Remote
Tom Stafford
@tomstafford@mastodon.online
mastodon 4.8.0-nightly.2026-10-06"A vast confusion of vows, wishes, actions, edicts, petitions, lawsuits, pleas, laws, proclamations, complaints, grievances, are daily brought to our ears. "
2015 Followers
829 Following
50 Posts
Joined October 31, 2022
Github:
Newsletter:
Ignore this guy, join a Union
https://www.insidehighered.com/opinion/views/2026/08/27/why-postdoc-unionization-bad-idea-opinion
This is why @pluralistic@mamot.fr is such an essential writer: he can tell the history you lived through and make you see it again. In this case the story of superior technology - the VCR - which was disinvented as media corps got their hooks into regulators and law. Why *are* there DVD players and not DVD recorders? Why *do* DVD players have anti-features like region encoding that nobody actively wants?
Another world is possible!
https://pluralistic.net/2026/08/25/gammamax/#felony-contempt-of-business-model
"The word “risk” originally comes from “rhizikon”, Byzantine Greek meaning “a sharp rock or reef”, later “any hazard to shipping”...The word “average” comes from an Arabic word “awariya” meaning “lost merchandise”; the “Law of General Average” is the principle that when stuff has to be chucked overboard in a storm the losses will be shared out fairly between the merchants so that the crew are can grab the heaviest objects as they see fit without worrying about who owns what."
/1
No need for an inner critic when you have the replies on social media, am I right?
"In short, the technologies through which we see the public shape what we think the public is. And that, in turn, shapes how we behave politically and how we orient ourselves."
https://www.programmablemutter.com/p/were-getting-the-social-media-crisis
("The bigger problem isn't disinformation. It's degraded democratic publics")
Depressing. I knew internet hero @nixCraft@mastodon.social would have a page with exactly the information I wanted, but the search engine puts the option to generate an AI summary above the result I want.
It's not "Search assist", is it, in this context? It's stealing traffic and credit
"Regrettably, some academics are drawn to big theories that seem too grand for clarity or mere evidence."
Billig, 2011 on academic writing
I've started getting Google Scholar alerts which announce work by legit academics but the titles click through to porn sites.
Does anyone know why that would be?
Some kind of SEO scam? Some kind of phishing? Something more targetted against the affected academics?
#AcademicChatter #ScholarlyPublishing
Testing alternatives to the engagement-maximising, polarisation-inducing, algorithms of most social media feeds
https://open.substack.com/pub/tomstafford/p/another-algorithm-is-possible
Part three of my series of posts about signalling theory, and a fantastic case study of market change produced by the introduction of AI to a freelancing platform, giving everyone easy access to instant, bespoke, job applications. It didn’t end well, and the insight into why comes from understanding the nature of honest signals.
https://tomstafford.substack.com/p/when-trusted-signals-collapse
Replying to
I think signalling theory is useful for understanding the effects of LLMs - they don't just *produce work*, they disrupt the indicators that we are used to relying on for judging how much effort someone put into communicating with us.
When someone writes to me now, I cannot as easily judge the level of care they took, their expertise in a domain, or bespoke knowledge about me that they have used to customise their communication
Problematic!
2/
In other news, I am absolutely delighted with my new profile picture
Replying to
@ana_valdi a) telling the truth is defence against defamation but b) if I was in your position I would want to speedrun a law for journalists course, or recruit a journo to advise!
Nice note on the publishing history of Akerlof's classic "The Market for Lemons" paper (1970).
Rejected by two top journals for being "trivial", and a third for being "incorrect".
