Google Notebook LM / WitE2 - Great Tool to learn the game

[terry1040]

I just discovered Google's Notebook LM (https://notebooklm.google/) as a great Learning tool for WitE2.

I created my own private WitÉ2 Notebook linking the 2 main WitE2 PDF documents "Living Manual V1.04.05" and "What's new" with about 10 advanced WitE2 tutorial videos from Matteo Bosso (who is running a highly recommended Youtube Channel about WitE2 [https://www.youtube.com/@fraulens])

Now the magic happened.

Based on my selected focus areas and questions the Google AI automatically generated amazing things:

⦁ An 45min Audio Podcast, where to 2 (computer generated) persons are debating about the best strategy as a Soviet Player for 1941. At any point, I am able to join their discussion and interactively ask additional questions on which the computer persons would react to. Now i can listen to an audio podcast about WitE2 while drving my car.
⦁ A 7min video podcast (coming acrosss similar to a powerpoint presentation) where my requested focus is about the best initial use of the soviet airforce.
⦁ An interactive quiz, based on all kinds of rules questions in a multiple choice approach, again with a selected focus if requested.
⦁ A set of aout 70 Flash Cards (that also could serve as quiz) to answer all kind of rule questions. It also provides further deep dive details on the answers.
⦁ A timeline, that summarizes about the main events during the games, like listing the event when the Soviet player can start forming up Guards Rifle Corps.
⦁ A mind map, that is automatically generated from the rulebook and provides a topic related structure of the manual content.

I AM TRUELY IMPRESSED. HIGHLY RECOMMENDED.

Cheers
Terry

[Toronja91]

Cool tool I'm going to look into it. I'm going to DM you my own notes so you can feed it into the machine lol.

[MarkShot]

About 1 year ago, I used OLLAMA and AnythingLLM to build a game expert from manuals, forums, YouTube, WiKis ...

The idea was to allow me to switch games easily that maybe its been a year since I played. The project didn't work out so well as the AI was having too many hallucinations. But if you use a large context (like 256K tokens) instead of RAG, then there should not be much model adlibbing.

This was done with a 4090 (24GB VRAM), but I plan to add a 6000 (96GB VRAM) just for AI. The 4090 is more than enough for games at 1080P/60. But with LLAMA.CPP as your inference engine, you can use both GPUs for an effective 120GB. At q=4, you can easily use models to around 140-180B. (More in the future using IKLLAMA.CPP which is now under development.)

PC games are my hobby in retirement besides AI. But I have decided to just rotate less games and not depend on AI as a memory tool. The first data is already in and heavy use of AI results in cognitive decline. Like the obesity epidemic in the USA, you are going to see 30 year olds with dementia in 5 years or so.

So, I do not intend to let AI do my thinking, but mainly explore the math and tech behind it along with the impact on the field of software engineering (my former profession).

Just a heads up for you.