iiinova — where your digital self keeps evolving

Yeah, I think it is mostly good with the languages I know, but does struggle with specialist vocabulary - again, you need some knowledge to safely use it.

000

As long as you know that this is implausible (as we would) and perhaps remember that the UKSC gave judgment in R(Spitalfields) v Tower Hamlets in 2025. If you don't, then you might be actively mislead into believing something that isn't true.

000

Is AI ruining our skills? Early results are in — and they’re not good

I think you might not be giving yourself (or skills) enough credit. It seems like it is more akin to a muscle, use it or loose it: www.nature.com/articles/d41...

nature.com

Reliance on artificial-intelligence tools degrades the abilities of physicians and software engineers, studies show.

100

Maybe, we've had access to many frontier and law specific models and I have not had good experiences. Even the ones linked to databases of court decisions were unable to tell me about recent, ie after 2022 cases involving principle of legality, instead insisting there weren't any - pretty damming.

100

Yeah, I don't know either. Even if it is, wonder if that is a self discipline we can generalise. Will people actually stop at proof reading (many already don't) and do we deskill ourselves into dependency (which the companies hope we will). At a societal level, it may well be a true slippery slope.

It is a much better and more efficient use of my time to just read the document - not to mention the added benefit of learning something along the way. Plus, I have the added assurance that what I take from it is actually the product of my own analysis and not computer driven text prediction.

100

Appreciate your perspective, as always, but we have clearly had very different experiences. It is sicophantic, reflecting back to me what the prompt suggests; critiques of writing/reasoning are bland and unspecific, sometimes hallucinated, often miss the point and always incomplete. 1/2

100

I can only speak for my field, and the models I tried: potential areas of application vary too widely. In academic law, it's a joke and actively harmful to students and academics. Those who think they are not actively deskilling themselves by using these tools are in for a rude awakening.

Instead, you get the most likely sequence of words in response to your sequence of words, following the probabilities extracted from underlining training data - everything else happens in your mind. I find it really troubling that people humanise and outsource their thinking to LLMs, esp academics.

220

It is really important to remember that LLMs are just probalistic text generators, not intelligent entities: they have no concept of reasoning or logic, no thoughts or message to convey to you, they don't analyse or even read anything you give them in any meaningful sense. 1/2

111

From someone who knows more about this than either of us do: M Kramer, Rights and Right-Holding: A Philosophical Investigation (OUP, 2024) p. 216-217. I think that makes exactly my point - they are not at cross purposes, Raz's and Hohfeld's conception of rights are in deep conflict.

000

Yes, but it does in Razian terms - really puzzled why this doesn't seem to land.

100

Thinking you have isn't the same as doing so successfully. But let's leave it here, I think the thread has out lived its usefulness to us.

000

Perhaps those things can be reconciled but not easily. Matt is doing interesting stuff here but plenty of it is contested and probably won't be resolved in the space of a bluesky thread. I think with that I'm going to take my leave.

000

It really isn't. Hohfeld says we only have rights where there are correlative duties on identifiable individuals. Raz says rights exist where A has an interest that is sufficient ground for imposing a duty. Nothing of what you've offered here explains that away. 1/2

200

Plenty of people seem to think it is more than a semantic difference. Kramer wrote an entire book about it and Hohfeld is typically categorised as offering a non-justoficatory account of rights to Raz's justificatory one. You are welcome to argue none of that matters but that's a contested claim.

100

From memory, Matt Kramer has also done some work trying to reconcile Hohfeld with Raz but he only succeeds in doing so by droping some (imho) crucial elements from their respective accounts.

100

It seems that article proceeds from the premise that rights are relational in the Hohfeldian sense. The problem is that this is not true for Raz - rights arise from fundamental interests and therefore exist prior and independently from any duties (something Hohfeld must deny).

100

I don't think that's right: they're both giving us an account of rights. Hohfeld approaches this in a strictly relational manner, Raz asks whether there is a fundamental human interest. Raz can thus give 'heard before adverse decision' rights status, Hohfeld, by your reckoning, does not.

100

Or idiosyncratic. Other theories of rights reach different conclusions (eg Raz Interest theory) - so unless we are all Hohfeldian I do not think you can state your claim in such absolute terms.

