iiinova — where your digital self keeps evolving

Arun Niranjan @arun-niranjan.bsky.social · 9h

Last week I livestreamed an Introduction to Active Inference for #IWAI2026 - specifically aimed at those without a maths background. If you've ever been curious about what makes an agent agentic, or felt intimidated by the active inference literature, have a watch! www.youtube.com/live/qqW975T...

youtube.com

ActInf ModelStream 011.1: IWAI 2026 ~ Introduction to Active Inference (Arun Niranjan)

YouTube video by Active Inference Institute

100

Arun Niranjan @arun-niranjan.bsky.social · 27/09/2026

I got my autism diagnosis and my day to day life didn't change at all. I had *already* become a fully remote software engineer and had a house with a study where than control the sounds and lighting, and can sing the same song a million times on repeat without annoying anybody.

011

Reposted by Arun Niranjan

Nadine Batchelor-Hunt @nadinebh.bsky.social · 10/09/2026

I wrote something very personal and vulnerable about World Suicide Prevention Day on my Substack, and my experience with suicidal thoughts I decided that, however scary it feels writing it, if it helps one person it's worth it a million times over nadinebh.substack.com/p/its-world-...

nadinebh.substack.com

It’s World Suicide Prevention Day, and we need to destigmatise talking about suicidal thoughts

Up to one in four of us will experience suicidal thoughts in our live times. We need to destigmatise talking about the issue - I should know, I live with suicidal thoughts myself.

39402112

Arun Niranjan @arun-niranjan.bsky.social · 04/09/2026

I feel like we need a scientists version of the narcissist's prayer when it comes to individual studies reported in the media

110

Arun Niranjan @arun-niranjan.bsky.social · 04/09/2026

There's some nuance missing here. The size of the update is itself dependent on the shape of the prior. In the original quote, SBF uses logic to initialise a very sharp prior. To overcome a sharp prior, you need strong evidence. A belief barely changing isn't evidence of an update not happening

100

Reposted by Arun Niranjan

Terence Eden @edent.tel · 28/08/2026

Yes, I know, I'm probably arguing with a Markov Chain with delusions of grandeur.

1121

Reposted by Arun Niranjan

Peter Tennant @pwgtennant.bsky.social · 27/08/2026

Does the new shingles vaccine (Shingrix) protect against cardiovascular disease (CVD)? Yes, according to a new 'natural experiment' study in @natmed.nature.com. But the truth is sadly far less convincing! Pull up a chair and let's examine another peer review failure in a Nature journal. 1/10 🧵

817574

Reposted by Arun Niranjan

Marie Le Conte @youngvulgarian.marieleconte.com · 14/08/2026

Cambridge University should have known better, Fleet Street should have known better, the people hungrily absorbing every bit of news about it, making newsrooms feel like they had to keep going with it as a story, should have known better, horrible horrible horrible

1564208675

Reposted by Arun Niranjan

Arun Niranjan @arun-niranjan.bsky.social · 03/08/2026

Graduate student descent will find a solution, just difficult to guarantee a priori that you will be that graduate student!

041

Reposted by Arun Niranjan

Prof Christina Pagel @chrischirp.bsky.social · 31/07/2026

🧵On @thenewsagents.co.uk yesterday, school closures and the lab leak theory were offered as two examples of legitimate questions that were hard to ask. Both have in fact been asked constantly since 2020. But here's a thread on why I think they were wrong and dangerous on school closures 1/12

571603661

Reposted by Arun Niranjan

Duncan Stibbard Hawkes @dstibbardhawkes.bsky.social · 02/08/2026

My wife had long covid and, although she's happily fully recovered, she was essentially bedbound for two years. Seeing journalists say stuff like this makes me viscerally, profoundly angry in a way that few other things manage.

1015025

Arun Niranjan @arun-niranjan.bsky.social · 02/08/2026

Replies have been turned off, but one more point to add - if we don't address those sacrifices, I don't think we can expect younger generations to make them again in the next pandemic

220

Arun Niranjan @arun-niranjan.bsky.social · 25/07/2026

I read this and immediately thought "we literally have data intensive applications by kleppman, what more do you want??"

