Tris Sherliker

..and why on Earth do so many LLMs think the answer is 'a spider'?!

[This]{data-ink=p0s0000 data-adv=1.568} [is]{data-ink=p0s0001 data-adv=0.453} [a]{data-ink=p0s0002 data-adv=0.507} [proof]{data-ink=p0s0003 data-adv=1.387} [of]{data-ink=p0s0004 data-adv=0.662} [concept]{data-ink=p0s0005 data-adv=1.827} [for]{data-ink=p0s…

I realised today that this: … is the vibe I get when people put their name and face to AI-generated output, and publish it as if it’s their own work.

When my wife said to me “Let me show you a t-shirt I saw…”, I wasn’t sure what to expect, but it definitely wasn’t an obfuscated bash script printed on the back designed to print a happy Easter egg message.

I’ve just been reading Filippo Valsorda’s well-written post about why, in the age of LLMs and automated security analysis, Vulnerability Reports are not Special Anymore.

I was using Claude to set up dotfile syncing before provisioning a new machine, and during the session I got this: > Updated plan User rejected Claude's plan: Plan: Exit accidental plan mode Plan mode was triggered accidentally mid-conversation.

I say on its Privacy page and homepage that BunTool doesn’t log any personal information.

There are only two hard things in Computer Science: cache invalidation and naming things.

My side project has been entirely handcoded in the backend, until today.

Working out the date of a web page can be useful (especially for prior art in patent cases, and for evidence in litigation), but it can also be difficult.

Whenever I find myself doing a task for a third time, it’s probably going to come up again and probably worth writing a script to get it done faster when it does.

I had a negotiation a while ago which very nearly derailed itself.