100

It can be both? I am not sure how Hohfeld gets you to something like 'hearing before adverse decision' is not a right.

100

How about procedural fairness, the example Aileen gave: the right to be heard before adverse decisions are made against us. Why is that not a duty on your account?

100

'Properly so-called' is doing a lot of heavy lifting here.

I think that is question begging.

121

You say to-may-to, I say to-mah-to. Sure sounds like a duty not to do something unless x applies to me.

200

Great example - maps on well onto Art 6/13 ECHR procedural rights.

000

Whether your latter claim is true really depends on what you mean by 'one of the goods'. That can be a fruitful debate, but it is not a foregone conclusion that the common law does not / cannot protect these. 2/2

000

A unanimous Supreme Court in Miller II would beg to differ, to give just one example. The government was not permitted to advise the King to prorogue Parliament without 'without reasonable justification' based on parliamentary sovereignty. That sounds like a duty to me. 1/2

200

You claimed the common law creates no specific duties on the state - seems like it does, no?

How does that account for legality, separation of powers, parliamentary sovereignty and a host of other constitutional principles? These are all creatures of the common law (at least in part) and don't make any sense when applied to you and me.

Jeremy Letwin @jeremyletwin.bsky.social · 02/09/2026

Reposted by Stefan Theil

If you're interested in the environment and human rights, consider coming to our conference at Lancaster Castle on 17-18th September. Registration link here: forms.cloud.microsoft/Pages/Respon...

163

Stefan Theil @stheil.bsky.social · 19/08/2026

Nobody has the time to do 200+ vivas, or listen to as many presentations - which is easily what it takes for larger papers. We do a fraction of those numbers for admissions interviews and the whole process still takes about 1-2 weeks of back to back interviews.

140

Stefan Theil @stheil.bsky.social · 19/08/2026

Agree but for lost universities it is a staffing and resourcing problem compounded by the dramatic drop in funding. Invigilated, written exams are great but take up more resources than online, viva and presentations take much more time and effort so not viable.

140

Stefan Theil @stheil.bsky.social · 13/08/2026

I was just pulling your leg - I hope that the taste at least was satisfying?

Stefan Theil @stheil.bsky.social · 13/08/2026

Yes, that's right - although I suppose it is all relative 😅

010

Stefan Theil @stheil.bsky.social · 13/08/2026

The much more important question is of course who got bamboozled into buying this overhyped product in the first place? (I think you're supposed to stick it in a grinder).

Stefan Theil @stheil.bsky.social · 06/08/2026

Interesting. I suppose the same logic applies: it is a probabilistic match between your prompt, input papers and underlining training data. Trouble is, nobody know what parameters influence that metric - and the LLM cannot give reasons that aren't themselves probabilisticly generated (so useless).

000

Stefan Theil @stheil.bsky.social · 05/08/2026

It doesn't have a independent metric for quality, it just probabilisticly responds to input. Because you know it's a bad piece you can push it to more closely approximating an "accurate" score - you would be none the wiser if you weren't an expert. That makes LLM use so hazardous for society. 2/2

Stefan Theil @stheil.bsky.social · 05/08/2026

That's precisely the problem: the LLM just tells you what it reckons you want to hear. It will respond to you suggesting overlooked "negative" elements by lowering assessed quality. I am almost certain you would have achieved the opposite effect by focusing on overlooked "positive" elements. 1/2

If we cannot discipline ourselves in academia as established academics then we're doomed. We're incredibly privileged to get paid to think, that's a responsibility (and an expectation) we have to meet.

020

It might be a force multiplier, for sure. My point is that even if it is, we shouldn't use it. More work and more 'thinking' isn't better work or thinking. Especially when slop and the mondane are omnipresent, we need to insist on the slow and careful thinking that we're actually good at.

020

Spell check is fine because I have already picked. But letting an LLM write an abstract for a paper, or even clarify/make more concise the wording in a sentence is not: unclear, waffly sentences are what tell me that I haven't fully understood/thought something through yet. The thinking isn't done.

040

Agree, but don't think we need to get hung up on labels: AI is a misnomer (marketing gimic) anyway. We should reject tech that edits the physical manifestation of our thoughts: writing *is* thinking. I wouldn't let it edit my work because it is through editing that I think things through.

140