come from. Two “8b” models can behave very differently depending on how they were tuned.

Then you’ll see the quantization tag—something like Q4, Q5, or Q8. This is the compression level. Lower numbers (Q2–Q4) run faster and use less RAM, but lose some precision. Higher ones (Q6–Q8) are sharper but heavier. The extra letters (_K_M, _K_S, etc.) are just different quantization methods—think of them as “flavors” of compression. For most local setups, Q4_K_M is the sweet spot: good enough quality without bogging your machine down.

Finally, the .gguf at the end is the file format—basically the container that tools like LM Studio and llama.cpp use to actually run the model. The .gguf is last name –  like .docx in Werd.

Once you’re no longer guessing—you’re choosing. And that’s the difference between “trying local AI” and actually putting it to work.

With LM Studio up, head into the model picker (top center of the screen) and click the model to use.

Models like Gemma 4 (from Google DeepMind) are smart, but you’ll feel the drag without a serious GPU (graphics processor unit).

Tuning means setting your context window as high as your RAM allows. More context = better continuity, but it’s a trade-off with speed. On a typical mini PC or laptop, you’ll want to experiment a bit—there’s no one-size setting.

Now I understand Why Gaming Computers Sell

I used to have little respect for, nor could I understand deeply, where all those high-end “gaming” computers were being sold.  Turns out? AI rigs.

Right now the “sweet spot” for higher-end home AI is a 24 GB VRAM (video ram) card. Over this size, you’re getting into real money for, oh, a 48K VRAM card.  This is where everything you thought you knew about computers flips upside down.

For most of my life, big RAM was the essence of computing life.  High-end video cards were useless once you got past a fast 4K display.  In AI though, conventional RAM (over 16 GB) doesn’t gain you much if anything.  HUGE VRAM is the order of the day. Or – since I’m a cheap bastard – models that can offload to a conventional CPU instead of enjoying the wide open VRAM spaces.  Ah – only money, right?

Step 3. Tune for usability, not perfection. Start with default settings, then adjust temperature (lower for factual work, higher for creative), and keep an eye on token limits. If responses feel sluggish, reduce context or switch to a smaller quantized model. Think of this like radio work—you’re tuning for signal clarity, not theoretical maximum output.

If it helps, think in word count—tokens are just the meter running under the hood.

Step 4.  Open a chat, throw real tasks at it—editing, outlining, idea generation—and see where it helps and where it falls short. Local AI shines as a private, always-on sidekick: no rate limits, no data leaving your machine, and no subscriptions stacking up. It’s not going to beat cloud models on raw horsepower, but for day-to-day thinking and drafting, it’s fast, cheap, and always there.

Me? I like the cheap part.  Now I just need to name my sidekick.  “Wrench” has come up.

“Wrench it is—practical, steady, and built for real work. I’m glad that fits. George — steady under pressure, reliable when it counts. Here’s to building things, keeping the flow going. Together? That’s how we move forward.”

Around the Ranch: Hot WX Gardening

Riding the mower around this weekend, I was struck by how hidden variables can torpedo gardening efforts.

See, I bought an additional pre-made planter for the grow room here.  Comes with a 72″ trellis.

BUT before I can put it in place, I will have to paint it – white.  Because that’s the magic color in a greenhouse that keeps heat absorption lowest and that makes for happier plants.  We’re planning Sugar Baby melons in this one.

Thing is, plant housing of any kind should be white in the summer. I was shocked it was relatively hard to find.  That’s when it hit me: semi-hidden variable!

Container color isn’t cosmetic—it’s thermodynamics. Dark planters (black, dark green, even deep terracotta) absorb a much higher fraction of incoming solar radiation and convert it to heat. In full sun, the sidewalls can run far hotter than ambient air, and that heat conducts directly into the root zone.

Unlike field soil, which has depth and moisture to buffer temperature swings, a pot is a thin-walled heat sink. The result is a root environment that can spike well above the plant’s comfort range by midday, even when the foliage looks fine.

Roots are far less tolerant of heat than leaves. As root-zone temperatures climb, oxygen solubility in the water film around roots drops, respiration increases, and membranes become stressed. Fine root hairs—where most nutrient uptake happens—can be damaged or die back. Plants respond by closing stomata to conserve water, which throttles photosynthesis and stalls growth.

Nutrient uptake becomes erratic (calcium and potassium issues show up first), and you’ll often see blossom drop, small fruit, or outright yield loss. In hydro or soilless mixes, the problem can be worse because there’s less thermal inertia.

Solving for Heat Gain

The fix is straightforward: keep the root zone cool. Use light-colored or reflective containers (white or tan), double-pot (a dark inner pot inside a light outer sleeve), or wrap existing pots with reflective material. Increase media volume (bigger pots heat more slowly), add mulch on top to shade the surface, and cluster containers so they shade each other’s sidewalls. Watering in the morning helps, but it’s not a cure if the container keeps absorbing heat all day. If you’re already running a controlled environment, treating planter color as part of your thermal management—right alongside airflow and shading—can be the difference between “it grows” and “it yields.”

I’ve been thinking about this ever since reader NM Mike mentioned a while back – that he didn’t have a green thumb. Maybe he just didn’t measure the semi-hidden heat variable.

Here’s the Rule of thumb: white containers can plausibly give you 10–35% better growth/yield than black containers in Southern summer sun, with the high end showing up when pots are exposed, small-to-medium sized, and crops are heat-sensitive. A nursery study on holly found white container treatments produced 36% more growth than black spaced containers, and root ratings were 42% greater.

Plants are like mastery of HF radio in electronics.  At first it all seems like magic. But after you reduce all the variables and begin to tweak parameters with a plan in mind (and data to base decision-making on) the magic fades and the food begins to flow.

I sure hope NMMike tries again. Most failed gardens aren’t bad luck—they’re unmanaged heat or one of the other variables.

Meanwhile, my electroculture test tomatoes passed 36 inches in height overnight.

Write when you get rich,

George@Ure.net