Today one of the "most-cited papers in modern economic theory" (and possibly anywhere? Google scholar reports 48,913 citations as of today)
Replying to
@streetartutopia blocked and reported. Here is my report "High profile account. More responsibility not to degrade the discourse by sharing fake (AI generated) images. Engagement farming on Israel-Palestine conflict in this way risks undermining trust in real claims around destruction in gaza"
New newsletter
https://tomstafford.substack.com/p/the-mythical-median-voter
It's about the how we imagine the voting public. And how this is distorted by polling, and faulty intuitions about statistics. And there's n-dimensionap geometry and a statistic gag in there too
Reasonable People #100! Stated preference, revealed preference and the complications brought by circumstance
https://tomstafford.substack.com/p/the-surprising-deceptions-of-individual
A flying visit to @cwts@social.cwts.nl
An inspiring research unit with a great culture and ethic, and great plans
Good posters by the lifts too
Compare/contrast: ward-by-ward Sheffield City Council election results
1/ when mapped to ward area (SCC, via @index)
vs
2/ an equal-area representation (from me)
A big regret of my year off is that I didn't schedule "keeping in touch" calls with my boss, elaborately staged on a Jason Bourne theme
Me video calling from the embassy basement, hands bound with cable ties
Me video calling from a high speed train, sniper fire cracking glass inches from my head
Me video calling while trading brutal Muay Thai blows with numberless henchmen
Final video call. On a roof. My telephoto lens snaps into focus to show my boss's office and the back of his head
LLMpedia: A Transparent Framework to Materialize an LLM’s Encyclopedic Knowledge at Scale
Previously the factuality of LLMs has been assessed by querying them. This project generates an entire encyclopedia from model parameters, allowing an alternative — and more pessimistic — assessment of their accuracy
Paper: https://arxiv.org/abs/2603.24080
Site: https://llmpedia.net/
Replying to
An obvious application is #ResearchFunding
Research Funders increase rising application numbers. Some - maybe not all, but definitely some - of that is due to AI. If so, then we should expect not just increased competition / reduced success rates, and increased funder workload, but ALSO loss of the reliability in the signals reviewers have used to assess quality
Direct evidence for this: https://osf.io/preprints/metaarxiv/d8gcu_v1
3/
Replying to
Mixed feelings about the fact that my clickbait headline for my non-clickbait newsletter piece is proving popular
https://mastodon.online/@tomstafford/116237654310008878
Replying to
@neuralreckoning really interesting. The suggestion highlights the fact of how little actual development of proposals happens in normal grant review - funders recruit external reviewers, but they typically just criticise, sifting better from worse proposals,. without contributing to making individual proposals better
Evolution, signalling theory and (only safe for work) badger pictures. What’s not to like? https://tomstafford.substack.com/p/lets-talk-about-the-badger-penis
Lots of people are misinterpreting my latest newsletter
https://tomstafford.substack.com/p/hostile-interpolation
(not really, but the newsletter discusses the particular rhetorical style which positions readers as stupid or misled)
Replying to
Credit to OP meme creator @Shine_McShine@neopaquita.es https://neopaquita.es/@Shine_McShine/115140078225044394
South Africa withdraws AI policy due to fake AI-generated sources
Replying to
@moritz_negwer yes, great point
From a science funding and costly signalling perspective, one question is how funders could impose application costs which are disproportionately costly for 'dishonest' signallers (and what that would even mean?)
Mythos beer on offer at Tesco's.
I can only presume it is an Anthropic tie-in
Replying to
@steveroyle@biologists.social well part of the answer to that second question is illustrated by my new pic - different information in different spatial frequencies
"an AI coding agent deleted his firm’s entire production database. The AI agent’s misdemeanors were then hugely amplified by a cloud infrastructure provider’s API wiping all backups after the main database was zapped."
Replying to
@emjonaitis glad thoughts were provoked. Yes, no easy answers!
A real worry is the flight to "trusted" ways of distributing funding will have huge encumbancy bias, so I'm really hopeful we can (collectively) find viable alternatives
Replying to
So what's to be done? 3 response:
#1 Double down on the old signals
Maybe that means more things like interviews, to check researchers really understand the proposals they have written (risk of bias!). Some funders already have interviews for their major funding schemes (Wellcome, ERC), others are introducing them (MRC in the UK)
Maybe it means letting reviewers adjust, getting used to seeing "excellent" proposals and shift their baseline for giving excellent scores
Maybe
5/
Replying to
Option #3: New funding mechanisms
Give the money out in different ways. Is the proposal dead?
One suggestion from Galdin & Silbert (https://tomstafford.substack.com/p/when-trusted-signals-collapse) is that employers (i.e. funders in my context) need to trial more short term contracts, so they can collect reliable signals from workers before committing to longer jobs
More radical is push funding - rewards for research achievements rather than plans
https://goodscience.substack.com/p/the-case-for-pull-funding-at-the
There must be others. Let me know!
/7
Replying to
Related: Our short report on our @RoRInstitute workshop on funder Demand and Demand Management is now live
https://rori.figshare.com/articles/report/_b_Demand_management_findings_from_a_multi-funder_workshop_b_/32044422?file=63826728
Also related: @ERC_Research have just announced additional demand management mechanisms https://erc.europa.eu/news-events/news/erc-president-explains-stricter-application-measures-amid-rising-demand-funding
As did the @snsf_ch in February https://www.snf.ch/en/1OsSI1dScWZlkNAf/news/greater-demand-smaller-budget
4/
Afternoon boost for this morning's newsletter - a really neat set of studies which show how collective intelligence can be enhanced if groups structure their discussions, breaking down big questions into smaller parts. Even when the component answers are still uncertain, the overall accuracy of the estimate improves!
https://tomstafford.substack.com/p/the-wisdom-of-deliberative-crowds
Where do British political scientists preprint?
If the answer is the Elsevier-owned SSRN can you suggest alternatives?
EDIT: My sample (n=1) report they use OSF, which is good enough for me!
Replying to
@stuartyeates@cloudisland.nz interesting. Does that mean that one of the affected academics needs to cleanse their scholar profile??
Replying to
Managed to break the thread somehow, options #2 and #3 continue here https://mastodon.online/@tomstafford/116435684407415961
Tuesday boost for my post about evolution, signals and ....the badger
https://tomstafford.substack.com/p/lets-talk-about-the-badger-penis