David Wiley has advanced his thinking of open education and artificial intelligence beyond 'open prompts' to a more broad-based Open Educational Language Models (OELMs) that "bring together a collection of openly licensed components that allow an openly licensed language model to be used easily and effectively in support of teaching and learning." He doesn't go deeply enough, however. He includes in his proposed stack 'open weights', which are the so-called 'open LLMs' offered by Meta, Mistral and IBM, followed by 'fine-tuning with OER' (whatever that might mean exactly) and retrieval augmented generation (RAG) with OER. But there's (at least) two more crucial layers, and so I've modified his diagram: open data, which is the content used to train the AI (a recent example is the ); and open algorithms, the part nobody (not even OSI) includes as part of 'open', which is the design of the core AI architecture. And because we must must must have a new algorithm, I call it Open Educational Artificial Intelligence (OEAI), pronounced 'aye'.
Today: 100 Total: 160 [Direct link] [Share]
