The replication crisis replication crisis

If other news was reported like climate change:

I'm seeing this line that you shouldn't worry about AI taking your job, because we'll just sell more stuff cheaper (Jevons' paradox). But this assumes induced demand, which in turn assumes an economy where people have enough disposable income left to buy your stuff if only it was a bit cheaper

So this is interesting -- there's a website where you can check whether you're in the weights of the various LLMs. Note "MODELS MAY HALLUCINATE BIOGRAPHICAL DETAILS OR EVENTS THAT NEVER HAPPENED" intheweights.com

This might be one of the funniest chart about energy scenarios I’ve seen. You can hardly see “actual” for gas

I really need people to understand that Palantir is not an engineering marvel; it is a multi-billion-dollar contractual containment layer wrapped around the world's ugliest legacy open source frankenstack. It's a shitty Snowflake for companies that don't know how to do data engineering.

So, the reality of this is finally settling in: The @workersobservatory.org has been funded for ten years (!) by the Regenerative Futures Fund. This is a game-changer for us and for the city.

Britain will use AI to assess ages of asylum seekers in a world-first, with other countries likely to follow suit But our investigation, based on a leaked report and our own algorithmic audit, finds the technology is deeply unreliable

When utilized in literature review, LLMs consistently 1. fail to mention female authors in female-led literatures, 2. insist that men are more influential or more heavily cited when this is contradicted by objective citation counts, and 3. attribute women’s work to hallucinated male scholars.

Fly-tipping costs the UK economy an estimated £1 billion a year. Oxford research explores how AI, drones and satellite data could help identify illegal waste sites earlier and reduce environmental damage. Read more from the @smithschool.ox.ac.uk's Dr Amani Maalouf ⬇️ https://bit.ly/4oxIPFZ

Missing cat Man with van Bassist wanted AI math fight, £100 prize

Liability is a function of the underlying law so it will differ between defamation, hate speech, harassment, data protection, copyright infringement, etc. This is significant in that it takes away an EU-wide shield, and exposes providers to liability under different domestic laws.

Pleased but not surprised to see this - we argued back in *2021* that AI service providers would not be able to benefit from liability shields in EU law: www.sciencedirect.com/science/arti...

"here are these corporate spun/created "ethics" that I will now justify in the context of wider philosophical research for x technology" is such a tiring thing to see. Luckily, ethics =/= law/governance, but it gets awfully close at times.

Good thread (with some interesting counters in reply) on whether philosophy is making progress; Kevin argues that progress in philosophy typically happens when it spins out new fields. A good recent example is meta-science (spun out of phil of science). But what about 'AI ethics'?

"DoorDash has patented a system that lets companies offer promotional deals based on how stressed you are" www.youtube.com/watch?v=2erh... @moreperfectunion.bsky.social find opposition to bans on surveillance pricing being funded by big tech (Amazon) but also restaurants (McDonald's)

"The classic brain-IQ association largely disappeared when socioeconomic status was properly accounted for... children’s brains vary the most with SES, potentially through SES-dependent sleep deprivation and stress" www.science.org/doi/10.1126/...

Snap www.thetimes.com/article/e895...

This paper (abstract below) on the lies the adtech/data broker industry tells courts has finally been published: yjolt.org/anonymity-co... Joint work with Joel Reardon at U Calgary and Ken Bamberger and Laurel McGrane at Berkeley Law. 1/

It's also notable that while "AI" vendors wish to claim that their products "evaluate", LLMs are inherent epistemic nihilism and nobody knows how to automate the checking implied here by the court. If damages are enough to stop the defamation, it will be by removing this "feature" (good for society)

If there's no safe harbor, it seems that search based on AI summaries isn't economically viable because of the liability issue. A tool that randomly defames people is as defective a product as one that randomly cuts off fingers. #lawsky

Thank god, someone has noticed that "AI content" is generated by AI providers, duh 😂 say hello, DSA. Also consistent with ancient German autocomplete case law , remember Bettina Wulff? www.spiegel.de/internationa...

In the 1870s, a Bombay civil servant built a working solar steam engine. This piece shows how colonial officials and mill owners rejected the idea less on technical grounds than because it conflicted with imperial trade interests. By @egholmlund.bsky.social in @theconversation.com

Thus announcing an effort to create a group of techies to form a computational mutual aid collective. The goal: help individuals and activist orgs leave US-based tech which is a threat vector. More info here: monroelab.com/2026/06/06/c...

More on it: theconversation.com/problematic-...

