Stephen Downes

My own view of mental imagery is that it is perceptualist, that is, "seeing and mental imagery are fundamentally the same kind of mental state, differing only in degree or in terms of cause." This paper looks at experiments conducted in 1910 by C. W. Perky that many have taken to show that perceptualism is true; there's no difference between our perception of a thing and an image of a thing. As Nigel Thomas writes, "although perception is usually more vivid... than mental imagery, the experiment appears to show that this is, at best, a mere difference in degree." In this paper, Dan Cavedon-Taylor argues against this interpretation of Perky's results. "Given the deceptive nature of Perky's experiments, we cannot be confident that introspection and genuine indistinguishability was responsible for subjects' reports, rather than task demands." Maybe, but I think there are many other phenomena that lead to the same conclusion - the phantom limb syndrome, for example - where no effort has been made at all to fool the perceiver.

Mozilla CEO Anthony Enzor-DeMeo writes, "we are in jeopardy of a handful of closed-model companies controlling the browser, the search engine, the AI model, and the surrounding services that connect them." The intent here is to "offer an alternative to global tech monopolies controlling the next digital era." Fair enough. To make this work, you first download the Mozilla Smart Window, which integrates with the firefox browser. When turned on (because you can always turn it off) it gives you an 'AI panel' alongside the web page you're viewing. This panel is powered by an AI model you select - the agreement with Mistral, a French open-weight AI provider, makes one of their models one of these options. What's interesting is that it learns from and remembers what I've been browsing, so it can tailor results to that context. I'm trying it now as I write (it is not writing for me) - it got my first question ('who is the author of this page?') wrong.

Allocentric navigation "involves using external cues and environmental landmarks to build a mental map, allowing individuals to navigate from a bird's-eye perspective" and is contrasted with egocentric navigation, which "relies on self-centered cues, such as body orientation and movement." The point of this article is to summarize reserach (15 page PDF) showing that allocentric navigation is turing complete - that is, anything you can do with logic or computation, you can do with an allocentric navigation system. This is important because "cognitive maps do not normally store and process symbols in any linguistic or language-like sense." So "symbolic processing therefore need not be treated as architecturally basic but may instead be reconstrued in terms of map-like resources already known to exist in the brain."

This article offers "a small collection of useful CSS techniques and a quick reminder that print style sheets are still a thing." Now I haven't pinted anything from the web in maybe a decade, but I do know people who do. Specifying a print style shgeet ensures that they get something readable, not a mess with pointless navigation and other formatting meant for web users. Print stylesheets can manage things like page margins, page breaks, widows and orphans, and URLs you can see. Even if you're using AI to design your page, you still need to tell it to include print styles. So this is a reminder worth taking note.

"With AI watermarking," writes Laurens Hof , "inference providers mark their output as AI generated." But there might be hundreds of different AI providers, each with their own key. So "if someone encounters a piece of content on social media, they will have to go to 100 different systems, each run by the inference provider, to check if it contains the watermark of that specific provider." This is not practical, obviously. And it still depends on the goodwill of the AI generators in question to (a) actually watermark the content, and (b) not use the system for some other purpose, such as advertising or surveillance. Good article that expands on the argument in detail.

The story is in the headline - "The podcast I was booked on turned out to be a social engineering scam to get me to install malware." Normally I wouldn't link to a LinkedIn post but I think the lesson is clear and urgent: "Never, ever install custom software for a meeting or podcast." Joan Westenberg described in her post how "their prep was impeccable w. better interview questions than most journalists" and how "they sent screenshots of themselves 'on the call' to pressure me." Here's a Claude analysis of the software they asked her to install.