Stephen Downes

This is a detailed description of how to approach writing a specification for AI agents. Long story short: don't try to do it all in one go. Start with a high level description of what you want done, then work the AI to refine it into something comprehensive and robust. "Simply throwing a massive spec at an AI agent doesn't work - context window limits and the model's 'attention budget' get in the way. The key is to write smart specs: documents that guide the agent clearly, stay within practical context sizes, and evolve with the project."

This is interesting. "MELODY (Music Education for Learning Opportunities and Development of Youngsters) is an Erasmus+ project co-funded by the European Union with a mission that is both innovative and timely: to use the universal language of music as a powerful educational tool to enhance children's participation in democratic life." There's a project Handbook of Best Practices and a toolkit available on the project website. Related: How to solve the tenor shortage, via Chris Corrigan.

To be clear, what's interesting here isn't the software itself, which is "a rough proof-of-concept built for informational and exploratory purposes." No, it's that the software exists at all. A couple of nights ago I read on Mastodon, "I wonder if it's possible to build a standalone application that connects to Brightspace and analyzes all course materials and info using a local LLM and then provides a Coach to help me learn? (turns out, yes, and it works well)." This article from the next day describes the process, from the creation of a specifications document to the prompt to the application itself. While watching the Olympics. In one night.

Report authored at UNC Charlotte. As has become popular recently, the paper differentiates AI literacy, AI fluency, and AI competency. And it stresses four 'pillars' of comprehensive AI literacy: "understanding the scope and technical dimensions of AI, knowing how to interact with (Generative) AI technologies, being able to apply principles of critical, ethical, and responsible AI usage, and analyzing the implications of AI on society" (ironically whatever algo they were using spelled 'usage' as 'U.S.A.ge'). The authors stress the fourth pillar, arguing for "a systemic shift toward comprehensive AI literacy that centers human agency - the empowered capacity for intentional, critical, and responsible choice." Agency requires some preconditions: "True literacy involves teaching about agency itself, framing technology not as an inevitability to be adopted, but as a choice to be made. This requires a deep commitment to critical thinking and a robust understanding of epistemology." There's an image but my image uploader is broken today.

The further we look back in time, the more compressed it feels, and Maricopa Community Colleges' early innovations in open and online learning feel very compressed in the mid 2020s. It's hard to capture those days of the 1990s and while this e-book is reasonably comprehensive, describing the large number of initiatives that came out of the system, it doesn't really have the feeling of being there (that's not really a criticism, just a sentiment). So if you're reading this - and you should - you should supplement it with a review of Alan Levine's 2003 article that links to coverage here and here and even in this here newsletter. Via Alan Levine, naturally.

It always interests me to know when people talk of concept-formation in learning and intelligence just what sort of account of 'concept' they are using. Consider, for example, a 'cat'. Suppose we encounter one that is 25 feet tall. Is it still a cat? Is it a new type of cat? Or just an existing cat (a tiger, say) with extraordinary properties. Anyhow. This article (24 page PDF) takes these questions seriously, and in particular, examines Waismann's notion of open texture through the paradigm of the prototype theory of concepts, a theory that in turn evolved out of cognitive linguistics and can be contrasted with empirical and formal theories of concepts. In particular, it involves the idea that concepts can be open ended and vague, similar to Wittgenstein's 'family resemblances', such that (say) different entities can be more or less instances of a given concept (that is, being an instance of a concept is not an 'all or nothing' proposition). This paper is accessible and clearly written, and a good starting point for a serious inquiry into these ideas, if you're so inclined.