Stephen Downes

I have the feeling Ian O'Byrne may conclude by the end of his series that this approach isn't the best way to manage your online presence, but that said, running a 'homelab' from scratch is no doubt going to teach him a ton about the tools we use. "At its simplest, a homelab is just a personal computing environment you control. It might be an old desktop running Linux in a corner, a tiny Raspberry Pi serving your files, or a small server rack quietly humming in the basement. It's a space where you can experiment, break things, fix them, and learn how digital systems actually work." As he sets up his self-hosting environment, he'll learn about the real issues developers face. "Ultimately, this isn't just about technology. It's about understanding the systems that shape us, and imagining how we might shape them in return."

According to this article, California Community Colleges (CCC) is partnering with AI company Nectir to launch an AI learning assistant for its 2.1 million students." The point-form article reads like it was AI-generated, though there's no indication of this on the web page. Still, according to the article, "One of Nectir's first pilots at Los Angeles Pacific University found that after a full term of using the platform, students saw a 20% jump in GPA, 13% increase in final scores and a 36% boost in their intrinsic motivation to learn."

Your new acronym for today is Agentic Context Engineering (ACE), "a framework that treats contexts as evolving playbooks that accumulate, refine, and organize strategies through a modular process of generation, reflection, and curation." As Robert Rogowski summarizes today, "It overcomes two major failures in adaptive LLMs: brevity bias (over-compression of prompts) and context collapse (loss of detail during rewriting)." Indeed, from my observation, a lot of the'AI hallucinations' we read about can be avoided with better prompts. On the other hand, it is arguable that prompt engineering is just a way of smuggling human knowledge into AI systems. Either way, though, it does show that the context of an inquiry has a significant impact on the outcome. 23 page PDF.

I like the way this article topples the 'wisdom' pyramid. Consider AI: it "doesn't climb a virtual digital pyramid from data to information to knowledge to wisdom... Instead, it processes patterns, generates responses, and creates what appears to be wisdom through statistical relationships between tokens." This raises the question: why do we suppose humans are any different? "Perhaps the hierarchy we've constructed is less about the nature of information and more about how we've chosen to categorise our own cognitive processes." Perhaps "the process of generating appropriate responses to complex situations doesn't require climbing a hierarchy at all... Perhaps the value isn't in the accumulation but in the generation; not in reaching the summit, but in developing the capacity to respond thoughtfully to whatever terrain we encounter."

This is quite an interesting article on how we understand AI, though it feels like it ends too soon. I recommend reading it from the perspective of human learning. What I mean is, imagine the interviewee, Naomi Saphra, is talking about humans, not AI. Most of what she says still makes sense. For example, "Just as biologists must understand an organism's evolutionary history to fully understand the organism, she argues, interpretability researchers should pay more attention to what happens during training." And also, "The model already wants to learn the easy thing. Your job is to keep it from learning the easy thing right away, so that it doesn't just start memorizing exceptions. That might make it hard to generalize to new inputs in the future."

Maybe it's just me, but how is the question "what is returned when you search for an educational research article on Google Scholar?" answered by studying "2,500 articles published between 2010 and 2022 across six AERA journals"? I get that the idea was to test Google Scholar by finding other versions of the articles that were't locked up. But even if you reference "AERA's role as a nationally and globally important organization," it's not like AERA comes even close to being 'all of education research' let alone 'all of Google Scholar results'. Nor is it representative of all educational research in any significant way. Anyhow, the article was published at Teachers College Record, where it's behind a paywall. There's an open version here (though who is ever going to find this version of the article?). These people in education, still publishing in paywalled journals... I don't know, I mean, I just don't know...