Andrew Mobbs
mobbsy.uk
Andrew Mobbs
@mobbsy.uk
Me:
• Father
• Software architect
• Cambridgeshire for past few decades, originally Moray.
• Boring middle-aged man

Also @mobbsy@mas.to
This was somewhat regional - Halloween was quite a significant childhood festival in (at least) the north of Scotland in the 1980s. However, the traditions were somewhat different, with children "guising" - wearing a costume and performing a party-piece (e.g. tell a joke) in return for treats.
October 31, 2025 at 9:23 AM
Surely "Fiat Lux!"
October 23, 2025 at 1:26 PM
I'm not sure why they claim that. If we believe OpenAI, a ChatGPT query uses 322μl water used per query. The 48.8 litres leaked per person per day is equivalent to 151,655 queries a day. I don't think AI is the big problem here.
August 28, 2025 at 3:05 PM
Pom Bears appear to be a mandatory part of any cafe's child lunch box.
August 21, 2025 at 10:41 AM
www.economist.com/britain/2025... discussed this trend a couple of months ago.

The audio edition performance of that article was... memorable.
Inverted commas are falling out of fashion
Blame James Joyce. And “fashion”
www.economist.com
August 15, 2025 at 12:31 PM
It's important to hit those KPIs
August 10, 2025 at 7:47 PM
Hence, my response to "the LLM can't count letters" is "here's how to get the LLM to count accurately" not "so LLMs are rubbish".
August 8, 2025 at 10:53 AM
I suppose I do - LLMs are a tool that have rapidly transformed HCI with impressive results. They enable automation and increased efficiency in a lot of use cases.

We do need to understand use cases where they function poorly, but to avoid or work-around those areas, not reject the tool outright.
August 8, 2025 at 10:51 AM
Threads like bsky.app/profile/kjhe... where people mock or argue with the LLM about the result it gave are pointless. (Except to show that AI slop amuses even people who claim to dislike AI.)

The GPUs aren't ashamed, they'll keep generating SoftMax values just the same as they did before.
August 8, 2025 at 10:11 AM
When did believing marketing ever work out well?

We know LLMs are bad at tokenisation, so if your problem involves tokenisation, use an approach that works with the tool.

If you're just trying to prove marketing is lies, well, I suppose it adds a little to the body of knowledge on that matter.
August 8, 2025 at 9:39 AM
Similarly, the rhetorical phrase "nobody is claiming" has become demonstrably false for every non-trivial example.
August 5, 2025 at 1:12 PM
Everybody on Bluesky has chosen to use social media site that is structured to minimise the sort of algorithmic feed that influencers are good at gaming.

As such, we are probably the sort of people who don't like This Sort Of Thing, so tend to dismiss it as worthless. (I agree that it isn't).
July 30, 2025 at 9:58 AM
Reading this post was the first time I heard that he'd died. RIP.
July 27, 2025 at 7:47 PM
The website misfunctions --More--
July 24, 2025 at 5:33 PM