There are quite a few plausible (but not desirable) near future uses that I suspect there need to be anticipatory guard rails around - eg use of agents in critical safety checks, LLM-written instruction manuals - that may sound either trivial or unlikely, but anything seems possible currently

Ultimately looking at the application of AI in safety-critical systems, so cross-domain, but could eg be cyber physical. However, slightly more interested in whether quality assurance is actually possible in general purpose AI, or whether it needs to be context specific to be useful

thank you!!

thanks v much

Am very much leaning into some 2014 Twitter energy here but who knows! It might work!

- systemic sociotechnical approaches to safety testing and ongoing monitoring
- decentralised, frugal or small-scale approaches to model development

Basically, *not* your grandpa's AI, but approaches grounded in better outcomes for ppl and planet (assuming they exist). Is that you? let me know!

Two in the top 10! how exciting

I was being sarcastic I suppose. Theatre of cruelty disguised as govt policies

Excellent work all round

One significant reason I can't have a proper job is that all my best work happens between 10pm and 2am

Us English Literature types do tend to prefer a late evening over an early morning

preferable

Related to this, I'm speaking at the UK Internet Governance Forum today, on the challenges of fragmentation. But first, I suppose I need to get out of bed. ukigf.org.uk/2025-agenda/

faCt. I cannot, it seems, be vaccinated against prolific typos

I mean, ngl, it actually took longer to undertake the labyrinthine and ritualistic process needed to make the appointment than it did to get the three vaccines but I guess that me another matter entirely

And an anti-trans activist.

oh, that is a very sad thought

the chill wind, it whistles through my bones

I mean, I know it's a bit ridiculous but you can only really have respect for someone who is so committed to living their best life every day