i
AE54 DDH ()


AE54

0th Berlin Symposium on Artificial Teacher Avatars

Introduction

λόγος - Word

φωνή - Voice

πρόσωπο - Face

Foreword to Machine Didactics

Prolog

HMPL

dall-e%203%3A%20%22illustration%20on%20black%20background%20of%20concept%20human-machine%20peer%20learning%20where%20machine%20learns%20from%20human%20and%20human%20learns%20from%20machine%22

dall-e 3: "illustration on black background of concept human-machine peer learning where machine learns from human and human learns from machine"

Human–Machine Peer Learning (HMPL) is a proposal that is positioned at the very frontier between educational, cognitive, and computer sciences. HMPL's core precepts which I introduced in my 2022 and 2023 papers are simple:

Humans and machines can learn together.
Humans and machines can learn from each other.

Vocabulary learning

Front. Educ., 2023
Sec. Digital Education
Volume 8 - 2023 | https://doi.org/10.3389/feduc.2023.1063337
Proof-of-concept of feasibility of human–machine peer learning for German noun vocabulary learning

Reading acquisition

In the second Hromada & Kim (2023) article, we describe first, syllable-oriented exercise by means of which the Primer aimed to assist one 5-year-old pre-schooler in increase of her reading competence. The pupil went through sequence of exercises composed of evaluation and learning tasks. Consistently with previous HMPL study, we observe increase of both child's reading skill as well as of machine's ability to accurately process child's speech.

Teaser

Next talk:  Make Your Own Not-so-large-language-model  @ State Of The Art(s)_ GenAI Applied _
July 5th from 17:00-18:45 at Gallerie at Medienhaus (Grunewaldstrasse 2)

Keywords: Large Language Models, Low Rank Adaptation, Retrieval Augmented Generation, Direct Preference Optimization, Embeddings

AIED 2022 paper

Au revoir, ECDF

Make Your Language-Model-Driven Community Engine

Mid-Sized Language Model(s)

A Mid-Sized Language Model (MLM) is a generative language model is an advanced AI system comprising of maximum 10 billion (miliarden!) parameters, organized into multiple layers with attention mechanisms. These layers process and interpret vast amounts of text data, while the attention mechanisms allow the model to focus on relevant parts of the input. This architecture enables the model to understand and generate human-like language, perform nuanced tasks like answering complex questions, writing detailed texts, and engaging in sophisticated conversations, leveraging its deep learning capabilities.

Adapting an MLM to suit Your needs

Essentially, You have three options:

fine-tuning (in every training step, training process updates billions and billions of parameters)

training a LoRa (instead of updating billions of parameters, You update just few millions)

do Retrieval Augmented Generation

'I'-Avatarization

"I"-Avatarization is the process whereby a living human H consciously creates, develops, fine tunes and optimizes (his|her) own generative AI avatar datasets & models.

That is, using datasets (mails, chat transcripts etc.) to create a generative AI copy of one's self (an "I-Avatar") which could provide information in situation when H (her|him)self is not alive anymore.

Demo

On a machine amsel.udk.ai (running somewhere in this room), there are many nice Generative AI tools installed, including:

text-generation-webui web-based interface for work with language models

Training PRO extension for training LoRas for Mistral-architecture models

superbooga-v2 for easy RAG prototyping

Enhanced Educational Environment

We define an Extended Educational Environment (EEE or E3) as an immersive and interactive XR learning environment enriched with AI-driven artifacts and avatars. These avatars and artifacts, developed using tools like UnrealEngine and MetaHuman, each possess distinct "personalities" or "characters" reflecting their underlying knowledge bases and machine learning models.

udk.ai

Join the HuggingFace udk.ai community !

Helmeto 2024

mid-sized LLMs

We use mid-sized (< 8 billion parameters) large language models derived from Llama 3.1 8B  and Mistral 7b base models.

On top of these models, we subsequently train specific adapters by means of Low Rank Adaptation (LoRA) methodology.

Additionally, Retrieval Augmented Generation (RAG) is also deployed in order to increase response accuracy.

NARRATIVE PRIMER: EMPOWERMENT THROUGH GENERATIVE STORYTELLING

 

Introduction

prompt%3A%20%22provide%20illustration%20for%20%22Once%20upon%20a%20time%2C%20a%20book%20had%20been%20made%2C%20a%20book%20which%20contained%20all%20the%20other%20books%20...%22%2C%20grayscale%2C%20in%20style%20of%20Gustav%20Doree

prompt: "provide illustration for "Once upon a time, a book had been made, a book which contained all the other books ...", grayscale, in style of Gustav Doree

Once upon a time, a book had been made, a book which contained all the other books, including its own construction & programming manual ...

Narrative AI in Education

Narrative AI in Education refers to the use of artificial intelligence to deliver learning experiences through storytelling, blending traditional educational methods with advanced, personalized digital tools. This approach leverages technologies such as speech-to-text, text-to-speech, language models and, optionally, image generation to create interactive, voice-driven narratives that adapt to each learner’s needs, preferences, and abilities. Unlike conventional screen-based learning, narrative AI fosters engagement through immersive storytelling, making education accessible also to children with visual impairment or from screen-critical communities. 

Narrative Primer Artefact

By integrating speech-to-text (STT), text-to-speech (TTS), mid-sized large-language models (MLMs) and sufficient and necessary knowledge base stored in the vector database, the NP provides a personalized and interactive learning experience. 

It emphasizes the traditional educational practice of storytelling, enhanced by modern AI capabilities, to promote basic literacy, arithmetic, musical skills and uncorruptable personality.

Take home lesson

spectacular things are already happening in "open source" branch of AIED

with USA gradually becoming the prey of the dark side, immediate deployment of "walled garden" approaches is of utmost importance

all "bricks " to build Your educational "cathedral" are available out there (GitHub, Huggingface) and ready to serve

the future will be more weird than a dream and the key to that dream is ...

... education