God's work

Because I've been doing some work with it, I've written up an applied primer on optimal transport/transportation theory and its applications to measuring dataset distance: philswatton.github.io/2026/04/08/a...

Hopefully of use to someone else! Grateful for corrections on typos or misunderstandings

Trump threatens to wipe out Iran’s ‘whole civilization.’ EU has no comment. “Honestly if we reacted every time somebody posted something outrageous online, that is all we would do,” an EU official says.

When I think of the notion of 'living in truth' as expressed in Havel's essay and Carney's speech, responses like this are a choice to keep putting up the sign in the shop window

www.politico.eu/article/dona...

However, a fundamental limitation is that extant benchmarking approaches fail to consider realistic use of AI-generated text, including simple edits or variations in decoder strategy. They do not capture real-world performance.

The review concludes with recommendations for future benchmarks. 5/5

5. All detection solutions are highly vulnerable to simple perturbations and edits, and this is not reflected in most benchmarks
6. Among proprietary solutions, Originality appears promising but lacks sufficient evidence and likely suffers from the same limitations as other approaches

4/5

3. There is some evidence for successful in-domain generalisation to new models, but this is inconsistent and may depend on model family
4. Generalisation to new domains however remains difficult

3/5

The review was based on aggregating results from existing benchmarking efforts. We identify some core results:

1. SoTA detection depends on in-domain transformer encoder classifiers
2. Detectors perform best on the training data + models they were trained on

2/5

First part of report cover page containing title saying "Detecting AI-Generated Text" followed by "Informal Literature Review" and the first paragraph of the executive summary.

Another one of my reports written at the Turing is now available online, which is an informal literature review on AI-generated text detection. A brief thread as I've seen some discussion on the topic ⬇️

Link: www.turing.ac.uk/sites/defaul...

1/5

Hoping the reviewers were keen on it!!

I'm good thank you, hope you're doing well too and still enjoying the new job

Sigh...

Was bound to happen eventually despite the reams of evidence, disappointed to see it was the FT though!

This is a great point: anyone who's not been plugged into AI discourse and encounters contemporary models is going to get Eliza'd into Yudkowski-adjacent views on AI safety

Think this is basically correct, albeit some are making educated guesses while others are just making guesses.

Will v much keep an eye out! I implemented a bunch of these measures a while ago in the R package I started working on along w/ the basic VfS measures. Should get back to this at some point!

It's a remarkably general measure - v nice list of equivalences on the Wikipedia page (including one discovered by you!)

Guy who argues this aged well because in formal logic, X->Y is true whenever X is false

There was a lot in the review I found interesting and worth further reading or borrowing: his defence of the Enlightenment as an ongoing project, his notions of communicative rationality and deliberative democracy, his opposition to visionary thought. He seems to me to have left a very rich legacy.

Jürgen Habermas Showed What Philosophy Could Be The death of Jürgen Habermas has left philosophy and the Left poorer. Central to his work was a profound critique of irrationality in all its forms. Taken seriously, his philosophy provides an indispe...

Great obituary of Habermas by @mattpolprof.bsky.social: jacobin.com/2026/03/habe...

I only recently learned of the outline of Habermas' thought after reading Matt's recent review of his career: www.liberalcurrents.com/also-a-revie...

The ref vote was against the status quo, but without a clear alternative as part of the vote. I think a lot of the mess afterwards was due to this

I also think something to be said that parliamentary voting dynamics set a specific alternative against the status quo, which sometimes results in latter being preferred even though a majority don't like it (eg House of Lords reform).

(And I'm conscious I still need to write a response to your immigration piece -- its been hard to find the time!)

I've come to feel that referenda are a bad fit for parliamentary democracy. No easy way for the public to change its mind, not clear how to balance election vs referenda outcomes, etc. So PV critique I agree with, but I think q of what opposition means on a referendum context is difficult.

I ended up supporting a second ref in part because soft brexit was squashed as an option, and in part because of the inflexibility of leavers on Interpreting brexit as a hard brexit despite specific promises made by the leave campaign.

It's wrong to undermine an electoral outcome, but I am under no obligation to start supporting the policies of the winner, nor can it remove constraints on policy. The same is true of a referendum -- nothing in the result should prevent me from campaigning for its reversal.

But I think more concerning for me was that the referendum result became a way to browbeat opponents, especially under Johnson (proroguing parliament, removing the whip from anti-hard Brexit MPs). The refusal to consider soft brexit a legitimate interpretation of the result was troubling.

I wrote a blog post around the time of the indicative vote (I think I was doing my masters) arguing that remainers ought to support a soft Brexit, and that pursuit of maximalist outcomes was a strategic mistake.

This was a very enjoyable and thoughtful review.

I had a similar but trajectory with a different conclusion. I initially supported PV but came to dislike its populist elements. I also came to feel that the referendum vote had to be respected.

Fiasco - the People’s Vote seven years on A review of No Second Chances: the Inside Story of the Campaign for a Second Referendum

As promised, here’s my review of No Second Chances by @morganj0nes.bsky.social.

Spoiler alert – I really dislike the extreme, anti-democratic and counter-productive People’s Vote campaign.

www.thepathnottaken.net/p/fiasco-the...

Re UK vs US, it's only an effect on <0.2% in the latter if you assume disinfo was hyper-targeted only on those individuals. Otherwise it affect a larger part of the population.

Work mentioned above:
cetas.turing.ac.uk/publications...
cetas.turing.ac.uk/publications...

and in any case my main point was that I think the narrative of a stolen referendum was a failure of understanding on the left, rather than an argument about hesitating to take action vs attempts at disinfo.

2/3

Will try and respond fully when I get a chance but it's a busy week -- nb I agree re needing to take action (see work on AI election interference below where we suggest mitigations), but I think there is a clear body of evidence which points away from effect on *vote choice* ... 1/3