[September] Week 3 : Replacing myself using AI

I've been playing Armored Core, Baldur's Gate, and Starfield for the last few weeks. While studying these game extensively, I had an ingenious idea.

[Beginner's Guide to Dark Souls 3]

The beginner's guide series has had an unfortunate reputation of being a tutorial made to troll people with ridiculous glitches.

To remedy this unfounded notion, the next beginner's guide to Dark Souls 3 will use highly advanced AI technology which will both free up time to study other games while making quality videos that will help newcomers to Dark Souls 3. Here's how it works:

Instruction -> Guide

A new AI based on a custom fork of AutoGPT will take an input from the user.

e.g. Make a step by step guide on killing Gundyr. Explain like the target audience is new to the Dark Souls series.

This input will then be converted into a set of key locations, items, strategy, etc. These informations will be converted into in-game coordinates. Hence the "custom" part.

Guide -> Gameplay

Gameplay will automatically be performed using an AI controlled character.

While this seems like an impossible task at first, the game already has a very helpful system that does 99% of the work. The NPC invader/summon AI.

By simply hijacking the AI used for the NPCs, and placing the "player" marker on the other side of the boss fog gate, the player character will navigate, engage in combat, heal, etc all on it's own. Some modifications are made to pick up items and interact with in-game objects such as levers, NPC dialogue options, and doors.

The AI is not perfect. NPC phantoms have a tendency of eating damage with their faces which is the reason why they have so much health. And there are also situations where the program crashes, the recording software crashes, etc.

The AI is designed to check these types of fail states. Relevant information is fed to the first step and an improved set of instruction is used. The process will repeat until it reaches its success state.

Gameplay -> Video

The video footage will be fed into an editing AI alongside with the guide from the previous step. Timestamps of each action will be sent here as well. Each segment is weighted by importance, difficulty, and emotion. These stats are then used to choose the appropriate soundtrack for said segment.

The soundtrack is cut using the BPM and the gameplay is cut to fit this length. Appropriate subtitles are generated using calls to the GPT-4 API and simple edits like zoom-ins are automatically performed. There is an option to use AI voice lines(ElevenLabs). However this process requires human intervention as the result is not always perfect and is disabled by default.

After encoding the entire video the file will be uploaded on YouTube using an AI generated title(GPT-4) and a thumbnail(Stable Diffusion).

Of course, this is all 100% bullshit and I hope to troll more people using this premise as the intro.

Thank you for your support!