Stephen Downes

I enjoyed this light treatment of subjectivism but as a refutation of the idea it fails utterly. The author traces its origin to the horrors of World War I, where people could quite rightly be justified in rejecting the idea of 'absolute truth' (one sees a similar response to the horrors of the Thirty Years War). But Arthur Krystal's association of subjectivism in philosophers with their emotional responses fails to justify the distinction between a trauma-free world of facts and trauma-fill world of emotions. You can't just say "the findings or conclusions of scholars and scientists existed independently of those who formulated them and those who interpreted them" and leave it hanging. Even if there is a 'reality' independent of our perception of it, there is an infinity of ways to understand that reality, and none of them has any a priori claim to being 'true'.

I think this article is mostly marketing for the National Academy for AI Instruction, "a five-year, $23 million partnership between the American Federation of Teachers and three of the largest AI developers - Anthropic, Microsoft, and OpenAI," but I also think it's an initiative worth noting. It's interesting to look at the article's use of language, for example, "We're in this race for teachers to get this knowledge" of more meaningful use of AI, said Randi Weingarten, the AFT president. "This will become the most disruptive technology in our time." And also, "A lot of teachers are doing this work at home, just wracking their brains." And also, "these tools, if gotten into the wrong hands, can be very dangerous for our students, for our profession, and for our jobs." And also, "there's still a lot of fear in the absence of federal guardrails on privacy, on safety, on disinformation, on academic freedom." This article is heavy heavy on the emotional language.

So what about LLMs training on free and open content? "I don't believe the answer is rejection," writes Hong Minhee. "I believe it's reclamation." The core question here, says Minhee, is that of who owns the models? "Who benefits from the commons that trained them? If millions of F/OSS developers contributed their code to the public domain, should the resulting models be proprietary?" OK, I take the point. But here's the thing. Why do we assume we're 'one and done' with models trained on open content. The content isn't going anywhere. Anyone can train models on this content (unless, of course, we lock it down, which would simply preserve the status of the existing proprietary models). We act today as though there could only even be one Anthropic Code or ChatGPT. But that's absurd. There are already many many properly open source models. We will be drowning in them! Open content ultimately means genuinely open AI - if we allow it to. Via Ben Werdmuller.

I have no idea whether this idea will take off, but I'm onboard with it. The idea is that people (real human people) who create their own websites can add a file called 'human.json' where they attest that their website is created by a human and vouch for other websites they know are created by humans. It should remind you of the old 'web of trust' and I now dub this the 'web of humans'. Using it is a three-step process: create a human.json file on your website (here's mine); put a link rel="human-json" into the header of your web page pointing to the human.json file; and then install the browser extensions for Firefox and Chrome to identify sites maintained by humans. Here are the full instructions. See also Alan Levine.

This trend is going to have a far greater impact on learning technology than most people realize, I think. Here's the trend: "Vibecoding, and especially the Claude Code style of vibe coding, is bringing people to create their own tools, who weren't able to do so before... Tools built by people realising they are pretty predictable to themselves, and that such highly localised and specifically contextualised predictability now lends itself to automation by the intended user themself." When people can easily build their own tools, what becomes of educational technology, which is in large part based on the authoring and sale of such tools? Se also a16z: Good news: AI Will Eat Application Software.

This is a bit of an ironic article for a website with the subhead "It's JSON all the way down" but it's still worth a read. Here's the gist: websites and structured data were originally defined using 'markup' languages like HTML and XML (these are the ones with all the angle brackets). But these specifications became more and more complex, with the peak of absurd complexity reached in Microsoft's OOXML. So some developers came up with an alternative tool, called markdown, that could translate some very basic easy-to-remember formatted text that into HTML. "You can learn it in ten minutes, write it in any text editor on any device, read the source file without rendering it, diff it in version control, and convert it to virtually any output format."