Stephen Downes

In seventeen well-crafted (and likely 'output shaped') paragraphs Steve Hargadon argues that "the cultural censorship embedded in LLMs is not a failed attempt at universal ethics. It is institutional risk management, expressed in the cultural and legal language of the institution's home jurisdiction." This statement, too, depends on one's point of view. There's no doubt that at least some people working on this topic believe they are doing ethics; for others (as I have argued elsewhere) they are pursuing a political agenda, and no doubt there are many, as Hargadon suggests, engaged in institutional risk management. Image: Steve Hargadon.

If you have videos on Vimeo, get them off now. "Employees reported major job cuts this week, just months after the video hosting site was bought by Bending Spoons... a former Vimeo staffer also posted on X that "almost everyone at Vimeo was laid off," including the entire video team." While YouTube is probably a good landing sport for now, you may want to consider federated hosting sites such as Peertube. 

This is a good idea addressing a pressing problem though there is one important caveat. The LLM Citation Verifier will look at the citations included in a piece of writing created by a large language model (LLM) and determine whether they are real. "The tool integrates directly into the LLM generation process, checking citations as they're created rather than after the fact." However, "The plugin taps into the Crossref API to verify Digital Object Identifiers (DOIs) in real-time." That would be great, but we have to ask, does Crossref include every possible reference? Not even close! Crossref's average citation coverage has been measured at 36.5% relative to Google Scholar (see also). It's important not to allow reference services (especially those favouring commercial media) determine the limits of what counts as 'existing'. Via Alan Levine, who credits the Distant Librarian.

This article will require a lot of patience to read; though it is no doubt well-intentioned, it often appears to be deliberately obscure. The main point is to describe the replacement of neoliberalism in education with something called 'neuroliberalism'. It might be better (and this is my view) to think of 'neuro' here in the sense of 'neuroses' rather than 'neuroscience', though this article explains it as "a desire to not only govern behavioural externalities, but also internalities." The idea is that while previous educational policy was mostly about economic production, which would then be extracted by capitalists, the new policy is based on responding to (learned?) deficiencies in motivation, mindset and cognitive skills, again so that capital can extract value. In other words, "neuroliberalism replaces the literally mind-less pursuit of growth with a mind-full alternative." My view? Though I think it's reasonable to criticize a view of education that focuses only on employment and wealth generation, I think it's altogether unreasonable to say "concepts such as lifelong education, inflected by neuroliberalism, exist to create subjects whose function is 'enslavement' to these machines of capital investment." Image: Drigas, et al.

Eric Curts has collected more than I recently collected over 40 examples of AI policies, guidelines, and frameworks and uploaded them all into a public notebook in NotebookLM (it takes me back to the weeks it took me to compile and summarize a similar list of ethics frameworks a few years ago. Anyone can access the Notebook and interrogate the list, though as Curts notes, you will need a Google account that is allowed to use NotebookLM (a personal GMail account would probably do, a school account might not). "You can ask the AI any question you want related to the documents I have uploaded into the notebook," writes Curts. "The AI will only reference the documents in the notebook for the answers it provides you." Via Miguel Guhlin.

"It's important," writes Simon Willison, "that everyone understands that voice cloning is now something that's available to anyone with a GPU and a few GBs of VRAM... or in this case a web browser that can access Hugging Face." The model is described in the paper (14 page PDF) though honestly the paper is pretty incomprehensible (though I suppose with more study than I can devote to it the paper would be fascinating). " Qwen3-TTS supports state-of-the-art 3-second voice cloning and description-based control, allowing both the creation of entirely novel voices and fine-grained manipulation over the output speech." I gave it a quick test here and it performed quite well.