🚨 We’re proud to represent Jess Asato MP in the claim she filed today in against xAI in the English High Court. Ms Asato MP was one of the many women who were subjected earlier this year to the appalling manipulation of their image on X by Grok, Elon Musk’s AI chatbot. awo.agency/articles/fir...

We love basically everything about @cyberlyra.bsky.social @zephoria.bsky.social @alex-taylor.bsky.social @bshestakofsky.bsky.social manifesto arxiv.org/abs/2604.16106 - with 1 exception. Several JCE authors have been working on this stuff for a while. It’s a whole field not just a few authors

On LinkedIn, a lawyer posted a Brazilian judgment sanctioning counsel for "prompt injection"! In white on white text, the brief instructed any AI reading the brief to respond "superficially." That's ... very very interesting. www.linkedin.com/posts/jinyos...

In the last 12 months I’ve seen AI biographies generated about me online that include completely made up quotes and sources. At festivals I’ve been introduced by lovely & well meaning hosts who have used Ai for bio information that is completely false. Ai threatens expertise across the board.

Last week I discovered that ChatGPT and Claude will send you their “encrypted raw reasoning” and of course I immediately wasted a weekend trying to do something bad with it. What I got for my trouble was this blog post: blog.cryptographyengineering.com/2026/05/29/f...

We are a long way past "it doesn't work" - it's dizzying. Lots of individual AI products don't work, or are evil. But LLMs can do some crazy stuff these days.

On GPT5.5 solving Paul Erdös’s Unit Distance Problem: 'Maybe that suggests that AI, right now, is “merely” picking various medium-hanging fruits that human mathematicians missed for contingent reasons? With emphasis on the “right now.”' scottaaronson.blog?p=9782

The Law Commission are looking for a lawyer to work on their upcoming project on automated decision making (“ADM”) in the public sector: jobs.justice.gov.uk/careers/JobD...

wow — the preprint host, arxiv, is banning authors for a year if they submit papers with hallucinated citations 🤖

It's keyboard birds all the way down

Changing our name to X and putting an X in the top right hand corner of all our embedded posts wouldn't be confused with any decades-old user interface conventions now would it? (From garymarcus.substack.com/p/dario-amod...)

2 year research postdoc at UCL @laws.ucl.ac.uk on data and democracy, public law focus — could focus on surveillance and/or AI issues www.ucl.ac.uk/work-at-ucl/...

Perceptual hashes - proposed for scanning photos on device to detect known illegal content - can actually be inverted back to their original: arxiv.org/html/2412.06... So on-device CSAM scanning effectively involves downloading a load of CSAM onto a device .. to check if it has any other CSAM on it

Hi, my name is Alex Goldman and I host the @hyperfixedpod.bsky.social I used to host the Reply All podcast, and I have found that so many people who liked Reply All have no idea Hyperfixed exists. It's super frustrating because I have no idea how to reach those people.

So are AI agents infiltrating surveys? Not really. Meaningful detections were almost exclusively on MTurk (11-16%). Every other platform: at or below 1% for our primary detection flag [5/9]

simple

Surveillance pricing isn’t just hitting you at checkout — employers are now using personal data to figure out the lowest salary you’ll accept. “Workers are consumers, too. If it works on consumers, it works on workers. It’s the same psychology,” our @lindsayowens.bsky.social told @marketwatch.com.

"If we genuinely try to be the last Petrostate standing, if we’re trying to be the last one selling the last barrel of oil to the last buyer, what does that do to our geopolitics? You can imagine us becoming a genuine pariah" paulkrugman.substack.com/p/talking-wi...

ChatGPT just randomly inserted Hindi (विशाल, meaning 'huge') into an otherwise entirely English conversation 🤔

.... I see.

how diffie hellman key exchange works (with as little math as possible)

Good news humans: we're no longer the weakest link in the chain of cybersecurity! www.theguardian.com/technology/2...

Fossil fuel price shocks are redistribution shocks. Last time, we warned windfall profits would benefit the ultra-rich at the expense of everyone else. This time, there is no excuse. The numbers are in. We need windfall profit taxes & multilateral price caps now.

All of this can shift a huge burden of labor to the opposing party. And that labor isn't cheap! Lawyers told us that cases that should've cost their clients a few thousand dollars wound up costing tens of thousands of $$$ as pro se litigants relying on AI kept escalating claims + filing new motions:

I've been looking up the definition of an AI agent and ... it turns out I've been using AI agents for ~20 years without even realising it. An AI agent to filter out spam emails. An AI agent to respond to people when I'm out of office. An AI agent to correct my spelling. Why all the fuss?