Jamie And Lion (aka, SpacedOutSmiles)
banner
jamieandlion.bsky.social
Jamie And Lion (aka, SpacedOutSmiles)
@jamieandlion.bsky.social
Monotropic semi speaking adult writing about neurodivergent play & adventures. MTB, digital a11y, AAC user, podcast host & lion tamer (he/him)
Erm… “Why not food if food shaped”… luckily he’s already been fed :)
November 13, 2025 at 9:58 AM
Not always. I’m using eurollm.io for the thing I built that helps me microwave stuff.

It runs entirely on my Mac, it’s 100% private, uses very little power. Was trained by EU universities on licensed data + funded by EU grant.

I don’t use the big commercial LLMs. The little ones are very helpful
eurollm.io
eurollm.io
November 13, 2025 at 9:56 AM
Hehe. Doing the same dance here. Zzzzz.
November 12, 2025 at 5:54 AM
It feels like one of those situations where it’s a principle not ‘use case specific’.

It’s the risk… and having to explain what you’re doing to protect the risks.

Side stepping it reduces the potential attach surface in the future & avoids hard chats with procurement teams.
November 10, 2025 at 11:10 AM
Is the URL cache control public?

Back in my networking days we’d sometimes see weirdness like this where equipment across the network would hit the URL as part of cache checking so one user request turned into dozens.

Wouldn’t explain 295:1 tho!
November 10, 2025 at 11:07 AM
Could be timing attacks. Attackers are smart as hell.

Sometimes it’s better to just not take a risk when we don’t need too.

Why risk opening the door at all when it’s trivial to side step the issue with good data separation.

It also avoids a complex chat / analysis at the paperwork stage.
November 10, 2025 at 11:00 AM
Off the top of my head:

- exploits for cache keys where other orgs data can be returned.
- if caches contain user data, a cache leak can expose activity patterns / locations / internal IP addresses etc

Keeping it separate reduces the risk of bugs causing harm considerably.
November 10, 2025 at 10:48 AM
They are org specific so keeping them separate means there’s an entire class of ‘data mixing bugs’ which can be side stepped… plus paperwork explaining the risks can be skipped.

If an orgs data is in a shared cache then the vendor needs to provide more evidence that it cannot be mixed.
November 10, 2025 at 10:39 AM
My guess would be for user privacy & making data protection easier.

The models are read only blobs so the only user data is the prompts. Keeping it all separated makes it easier to get sign off.

We do this inside ERMI for similar reasons. Data isolation keeps the paperwork simpler
November 10, 2025 at 10:34 AM
90% sure it does, though most employers probably don’t have a good policy.

This feels like it would be open to a challenge… but like most equality act stuff it would need to be funded… so it’s often easier not to fight this stuff :(
November 10, 2025 at 9:29 AM
Slightly cheeky take…

Social media is when people who think it’s the real world are trapped in a fake world which often does them harm.

Video games are when people know it’s a fake world, but they learn real world things which often does them some good.
November 7, 2025 at 9:51 AM
Squeee! Hope the move goes well :)
November 7, 2025 at 9:45 AM
That sounds like an amazing experience.

I’ve wanted to visit since I was a teenager - I learnt quite a lot of Korean from my taekwondo days! - but I’ve never made the trip happen.

It’s a good movie. Had me bouncing and smiling. Def listen on good speakers with proper base :)
November 6, 2025 at 10:11 AM
It’s very good! That’s the other track bouncing around my head.

I just saw they have announced a sequel has started production and due for 2029 or so :)
November 6, 2025 at 9:53 AM