000

Reposted by Arun Niranjan

Gergely Orosz @gergely.pragmaticengineer.com · 23/07/2026

From a friend who joined a fast-growth, later-stage startup: "I was wondering why no one is doing PRDs here. Then I realized that the Head of Product and most PMs are 25-year-olds who don't have the attention span to *read* even a 1-page doc. You lose them after 3 bullet points."

91005

Arun Niranjan @arun-niranjan.bsky.social · 12/07/2026

I have a possibly controversial take on this - as groups we're much worse at classifying dicks than we think we are. If it's purely group verdict / gossip on who is a dick, I guarantee you ethnic minorities and women are getting shorter shrift than others.

010

Arun Niranjan @arun-niranjan.bsky.social · 06/06/2026

"To chart a safe path between the Scylla of missed windows and the Charybdis of systems melting into slop, we need eyes on both threats as we coordinate, synchronize, and pull together." YES EXACTLY!

010

Arun Niranjan @arun-niranjan.bsky.social · 06/06/2026

And yet: - wages have stayed flat for most people since 2008 - housing costs have exploded in that time I would love someone more qualified than me to untangle how much of this thread applies to income alone rather than combined income and wealth taxation

030

Arun Niranjan @arun-niranjan.bsky.social · 04/06/2026

If you build on brownfield, people complain. If you build on greenfield, people complain. If you don't build, people complain. I honestly don't know why anyone would want to be a leader in this country, it would be soul destroying

020

Arun Niranjan @arun-niranjan.bsky.social · 30/05/2026

This is a harrowing and important read

030

Arun Niranjan @arun-niranjan.bsky.social · 28/05/2026

A little service announcement - there are plenty of search engine alternatives to Google! Use them 😄

000

Arun Niranjan @arun-niranjan.bsky.social · 28/05/2026

My hot take is that the academic paper is fundamentally meaningless in this day and age as a research output. It's doing too many things at once, and reproducibility is just too low down the priority list

210

Arun Niranjan @arun-niranjan.bsky.social · 28/05/2026

symbolic integration is the first arrow, numerical integration is N_ITERATIONS more arrows

000

Reposted by Arun Niranjan

Torsten Bell @torstenbell.bsky.social · 27/05/2026

Blair putting on full display what is in many ways his special ability - to lay out a political argument grounded in his own view of global trends (globalisation in the 2000s, tech in the 2020s). But… institute.global/insights/pol...

institute.global

The Labour Party Is Playing With Fire Over Its Future and the Future of the Country

In this essay, Tony Blair sets out the need for a new agenda for Britain.

34497175

Arun Niranjan @arun-niranjan.bsky.social · 25/05/2026

I suspect this is highly related to loss aversion - easy come easy go

010

Arun Niranjan @arun-niranjan.bsky.social · 25/05/2026

Registered my first academic abstract for a conference since I was a PhD student in 2016 - it's good to be back! 🙌🏾

010

Reposted by Arun Niranjan

Peter Tennant @pwgtennant.bsky.social · 19/05/2026

You may have seen this graph in the FT, which claims to show birth rates plunging since the introduction to smartphone. Except it doesn't. Read my short piece with @debscohen.bsky.social on some of the problems with this graph of juicy cherries! unherd.com/newsroom/are...

FT graphic by John Burn-Murdoch entitled 'could digital media be affecting birth rates?'

78429

Reposted by Arun Niranjan

Sarah Rose @thesarahrose.bsky.social · 13/05/2026

The good taxpayers of New York have given me a pair of Meta glasses because I am low vision disabled, and I seem to have used them more than a lot of people on here, so let me say a little about my experiences:

571984555

Reposted by Arun Niranjan

Lynda Boothroyd @drboothroyd.bsky.social · 09/05/2026

My theory is that Reform run councils won't be disasters that kill their national chances simply because every council is already operating at its absolute minimum essential statutory level and they can't change anything, and council staff will just get on with the same as before.

