Tracking instruction-tuned LLM openness. Paper: Liesenfeld, Andreas, Alianda Lopez, and Mark Dingemanse. 2023. “Opening up ChatGPT: Tracking Openness, Transparency, and Accountability in Instruction-Tuned Text Generators.” In Proceedings of the 5th International Conference on Conversational User Interfaces. doi:10.1145/3571884.3604316.
After the launch of the popular large language model ChatGPT in November 2022, Michele Murgia (education coordinator AI Convergence) and assistant professor João Goncalves (ESHCC) envisioned a project that would harness the potential of AI-driven LLMs but make it suitable for academia. The name of this project quickly became the Erasmian Language Model (ELM).
ELM is meant to depose ChatGPT from its position in education by offering a viable and better alternative. An alternative that’s open source, continually trained with a critical socio-technological perspective, safeguards privacy and labour practices and is based on community-wide participation. Anything students and teachers can do with ChatGPT (e.g. learn how prompts work, learn about factual error, critical reflection on technological output, produce rubrics etc.) can be done with ELM. In fact, ELM offers more educational and research possibilities due to providing insight into its model and data.
Sounds like the kind of project we'd like to include.
https://ecda.eur.nl/erasmian-language-model-elm/
Sounds like the kind of project we'd like to include.
Not entirely clear where the official codebase or model files are. I've filed a question in this repository: https://github.com/Joaoffg/ELM/issues/1