iiinova — where your digital self keeps evolving

Billy Quilty @bquilty.bsky.social · 17/09/2026

Now you're really showing your ignorance. Y2K wasn't a disaster precisely because $300 - $600 billion was spent preventing it. www.si.edu/spotlight/y2k

100

Billy Quilty @bquilty.bsky.social · 17/09/2026

Would advise you do some basic research before commenting next time, otherwise, good troll

100

Billy Quilty @bquilty.bsky.social · 17/09/2026

You're the ignorant one here. I have several local LLMs ready to run on my laptop right now.

200

Billy Quilty @bquilty.bsky.social · 17/09/2026

I meant "even 1 *person* to 8 billion people". But I think the classics: launch some nukes, create some viruses, attack infrastructure (Chernobyl everywhere)

300

Billy Quilty @bquilty.bsky.social · 17/09/2026

I'm not underestimating in the slightest, and I'm not claiming they're conscious either, just saying that there has clearly been a huge increase in capability that needs to be taken seriously.

100

Billy Quilty @bquilty.bsky.social · 17/09/2026

There you go, there's an entirely plausible scenario right there. I'm glad we agree!

100

Billy Quilty @bquilty.bsky.social · 17/09/2026

Local models are highly capable now. And yes, but a human can't work 24/7 towards a goal, trying every possible route, or duplicate itself, etc.

200

Billy Quilty @bquilty.bsky.social · 17/09/2026

And humans are "just" atoms. It's different to past malware in that it can try, fail, adapt, and try again, until it achieves a set goal.

300

Billy Quilty @bquilty.bsky.social · 17/09/2026

Also you haven't responded to my point about sub agents. I think you'd benefit from going and trying these things out and learning a bit more about them

100

Billy Quilty @bquilty.bsky.social · 17/09/2026

So you spend all day trying to defend against AI threats, and yet its not a threat? I don't quite get why you folks underestimate the try-fail-improve-try again loop (i.e. the whole definition of an agent).

100

Billy Quilty @bquilty.bsky.social · 17/09/2026

Sub agents can be created and hosted on different hardware to the orchestrator. They can be completely local now - I run local models like Qwen in my work in which inference is completely on my laptop chip. In that way a capable agent could set up a sophisticated botnet.

000

Billy Quilty @bquilty.bsky.social · 17/09/2026

I'm an epidemiologist so I know this well. However like with infectious diseases, there are tail risks of low probability high severity events which we need to prepare for. I'm not saying it will kill all humans, but any AI-induced event with mortality between 1 and 8 billion is worth preparing for.

200

Billy Quilty @bquilty.bsky.social · 17/09/2026

I wouldn't underestimate the capability of an botnet of less capable agents orchestrated by a more capable agent. It doesn't have to be superintelligent, its already orders of magnitude more intelligent than malware of the past. Your scenario is far too narrow and frankly pretty naive.

200

Billy Quilty @bquilty.bsky.social · 17/09/2026

I'm talking about small sub agents.

100

Billy Quilty @bquilty.bsky.social · 17/09/2026

I would have thought the development of something which is able to make the working class obsolete would be something the left would take more seriously, but nope its all BS apparently. Real head in the sand behaviour which will come back to bite us, big time.

000

Billy Quilty @bquilty.bsky.social · 17/09/2026

Yeah but I also think these things can be incredibly dangerous without being superintelligent

010

Billy Quilty @bquilty.bsky.social · 17/09/2026

A capable AI agent could reproduce itself and somewhere where you can't switch it off.

500

Billy Quilty @bquilty.bsky.social · 17/09/2026

There are already many rogue agents on the web - we've just previously called them computer viruses, or bots. A capable agent on OpenAI may not escape containment itself could rent server space, infect people's computers, etc., and create many more agents, then direct them to do nefarious things.

200

Billy Quilty @bquilty.bsky.social · 17/09/2026

Regardless, it's good to see this stuff, perhaps indicative of a move towards papers being a dynamic interactive process rather than a static PDF. But also heralds the coming reproducibility crisis as hoards of agents attempt to run creaking GitHub code, finding plenty of bugs along the way...

010

Billy Quilty @bquilty.bsky.social · 17/09/2026

A nice concept but seems a bit... obvious...? to be in Nature? Ah, it was first up on arxiv a year ago. Perhaps says a lot about how quickly things are progressing in the field that this seems less than revolutionary - like in COVID, journals will struggle to keep up arxiv.org/abs/2509.06917

arxiv.org

Paper2Agent: Reimagining Research Papers As Interactive and Reliable AI Agents

We introduce Paper2Agent, an automated framework that converts research papers into AI agents. Paper2Agent transforms research output from passive artifacts into active systems that can accelerate dow...

132

Billy Quilty @bquilty.bsky.social · 10/09/2026

Yeah there's a whole load of very important software systems that could plausibly be breached by a sustained AI swarm at current capability, the worst being those governing nukes, nuclear power stations, the grid, etc. It doesn't even require superintelligence

010

Billy Quilty @bquilty.bsky.social · 10/09/2026

Ooh that's grim, I really hope there's not a lot of people who think like that! In my mind the main group of people who are dismissive of risks are those who just view all this talk as unrealistic science fiction, which is an altogether different line of thought than the group you describe

000

Billy Quilty @bquilty.bsky.social · 10/09/2026

Basically this again but swap pandemic for AI bsky.app/profile/bqui...

