Erasmian Language Model (ELM)
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.
What is the idea?
The idea behind ELM rests on actively democratising the means of AI production, because we ultimately want the labour by the academic community to be precisely for the academic community and possibly society at large – not the vested interests of commercial companies. ELM is accordingly a project that not only seeks to protect and foster public values from existing technologies that infringe on them, but also future ones.
The supporting team now consists of student assistants, PhDs and post-docs from various institutions and backgrounds, from computer science to law to the social sciences. ELM will be first employed in the minor AI and Societal Impact (EUR), the minor Computing Minds (ErasmusMC) and the Master Recht en Technologie (EUR). The Erasmus Centre for Data Analytics supports the technical infrastructure of the ELM.