Sound like we have related ideas on this. Here is the engine that does some of this
- Yes, I would NEVER lay off the overwhelming majority of my PRODUCT DESIGN EXPERTS and risk everything on statistically driven probability word-sequence software. That’s utterly retarded.
- The proof is in the complete lack of pudding. Ford management tried replacing their engineers with LLM prompts. The problem is, management majors don’t know a fucking thing about engineering. Do you know WHEN to apply Young’s Modulus when designing components of a steering sytem? AI DOESN’T. This is why Ford is literally in danger of being forced into bankruptcy due to mandatory, federally ordered recall work alone. The recall work on several of their products may end up costing more per vehicle than the cost to build them in the first place.
- MBAs have done more to screw up our economy, and utterly DESTROY the lives of our workforce than any other segment of society outside of the bankers who, upon obtaining 50% ownership in corporation after corporation, GAVE China 98% of our manufacturing base. At this point, to reindustrialize, the FIRST STEP would be having the tool & die shops workiong 24x7 just to produce more tooling to stand up more tool & die shops. If we wanted to ramp up military production like we did in WW2… we can’t. There IS no civilian product manufacturing base in the U.S. any more available to convert to wartime production. That’s why we keep RUNNING OUT OF MISSILES in these battles with the Houthis and Iran (it is, however, turning out to be a blessing in disguise, because the people of Iran do NOT deserve to have their country destroyed over the grandiose power-tripping aspirations of a bunch of identity-thieving Khazarian (but I repeat myself) mobsters LARPING as the descendants of a tribe who LITERALLY killed every family member, and then committed suicide, rather than be taken captive at the Siege of Masada in 73 AD. ANYONE running around claiming to be a Judean is a liar.)
Dropping. No, it isn’t. Do you know how much Nvidea is charging for those GPU cards that the Artificial Stupidity companies are buying? Over $100,000 a pop. Have you seen the price of memory lately? it’s NOT dropping. Have you seen the price of a kW-hour of electricity? It’s rising.
And we haven’t even had one full-blown attack on a data-center yet, but it’s coming, due to government refusal to reign in the Orwellian company with the Orwellian “Flock Safety” and their MASSIVE 4th Amendment violation scheme.
If you want to see the kind of thing that’s in a world-class chess engine these days, this talk from Rust Week 2026 gives a taste of it. This is all purpose-built hand-tuned stuff, not anything coming out of an LLM.
How can KEEP running out of missles when we can’t produce any in the first place?
One year ago, getting GPT-4 quality cost $2.50 / 1M input tokens. Today, GPT-4o-mini gives you that exact same quality for $0.15 / 1M tokens. That is a 94% price drop for the same results
The industry trend is called the “Tiered Super-Moore’s Law”
Its usper because the price drop of AI intelligence is happening 5 to 10 times faster than traditional Moore’s Law ever did. While traditional Moore’s Law saw chip power double every two years, AI is collapsing the cost of identical intelligence by roughly 10x per year
If you look back two years (to mid-2024), the math perfectly aligns with a 99% to 99.9% cost reduction
We have a minimal manufacturing base capable of doing peace-time production.
Unlike Russia and China, we have NO surge capacity, other than adding a second shift. Adding a third shift to a plant is never a long term solution, because that makes maintenance a weekly activity instead of a daily activity, and in the long run, ruins either production quality, or makes the cost per unit surge even higher, as you’re paying employees to stand around while maintenance is doing periodic maintenance and/or break/fix that normally would have been handled during nightly non-production hours.
Here’s a challenge for you. Locate a source of screws/bolts made IN the United States, which doesn’t have 100% of their output going to a defense contractor and/or the military itself. 40 years ago, you would have several to choose from within the nearest city. Now there are less than a handful across the entire country. Every screw and bolt in a car coming off a U.S. assembly line for civilian products was imported from China.
Then why are companies freaking out over the fact that the same amount of money for their LLS service that covered a whole month, now lasts less than two business days, starting as of last month?
> The industry trend is called the “Tiered Super-Moore’s Law”
I know Moore’s Law (I’m a computer engineer). Moore’s Law is already close to ending because device sizes are now down to around 100 atoms wide (and semiconductor properties are based on dopant concentrations to bulk material in the range of 1:10^9 to 1:10^15. More concentrated than that, and you have a conductor, not a semiconductor.
