Stephen Downes

OK, let's leave aside that the prediction (that everyone will be a full stack engineer in 2025) is absurd. Real developers aren't using Vercel or Replit, and there are still serious limitations to what Claude or ChatGPT will code. What about Marc Oehlert's prediction in this shirt article? "What if the systems get smart enough so that we don't need anyone creating compliance training content anymore? I mean we will but it'll just be Brad in HR bringing up his AI assistant to create the new course on ethics, or sexual harassment, or information awareness." Take it one step further - why do we need Brad in HR to be doing anything? What can't (potential) students just create their own courses? This will happen. Not by 2025. But soon.

This article takes some very different articles and weaves them into one coherent narrative. The fabric stretches a bit thin at points but the article overall succeeds. One is Yann LeCun's argument that after language models we will need 'world models' for reliable AI. "These systems would need to learn through direct interaction with the world, much like humans do." Quite right. Another is Tiera Tanksley's proposal for a critical race algorithmic literacy (CRAL) that leads Carlo Iacono to propose a framework for more ethical AI development, which would combine these. The article also mentions Steve Solomon on environmental concerns and Hartmut Neven on Google's 'Willow' Quantum computing chip, all (I think) from the 2024 Hudson Forum. Image: Insta.

This is what Wikipedia would call a stub - it introduces an article with enough content to make it register on search engines, but with the full text hidden away behind a paywall. It's also a very lazy article - "a randomised experimental study in the lab setting" - suggesting  "ChatGPT can significantly improve short-term task performance, but it may not boost intrinsic motivation and knowledge gain and transfer." Sure, students do better - but what about these vague and unmeasurable concerns like 'motivation' and 'metacognitive laziness'. I'd hide the paper behind a paywall too, lest anyone actually see it. I passed over an equally lazy article earlier today warning that AI agents may result in 'cognitive laziness'. It was about as sound as this formal academic publication. 

As a rule, I don't run large language models on my laptop, even though I could. I also don't run web servers on my laptop, for the same reason. Not that either would not be a nice feature to have, but we're not there yet. Still, as Simon Willison reports that "Meta's new Llama 3.3 70B is a genuinely GPT-4 class Large Language Model that runs on my laptop," we can certainly see the potential. And, he adds, as these powerful models get smaller and smaller, the worry that there may be an 'AI plateau' seems to be misplaced."

I'm not sure whether this post will be here tomorrow so I'll quote it at some length: "izzzzi is an experiment which might be called "slow social media" where we are exploring a multitude of constraints imposed on the standard mechanism of people making posts:

• posts are collected into a digest once a day.
• posts from yesterday are deleted, forever, every day.
• posts are a draft and can be edited until the moment that yesterday is deleted and tomorrow becomes today.
• posts are only visible between people who "add" one another ("mutual follows").

it's become a sort of collaborative daily newspaper written by friends. on the surface the parameters feel antithetical: it's ephemeral, it only changes daily, hardly anyone sees it. this is precisely what makes it interesting." It's one of these things that may flare up for a bit and then disappear. Like its posts. Via Ed Summers.

This is a presentation of multiple AI literacies, influenced by Doug Belshaw's work, describing them as: cultural AI literacies, cognitive, constructive, communicative, confident, creative, critical, and civic AI literacies. Each of these describes a specific aspect of a person's skills or competencies, for example, 'constructive AI literacies describe "understanding what it means to construct, build, or make something within AI-enabled environments". Similarly, 'cognitive AI literacies' involve "developing the skills necessary to navigate various AI environments to build knowledge and understanding effectively'. It's a useful way to look at AI literacies and makes me rethink again my approach to literacies generally.