Stephen Downes

I agree with the argument made in favour of open source AI offered here by Tim O'Reilly. "I believe that the big labs are making the same strategic mistake that Netscape and Microsoft made in the mid 90s. Yes, make the models more reliable for ordinary users. But don't shut down the options for developers who don't work for you to push the state of the art forward. As Bill Joy put it decades ago, 'No matter who you are, most of the smartest people work for someone else.' No one should have a monopoly on innovation, and no one should be building a moat to hold it back."

What I like about this article is that it makes it clear that there is no single way to support what they call education for the poor. As the diagram says, "no single layer solves it alone". And it offers an effective counter to single-minded programs (such as the religious and charter schools recommended in Deseret News, via the always negative Joanne Jacobs). My view: There's a lot of data showing a correlation between socio-economic status and learning outcomes, which is why support programs, including dental and health care, sit alongside good pedagogy and access to resources as essential elements in educational equity. As I said in the very first few issues of OLDaily some 25 years ago, if you want children to do better at school, feed them. (The downside of this article: it's from a very sketchy source, Unlock Knowledge and Talent Upliftment (UKTU), "an independent human-vetter informational framework". I probably should never have linked to it, but now I'm here, I may as well pass it along as a cautionary note).   

It's important, I think, to understand that this article is focused specifically on recall of facts, and not cognition generally (though there are people out there who will no doubt conflate the two). It's interesting because of the way it describes five sources of failure of factual recall based on "one of five knowledge profiles: (1) encoding failure, (2) recall failure, (3) direct recall, (4) recall with thinking, and (5) inference without encoding." The authors "suggest that recall is tightly coupled to the conditions under which a fact was learned. When the query diverges from the training-time context, phrasing, or ordering in which the fact was encountered, recall becomes harder."

This interview (25 page PDF) is accurately described by the abstract, which frankly I would not recommend reading beyond (why link it here? for the record). "Tracing GenAI's origins from an Epstein-funded symposium and the Pentagon's chequebook to the eugenic roots of Silicon Valley's culture, Popenici contends that AI is the political project of 'technofascist' billionaires rather than a neutral educational tool." The article is historically dubious, narrowly focused, and conspiracy theory clickbait. 

This article describes the practice of sharding databases in order to serve more requests across larger data sources than a single database could handle. In sharding, each of several (or several hundred) databases contains a certain segment of the data. This division of data is defined by a simple function (for example, each database might handle only certain record numbers - the first handles records 1-10, the second handles records 11-20, and so on). A router then sends the request to the right database depending on the request. Multiple identical routers may exist, and traffic to them is distributed through a load balancer. It's a good article, and only touches the basics of sharding (there's no discussion of how search is managed, for example), but well worth reading for a detailed outline.

This article starts from the perspective of a personal AI that learns about its user over time. We'll call that the 'personal context' for lack of a better term, along with the author, Duncan Davidson. How do we create and manage a personal context? Davidson lists some things that matter: provenance, for example, to distinguish between what the user wrote and what they read. Scopes are important, as they distinguish between work, home and family. Identities and types matter, to allow for a consistent ontology across sessions. Time matters: my views may change over time, and what I said recently on a topic should take precedence (all else being equal). The system should be available across devices and across locales, and interoperability should not require uniformity: "Different people and companies should be able to build different context systems for different ecosystems and trust boundaries. What they need is agreement on identity, permissions, provenance, and exchange, not one universal application."