hometown 4.5.11+hometown-1.2.1
Serving on a programme committee in 2026, one is confronted with the effects of generative AI and it isn't pretty. Submissions that would previously be obviously under-baked now have a distractor layer of polish and puffery... full of fancy-sounding, superficially on-topic but meaningless phrases.
I will be predisposed to declining future invitations unless/until we can create stricter anti-genAI policies... not sure exactly what they would look like, but need to lower the oblox levels somehow.
Interesting article about, essentially, the corruptibility of academics. Many aspects are weak: the "ivory tower" metaphor just needs to die, and most academics I know don't crave "power" nor doubt the importance of their research work. Nor would most be impressed by private jets or Jaguars; anyone who dealt with Epstein deserves what they get. But there is appeal in a wealthy patron who might by-pass the bureaucratic straitjacket and its "impact-starving" qualities.
https://www.theguardian.com/commentisfree/2026/feb/20/epstein-files-professors-academics
A while back I stumbled on this nice description of a surprising feature of Linux: you can sometimes use /proc//mem to write to memory that is mapped read-only. https://offlinemark.com/an-obscure-quirk-of-proc/
What I had not understood is: "sometimes" means *when*, and "write to memory" means *which copy* of that memory? Clearly not "always" "the mapped physical memory"... this would be a gaping security problem if an unprivileged sure could write to literally any physical frame mapped by their process.
So...
Trying to put together a few slides about "data sovereignty" -- or, more generally, legal and social and ethical concerns of distributed systems (especially when they get ever larger).
Does anyone have readings or data to suggest?
E.g. I'd be especially interested in illustrating the historical decline of e.g. in-house or ISP-provided services, and the rise of the hyperscaler-backed equivalents e.g. Gmail / Office 365 / ...
Idle thought about source-level debugging: to counter the "bouncing around" problem, why not have two line tables:
one for the leading frontier i.e. we are at line N if we have started any of line N's operations;
one for the trailing frontier: we are at line M until we have completed all of its operations.
The "current position" is really a pair (leading, trailing), and "stepping" proceeds until a change in either of these. Once nice thing is that each one proceeds in source program order....
I gained a dim-ish view of ARIA from a few early interactions, but attributed this to personal grumps. Some time later, I had certainly been surprised by how much of their funding is going to US organisations -- though the ones I knew about (two, and neither mentioned in this article) were all plausibly deserving cases.
So, not endorsing the accuracy or fairness of this article at all, but it asks worthwhile questions.