hunemoL

"Power of the community: humanity needs more language models!"

Motivation

"The goal of the Socratic dialogue in the form handed down by Plato is the common insight into a situation on the basis of question and answer." Sense and method of Socratic dialogues

Current state of the AI ​​infrastructure

1. a confusing range of AI-supported services for the average consumer
2. Operation of language models - training and surveys cause significantly higher CO2 emissions compared to search engines
3. human characteristics are attributed to the AI, keyword "hallucination", see [7].
4. public service media (Öffentlich-rechtliche Medien {ÖRM}) have extensive archives with which public service language models can be trained

Thesis 1: The voices of the decision-makers from the bodies of the public service media (ÖRM) are missing in the reporting on the subject of artificial intelligence.
Thesis 2: It is desirable that the decision-makers in ÖRM committees enlighten the general public as to whether ÖRM editors want to further develop their work and structures in view of the AI ​​phenomenon.
Thesis 3: The training of language models with content from ÖRM archives and from library stocks can make a significant contribution to the development of the identity of people and nations.

Call to Action

hunemoL network

1. Establish "Federated learning"-based language model infrastructure
2. Training language models, professionalizing quality assurance during training
3. Data owners train their own hunemoL instances with this data. This ensures a better quality of the language model compared to a centralized language model approach
4. hunemoL instance XY parses the prompt text and suggests one or more hunemoL YX instances for generating the response
5. Symbiosis: Instead of trying with a lot of effort to teach the machine the ability to "understand" the meaning of the language, to precisely distribute the tasks between man and machine in order to develop a sustainable, efficient, solidary hunemoL infrastructure
6. Leading public discourse on the role of language models as part of public service media
7. Display environmental pollution caused by AI infrastructure transparently

People's receiver and transmitter "hunemoL"

The goal

The goal is to realize "hunemoL" as a prototype of a chatbot, a text-based dialogue system equipped with "federated learning" features. "hunemoL" aims to provide a low-threshold, user-friendly AI infrastructure for the general public. "hunemoL" users are licensee payers , listeners and viewers, but also journalists and members of the broadcasting council of the respective public broadcasters.

hunemoL as a human-machine symbiosis

Use the strengths of the hunemoL participants and the possibilities of technology for the development of a sustainable, efficient hunemoL infrastructure in a targeted manner.

The language model as a cultural and social project

The provision of a low-threshold accessible hunemoL infrastructure can revolutionize the media landscape and accelerate communication processes to an even greater extent than the invention of book printing with movable type by Johannes Gutenberg .

hunemoL connects generations - in real time

• hunemoL is a mouthpiece, a medium that can pass on the voice, the moods, feelings, the messages, the knowledge of the older generations to the following generations in real time and keep them for the future
• With hunemoL the following generations get low-threshold access to the knowledge, to the wisdom and also to the emotional universe, to the creativity of the older generations in real time and for the future

and all of this in a sustainable, cross-border and language-independent way.

"hunemoL" life cycle

• "hunemoL" is being developed as open source software on behalf of public service media (ÖRM) . Project {hunemoL}. Just as, for example, the construction of transmission masts are commissioned by ÖRM
• "hunemoL" can be downloaded by the user and set up as a personalized hunemoLXY variant, a hunemoLXY instance

• "hunemoL" implements an API that allows users to:
  • rummaging through current programs and in the ÖRM archives
  • low-threshold communication with other hunemoL users
  • that hunemoLXY instances can learn from each other
  • Order notifications about upcoming broadcasts according to his/her interest profile
  • Order ratings reports for specific broadcasts.

• Translated with the translation function integrated in the browser Chromium version 112.0.5615.49 (Official Build) snap (64-bit).
• Deutsche Version im Flyer-Format - hunemoL: The humanity needs more language models!

bibliography

1. Swarm Learning und Federated Learning, 22.01.2023 - https://realtime.fyi/articles/git-for-future/swarm-learning-und-federated-learning
2. Föderales Lernen - https://de.wikipedia.org/wiki/F%C3%B6derales_Lernen
3. Federated Machine Learning - https://www.bitfount.com/pets-explained/federated-machine-learning
4. Federated Learning of Cohorts - https://en.wikipedia.org/wiki/Federated_Learning_of_Cohorts
5. The FTC should investigate OpenAI and block GPT over ‘deceptive’ behavior, AI policy group claims, 30.03.2023 - https://edition.cnn.com/2023/03/30/tech/ftc-openai-gpt-ai-think-tank/index.html
6. Hallucinations Could Blunt ChatGPT’s Success, 13.03.2023 - https://spectrum.ieee.org/ai-hallucination

License

HappyPony, 26127 Oldenburg unter CC BY-SA 4.0