“Tiered Super-Moore’s Law” sounds like something invented by people who have been LYING for the last few YEARS about what this software can do. The bubble is ALREADY breaking. And that bubble was based on marketing hype, not truth.
Prediction: 2-3 years from now anyone with a basic Chatbot will be able to feed it a screenshot of your current game (online or physical), then the Chatbot will tell you the strategic and tactical situation, and advise you on your next move. To assuage your dread of this (and its coming whether its my rules engine or the next tinkerer), I guess you could learn more about it, design ways to control or police it, and maybe even harness it for your and this communities’ eventual wargaming goals. Hating it and hoping it goes away might not be your best option.
Anyone here who has yet to send a screenshot to an AI assistant should try this now. Whatever you get back, it will be fascinating (and, yes, a bit of a scary or hilarious wake up call). Try free gemini or just chatgpt. Here is a sample to try (but you might need a full high-res image from your own pc or phone camera) :
I recently instructed Grok to teach me how to play Austerlitz: Napoleon’s Greatest Victory. I fed Grok the rules, scenarios, charts, maps, and errata, then ran the game from prompts from me, to include screen shots from Vassal during the turns. It was an interesting experience. In the end, we both taught each other. The biggest problem though was Grok’s eyesight while reading hex numbers and chart rows columns. Grok would constantly be a hex row off or read the wrong combat result wrong, even though it had a pretty good grip on explaining and interpreting the rules. Overall a pretty enjoyable experience. Played an entire game. It was basically a solitaire game with me playing both sides and Grok assisting where needed.
1 Like
Excellent. You essentially created your own Austerlitz University. The system I have developed solves the vision issue and can be applied to almost any game that has digitized maps, rules, and graphics. Did Grok weave in the actual history of the battle, the strategy writings on how to win the game, and the bugs in the game?
I prompted Grok to use historical goals but the scenario I used was the learning scenario which didn’t really allow for much historical perspective. I would describe Grok’s participation as similar to playing the game against another human who is more familiar with the rules, rather than an airtight AI protocol. There were times when I had to keep Grok on track. I would describe the experience as similar to using GPS where it helps to have a good sense of the intended direction and an idea of what the best route to arrive is while GPS focuses on letting you know which turns to take. on the chosen route. It was very conversational in nature, but Grok’s strongpoints were keeping the flow of the game per the Sequence of Play and clarifying rules. I can’t state that I felt that Grok had a clear overall status of the scenario strategy and how to win a chess match against me, but as I said, the prompts were primarily to aid me in a solitaire game and learn the game, so it was a success to that end.
I will probably try and take it to the next level next time and challenge Grok to a game to see how that works out. One disadvantage to the using the game that I chose, there are hardly any online AAR posts online, so there was very little for Grok to refer to besides the official game documentation.
Great. I have encoded the game into the legality engine including the playbook. If interested I can run the full training process which would allow you to provide your Grok a “champion playbook” based on 10s of thousands of automated playthroughs. Your grok gets the rules, a computerized map, the champion playbook and you basically playbymail against your grok.
Okay…what do I do?
Several possible next steps. You know how to converse with grok. What platform do you use? Cli… web app… api? Can you work with a git repo, or prefer a Windows exe app? BTW grok can probably do the github part for you if you Want to try. It will also be safer.
You can also just download the exe to learn the interface on an existing simper game while I train the system on austerlitz, which might take a few days
Actually, I am making it much easier. When ready, you will simply click on a URL which will open the Legality Engine UI. You will then pick your scenario (Austerlitz) and begin playing/testing after you prove you have the module (for bring-your-own copyright workflow). No git, no download.
I will deploy this later today I hope. But the actual AI training for this scenario will take much longer than training on Chickamauga and Arnhem. Until then Austerlitz will have a very basic AI enemy
Quite a while ago I also gave an AI maps and rules with the purpose of creating custom Java code for a Vassal module. The idea was to enable highlighting hexes in LOS and hexes you could move to.
Anyway it suddenly dawned on me, that when you upload stuff to an AI, it will retain it and incorporate it into its training/knowledge. And this might well be a copyright infringement - unless you got permission of course. It is probably not a problem if you run a local AI model.
I am not a lawyer, but it might just be something to consider 
This is what the legality engine does now. LOS, movement possibilites, and weapons range
Yea, but if I understand it correctly, it does it in a browser, not in Vassal.
Anyway, my post was about copyright, not about coding or AI as such.
Yes you have it correct. But to see the browser version, the tester/user must legally possess the vassal module