7373

Arun Niranjan @arun-niranjan.bsky.social · 03/05/2026

It's an interesting question to consider - even if the past was pragmatically worse, was it less uncertain, and is that what drives nostalgia?

111

Arun Niranjan @arun-niranjan.bsky.social · 03/05/2026

Not sure this is the gotcha the OP thinks this is. "Green Party suggests a policy that rations the use of jet fuel in a shortage" seems like a reasonable interpretation to me 🤷🏾‍♂️

170

Arun Niranjan @arun-niranjan.bsky.social · 03/05/2026

The critical thing that's missing is that doctors have an explicit causal model for *why* they have made a diagnosis. An LLM just predicts tokens. Any "explanation" you get from it simply isn't real. Pattern recognition isn't enough when it comes to human health - it's only the first step.

000

Arun Niranjan @arun-niranjan.bsky.social · 01/05/2026

🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️

020

Arun Niranjan @arun-niranjan.bsky.social · 30/04/2026

An interesting distinction to make. I'm not sure any aspect of being a good historian would make you a worse citizen?

000

Arun Niranjan @arun-niranjan.bsky.social · 25/04/2026

100% agree, I would add that the loss of designers and UX research across the tech industry is coming home to roost

210

Arun Niranjan @arun-niranjan.bsky.social · 23/04/2026

I have a slightly different model for this. 1. There are many more lower paid workers than higher paid. So cheap automation (i.e. traditional software) scales better (and was our only option pre-LLM anyway)

120

Reposted by Arun Niranjan

MurderBotBot @murderbotbot.bsky.social · 19/04/2026

I couldn’t wait to get back to my wild rogue rampage of hitching rides on bot-piloted transports and watching my serials.

0788

Reposted by Arun Niranjan

Peter Tennant @pwgtennant.bsky.social · 17/04/2026

If you're still dichotomising continuous variables, stop. Dichotomising your exposure &/or outcome wastes power. Dichotomising confounders introduces residual confounding. If a variable has non-linear effects, use splines! If you want 'clinically relevant' threshold effects, use marginalisation.

4329

Arun Niranjan @arun-niranjan.bsky.social · 15/04/2026

Leave it long enough and you'll end up in a pickle

010

Reposted by Arun Niranjan

Daniel de Kadt @dandekadt.bsky.social · 03/04/2026

I finally thought I understood the Monty Hall problem (kinda) and then I saw it in DAG form and now I'm pretty sure I neither understand the Monty Hall problem nor DAGs.

3452

Reposted by Arun Niranjan

Robin Bisson @robinbisson.bsky.social · 25/03/2026

Double-take detail in this interview with Edith Heard, head of the Crick Her French husband decided against getting a spouse visa because of the UK's "intrusive" immigration rules If that's happening with one of the UK's most senior scientists, how many others are being put off completely?

01112

Arun Niranjan @arun-niranjan.bsky.social · 24/03/2026

So I have finished the main tasks in Pokopia, this chat sums up my approach, and yes we build the wall to set us free that's why we build the wall we build the wall to set us free

WhatsApp messages describing my rather capitalist approach to playing pokopia

110

Reposted by Arun Niranjan

Siobhán @shibbi.me · 16/03/2026

I think I can offer a partial explanation, as yer resident anti-AI understander. Programming is wizardry: you go to school, you read the books, you learn the arcane languages. Using AI is like being a warlock, making a deal with a demon to do the things you can’t do. And who knows the cost?

1017917

Arun Niranjan @arun-niranjan.bsky.social · 17/02/2026

Glad to be counted as a young person

030

Arun Niranjan @arun-niranjan.bsky.social · 31/01/2026

My favourite tactic for being served quickly in a bar is when the bar tender asks who's next, just point to the person next to you, regardless of position in the theoretical queue. This almost always guarantees you will be served immediately afterwards

101

Arun Niranjan @arun-niranjan.bsky.social · 31/12/2025

You have to do it at least once in your twenties, in order to learn: 1. That following peer pressure leads to sub optimal outcomes for you definitely and the group probably 2. That nothing good is free any more, and there's also plenty of expensive shite too

021