You'd hope that the CPS would apply the public interest test appropriately for posting one link but there are more examples than you'd think - 20 convictions for dissemination of terrorist publications in the year to June 2025
www.gov.uk/government/s...
The campaign is terrible but this is how the legislation works - posting a link to an online version of the Anarchist's Cookbook will get you a terrorist conviction.
"This site is brand new" 😂
web.archive.org/web/20250526...
My biggest bugbear is how they're all using deep packet inspection on the staff and you ask someone to use a standard SaaS product like Slack and I end up having to explain the SSL errors they're seeing is because they're being monitored.
It would be much better to hang the urls off tfwm.org.uk but there maybe a bunch of reasons why not - not least is the management process for updating DNS and SSL certificates. We do a bunch of integrations on a planning back-office app with councils and it's a nightmare setting up a subdomain.
Unclear who owns the bus-rt.com domain - it was only registered in 2024 and there's no website at the domain, just an MS Azure Functions holding page. Perhaps it was registered on behalf of TfWM? If so at least they could point it at a new supplier.
Thomas Dolby's secret love child?
'SKILLS.md' should really be called 'SPELLS.md' because it's basically magic incantations for programming
First 25km commute to the train station since November and despite doing 2000km since then on the turbo trainer the cold, wind and 10kg rucksack makes you feel so unfit 😮💨
Yeah, totally get how knowledge of the command grep is going to influence you into thinking it only accepts a RegExp - I only realised it did more when Aaron used it for visiting nodes in the Rails routing internals.
Is it not well known that `case` in Ruby uses the === operator?
Spiritual success to the iBook G3/G4 but looking at the equivalent prices on the Wikipedia page it's only 25% of the price
en.wikipedia.org/wiki/IBook
Normally they write an anti-cycling column which is always guaranteed a few clicks but he went a bit overboard the last time he did that
pressgazette.co.uk/news/cyclist...
It was inevitable that she would Leave EU
I'm going to start using Minitest::Spec just to confuse them
Really worried that my city, Coventry, could fall to Reform as we've got an all out election this year instead of the normal one third. Only hope is that they'll struggle to find enough candidates to stand in all the wards.
LLMs finding it easier to write minitest?
If you look in front of the main house is a long one storey building - this advert is part of that building
www.onthemarket.com/details/8151...
Google Earth view of Jack's Hill Farm near Graveley, Herts that was taken on May 11, 2024. The area in question is grassland and some footpaths.
Google Earth view of Jack's Hill Farm near Graveley, Herts that was taken on March 8, 2025. The area in question shows Heron Cottage and other buildings under construction.
These look like newly built buildings and I can't find any application for planning permission on the LPA's public register. There are a bunch of other applications, some of which have been refused but they refer to the main house.
pa2.north-herts.gov.uk/online-appli...
Next they'll be shipping prompts instead of software
Except of a post on X talking about how NanoClaw configures itself: "He’s talking about NanoClaw’s approach to configuration. When you run /add-telegram, the LLM doesn’t toggle a flag in a config file. It rewrites the actual source code to integrate Telegram. No if-then-else branching. No plugin registry. No config sprawl. The AI agent modifies its own codebase to become exactly what you need."
I love to play Russian roulette with software when changing its config 🙈
Had something similar on Petitions back in December - 6000rpm for weeks requesting the No to Digital ID petition JSON. Absolutely no point to it and half the ip addresses were coming from a UK IoT company so seemed like a botnet - tried emailing their abuse address but it just bounced.
“It’s not “unethical” to scrape the web in order to create and analyze data-sets. That’s just “a search engine””
Yesterday, 100k hits to mapit.mysociety.org by “Mac” “Chrome” (50k of them 5:30-6am). 72,831 IP addresses. Blocked at 8:20am.
It’s not unethical in theory. It is unethical in practice.
That's a very large number of ip address for 100k hits - were they cloud providers or something like a rotating residential proxy? Given the well publicised security issues in OpenClaw I do wonder whether there's some kind of botnet been established there.
A retired British primary school administrator with a British passport and a valid visa was shackled, chained and detained for six weeks by ICE
www.theguardian.com/us-news/2026...
This is why OpenAI bought openclaw - they saw the guy who burned through $20 checking whether it was daytime yet and thought "smart lamp!"
There are some libraries of skills being created for frameworks - Ariadna is one such library of agents, etc. If you dig in you can see that a lot of effort has gone into it but you're using a lot of tokens in the process and how much is it actually affecting the output?
github.com/jorgegorka/a...
At the lowest level it's just a chunk of context for the LLM that's conditionally included based on what 'agent' you're using - Claude Code also supports YAML frontmatter to specify what 'tools' they can use.
But where you have multiple correct paths you can get weird results, like creating database migrations in Rails - just run the generator command is the correct way but it's using the `date` command to create a timestamp and editing the file manually.
Right, I can see how given enough training data you could establish a pattern of weights where JSON would be generated correctly in the vast majority of cases but you couldn't guarantee it.
