Welcome to DU! The truly grassroots left-of-center political community where regular people, not algorithms, drive the discussions and set the standards. Join the community: Create a free account Support DU (and get rid of ads!): Become a Star Member Latest Breaking News Editorials & Other Articles General Discussion The DU Lounge All Forums Issue Forums Culture Forums Alliance Forums Region Forums Support Forums Help & Search

WarGamer

(18,928 posts)
12. sigh...
Mon Jun 1, 2026, 11:32 AM
Jun 1

Some people are so confused about AI.

Example:

(Ask to Gemini 3.5) Who commanded the 3 spearheads of Operation Barbarossa in 1941?

(Gemini) Operation Barbarossa was divided into three massive Army Groups, each assigned a specific geographic objective and commanded by a senior German Field Marshal:

Army Group North (Heeresgruppe Nord) Field Marshal Wilhelm Ritter von Leeb

Army Group Centre (Heeresgruppe Mitte)Field Marshal Fedor von Bock

Army Group South (Heeresgruppe Süd)Field Marshal Gerd von Rundstedt

(Asking to Gemini 3.5) From which sources did you steal that data? A book like War without Garlands?

(Gemini) The argument that Large Language Models (LLMs) simply "steal" or copy from specific copyrighted books—like Robert J. Kershaw's War Without Garlands—fundamentally misunderstands the architecture of artificial intelligence.

An LLM does not operate like a search engine pulling up a hidden digital library, nor does it keep a repository of text files to copy and paste from. Instead of relying on a specific document, an LLM generates information by drawing upon a vast, synthesized compendium built from thousands of textbooks, public archives, and official war records.

Here is an objective breakdown of how an LLM processes and retrieves historical data:

1. Training is Learning, Not Storing
When an AI is trained, it processes millions of documents—including digitized military histories, academic journals, public domain records, and encyclopedias. The AI does not store these texts. Instead, the training process breaks the text down and analyzes how words, concepts, and facts relate to one another mathematically.

Once training is complete, the original files are gone. What remains is a complex web of statistical associations. When asked about a historical event like Operation Barbarossa, the AI isn't opening a digital copy of a specific book; it is dynamically reconstructing widely accepted historical facts based on the patterns it learned across its entire training data set.

2. A Compendium of Public Domain Facts
The names of the commanders who led the German thrusts in 1941—von Bock, von Rundstedt, and von Leeb—are not the intellectual property of any single modern author. They are matters of public historical record found natively in:

The German Federal Archives (Bundesarchiv)

Official Soviet defensive logs and declassified red army reports

Standard academic textbooks and universal encyclopedias

Because these facts are identical across thousands of independent historical sources, the AI treats them as a shared consensus of human knowledge. In copyright philosophy, there is a strict divide between facts (which belong to the public domain) and expression (the specific, artistic prose an author uses to describe those facts). An LLM synthesizes the facts without duplicating an individual author's copyrighted prose.

Recommendations

2 members have recommended this reply (displayed in chronological order):

No. QueerDuck Jun 1 #1
Why not? highplainsdem Jun 1 #3
A 50% STATE TAKEOVER of private companies is not the answer... it's overreach. QueerDuck Jun 1 #24
Ownership of an AI built on public data is just. It gives the public class action standing in federal court. ancianita Jun 1 #28
How does that help pay for the food, health, and shelter of the tens of millions of people who will lose their jobs Doodley Jun 1 #2
It isn't enough. Ideally, generative AI models trained illegally should be destroyed, with new legal AI models highplainsdem Jun 1 #4
I understand what you are saying, but the genie is already out of the bottle. This is where we are, and Doodley Jun 1 #6
Just using AI dumbs users down and makes them dependent on AI. There are few if any good uses highplainsdem Jun 1 #9
With all due respect, have you used it much? Doodley Jun 1 #14
I've used it enough to know how it works and how tempting it can seem. And I've read thousands of highplainsdem Jun 1 #19
With all due respect, I don't think you understand how incredibly useful AI is. Doodley Jun 1 #21
With all due respect, you asked AI to give you ideas and tell you where to shop. IMO that's an example highplainsdem Jun 1 #22
With all due respect, you are making a lot of assumptions. Doodley Jun 1 #23
My "assumptions" were based on what you'd told me. highplainsdem Jun 1 #29
Try all it's tech we paid for. In-Q-Tel funds or should I just say CIA. Blue Full Moon Jun 1 #5
I don't want to own it. I want it to not exist. WhiskeyGrinder Jun 1 #7
Agreed. MineralMan Jun 1 #8
Same here. But as I said above, in the meantime I like seeing regulations and penalties on AI suggested highplainsdem Jun 1 #11
YES! Emile Jun 1 #10
sigh... WarGamer Jun 1 #12
It isn't true that "Training is Learning, Not Storing" - that's AI company propaganda to try to defend highplainsdem Jun 1 #15
Thanks for the information. There is also a lot of people who have formed an opinion about AI Doodley Jun 1 #16
See reply 15. highplainsdem Jun 1 #20
Well, he's halfway there. BlueTsunami2018 Jun 1 #13
Bingo!!! Doodley Jun 1 #17
While everyone is debating the issue xuplate Jun 1 #18
With what? 40% don't have enough cash to cover a $1000 emergency expense. thesquanderer Jun 1 #30
Norway has a sovereign wealth fund on oil and they're almost always ranked at the top of the list of happiest nations. Uncle Joe Jun 1 #25
The AI bubble is on the cusp of a crash Bev54 Jun 1 #26
Didn't Cervantes Write A Book About Bernie? ColoringFool Jun 1 #27
Latest Discussions»General Discussion»Bernie Sanders: The Publi...»Reply #12