Christian Schmidts & Prof. Daniel Hromada
daniel@udk-berlin.de / c.schmidts@udk-berlin.de

presented at

UdK Berlin Campus Kollisionen 2025, Aula Medienhaus, 6.1. - 10.1 10:00 - 16:00

Outcome

A symposium during which "artificial avatars" (AAs) of Your making will discuss with us humans the question "What is art ?"

Synopsis

DAY 1 :: 6.1. :: Introduction to technologies we will use: voice cloning, fine-tuning of large language models, talking head generation, Unreal game engine, github

DAY 2 :: 7.1. :: Bringing an exemplar AA into existence

DAY 3 :: 8.1. :: Bringing other AAs into existence

DAY 4 :: 9.1. :: Making AAs communicate among themselves & with us, room preparation

DAY 5 :: 10.1. :: 1st UDK.AI Symposium on Artificial Avatars

Day 0 :: 6.1.2025 :: AE550106

Today we will tell You what this Kollisionen seminar should be about, what technologies we will use and where do we will, ideally, get...

Introduction

Symposium

Definitions

fine-tuning & optimizing

dall-e%203%3A%20%22neo-victorian%20nanotech%20artifex%20john%20percival%20hackworth%20fine-tunes%20a%20large%20language%20mode%20for%20%20his%20solarpunk%20young%20lady's%20illustrated%20primer%20artefacts%22

dall-e 3: "neo-victorian nanotech artifex john percival hackworth fine-tunes a large language mode for his solarpunk young lady's illustrated primer artefacts"

During today's workshop, I will introduce You to two methods how You can fine-tune Your model to allow it to provide better inferences for the task You want to solve.

Lora training :: here You just have a bunch of data

Direct Preference Optimization :: here You have data which contains list of tasks (e.g. questions) and associated correct and incorrect answers

(Teacher) Avatarization

In the context of Artificial Intelligence in Education (AIED), teacher avatarization is the process of transforming the dataset produced by one concrete human teacher (or a precisely-defined group of human teachers) into a multi-modal generative artificial intelligence system with which future generations of learners can seemlessly interact.

Generative KI

Generative künstliche Intelligenz (KI) widmet sich der Erstellung neuer, oft unvorhergesehener Daten oder Inhalte, die das Ergebnis des Lernens aus bestehenden Daten sind. Diese Modelle 'verstehen' irgendwie die Struktur und Verteilung der Daten, auf denen sie trainiert wurden, und versuchen, neue Muster zu erstellen, die mit diesen erlernten Mustern übereinstimmen. Generative Modelle können für verschiedene Zwecke verwendet werden, wie zum Beispiel die Erstellung von Bildern, Texten, Tönen oder Videos und werden oft in Bereichen wie der künstlichen Inhaltsproduktion oder Sprachsynthese eingesetzt.

Neuronale Netze

Neuronale Netze im Kontext der KI sind keine echten Nervenzellen, sondern softwarebasierte Modelle, deren Architektur von der Art und Weise inspiriert ist, wie das menschliche Gehirn Informationen verarbeitet. Diese Modelle bestehen aus "Schichten" von Datenstrukturen, die "Neuronen" genannt werden. Die Neuronen sind miteinander verbunden und ihre Verbindungen haben ein bestimmtes "Gewicht". Im Lernprozess passt das System diese Gewichte - auch "Parameter" genannt - allmählich an, um den Fehler zwischen seiner Vorhersage und dem, was vorhergesagt werden soll, zu verringern.

model

A trained neural network is stored in a file. This file is called "a model". When You want to use it - either for inferencing or training or both - You need to load the model from the disk into memory.

Based on the amount of "parameters" (e.g. numbers which represent the synaptic weight) the model contains, the process of loading into memory shall be or shall not be succesful ;)

training & inferencing

In machine learning and AI, we speak about

"training" when the AI is learning from the data we provide it

"inferencing" when the AI is using its current "knowledge" to solve new problems (e.g. problems which maybe weren't in the training data at all)

Technologies

0th Avatar

Participants

at%20least%20one%20non-human%2C%20non-artificial%20being%20will%20also%20attend%20the%20Symposium

at least one non-human, non-artificial being will also attend the Symposium

all open-minded individuals (that is: YOU) who wish to learn more about open-source generative AI

Christian Schmidts & Daniel D. Hromada

cybrid of romantic poet John Keats (Avatar issued out of 0th symposium)

AAs of Your own making

all species welcome !