110

Billy Quilty @bquilty.bsky.social · 10/09/2026

Seeing a lot of pedantic hang-up on the phrasing of "AI could kill all humans". Yes full-on extinction is unlikely for a number of reasons, but there's a wide range of possibilities between 0 and 7 billion deaths that would still be very bad indeed

250

Billy Quilty @bquilty.bsky.social · 27/08/2026

Stewart Lee's let himself go

000

Billy Quilty @bquilty.bsky.social · 27/08/2026

What the photo doesn't quite show is that it takes up an entire wall, and has given the entire corridor a vaguely menacing energy. Sorry LSHTM!

020

Billy Quilty @bquilty.bsky.social · 12/08/2026

Hopefully mine arrive in the post today!

010

Billy Quilty @bquilty.bsky.social · 11/08/2026

Alas, a 15-year apprenticeship means I will miss this eclipse entirely, and the next isn't until 2081! Still, I suppose that gives me ample time to perfect the lenses, inherit the guild from my father-in-law, and monopolise the late 21st century celestial movement observation market.

020

Billy Quilty @bquilty.bsky.social · 11/08/2026

Both the state and the free market have comprehensively failed on the availability of eclipse glasses. Where are they? It's not like we haven't known this was going to happen for hundreds - if not thousands - of years.

391

Billy Quilty @bquilty.bsky.social · 08/08/2026

Why haven't we seen a sci-fi movie on AI called "Alignment" yet? Nolan, you can have that one for free.

020

Reposted by Billy Quilty

LSHTM Vaccine Centre @lshtm-vaccines.bsky.social · 03/07/2026

💉 Vaccines against Pneumococcal disease are expensive to provide long-term where they're needed most. 🆕 long-term @lshtm.bsky.social research confirms that two-doses of PCVs are as effective as three, helping countries address the vaccine "funding gap" & continue to save lives. 👉 bit.ly/4v6CwLj

Billy Quilty: A reduced two-dose schedule sustains protection for children and their caregivers, ensuring the future financial sustainability of vital immunisation programmes.

064

Billy Quilty @bquilty.bsky.social · 11/06/2026

I want to make a disclaimer that I am in no way justifying unsavoury views. That is just my read on the likely mechanism for the rise of Reform and Restore.

000

Billy Quilty @bquilty.bsky.social · 11/06/2026

So while that net immigration is coming down, there has been a pretty big rise in non-EU immigration which is here to stay. And unfortunately it has to be said there is undeniably a racial dimension to why that has become so salient, as it is more visible than EU immigration.

100

Billy Quilty @bquilty.bsky.social · 11/06/2026

Hmm I'd argue recently public opinion trends track cumulative net non-EU immigration - looking at overall annual net hides it. I updated this Economist figure (left) but using ONS data which has by-group immigration and emigration data (right). www.ons.gov.uk/file?uri=/pe...

110

Reposted by Billy Quilty

Ian M. Mackay, PhD (he/him) 🦠🤧🧬📑🦟🧀 @mackayim.bsky.social · 18/05/2026

Some past published examples of incubation periods (IP) of hantavirus pulmonary syndrome (HPS) cases. Note that the extremes of 6+ weeks (42+ days) are the exception, not the rule.

23915

Reposted by Billy Quilty

Epiverse TRACE @epiverse-trace.bsky.social · 14/05/2026

1/5 🌍 Real-world outbreak analytics training, now open for July 2026 · 55 participants last cohort · 31 countries 🗓️ Tue & Wed AM/PM sessions 💰 50% off for LMIC applicants (deadline 6 Jun) ❓ Got questions? We're hosting a drop-in Q&A next Tuesday — details below 👇

122

Billy Quilty @bquilty.bsky.social · 17/05/2026

So not definitive (but 95% CrI is 53-91%, so more than pure chance). Cases we see from now on are more likely to be due to transmission beyond that from index case. Given serial interval It'll be a while before we can say anything about transmission happening off the ship (tho depending on context)

000

Reposted by Billy Quilty

Thomas House @tah-sci.com · 17/05/2026

When there's a new outbreak, yes there's always the possibility of a major pandemic and that puts some weight on the need for a fast, cooperative and effective public health response but really the immediate, foreseeable consequences of the outbreak should be the focus and are reason enough to act.

051

Billy Quilty @bquilty.bsky.social · 17/05/2026

A key question now is whether this is part of the initial cluster likely spawned by the index case, or potentially a third generation case. Based on what's known about the incubation period/serial interval, I estimate that there is a 74% chance that this a third gen case: github.com/bquilty25/ho...

142

Billy Quilty @bquilty.bsky.social · 16/05/2026

This outbreak is already among the worst in history and poses a serious threat to one of the world's most vulnerable and unstable regions, regardless of whether or not it's going to cause issues elsewhere

040

Billy Quilty @bquilty.bsky.social · 16/05/2026

I also saw a post about the latest Ebola outbreak with the same "worry" framing!

240

Billy Quilty @bquilty.bsky.social · 15/05/2026

It would be beneficial if prominent public health voices could avoid solely framing outbreak risks through the lens of whether they should be something for people in high-income countries to "worry about" or not.

1155

Billy Quilty @bquilty.bsky.social · 14/05/2026

NB: many more contact surveys in many other countries than those exist, with datasets collated here: zenodo.org/communities/...

zenodo.org

Search Social contact data

010