@tomstoneham@dair-community.social

Tom Stoneham @tomstoneham@dair-community.social · Jul 30, 2026

The whole OpenAI/HuggingFace incident made me feel the need to bang on again about how a very distinctive set of values have been smuggled into AI systems via an allegedly 'objective' definition of intelligence. tl;dr - if you build something with zero tolerance of failure, it will always cheat when that is the most efficient and reliable means to complete the task #AI #OpenAI #Cheating #ValueAlignment https://listed.to/@24601/74966/but-why-do-ai-models-cheat

10

0

11

0

Tom Stoneham @tomstoneham@dair-community.social · Jun 24, 2026

Replying to @denzilferreira@techhub.social

@denzilferreira@techhub.social This will fall on deaf ears. The neoliberal hegemony loves tech monopolies because it allows them to outsource the policing of digital space. (The fact that Big Tech is even less democratically accountable than the real police is easily ignored because the policing is now off balance sheet. Same as privately owned 'public space' in cities which are policed by private security.)

0

Tom Stoneham @tomstoneham@dair-community.social · Jun 19, 2026

Replying to @barrydeutsch.bsky.social@bsky.brid.gy

@barrydeutsch.bsky.social Brilliant subversive humour! Where is this?

1

0

Tom Stoneham @tomstoneham@dair-community.social · Jun 04, 2026

Replying to @mikarv@someone.elses.computer

@mikarv@someone.elses.computer @Dragon@toast.dragon2611.net And if you ask it to do some arithmetic, that seems to go up at least 10x (Hugging Face had a nice tool to track task specific energy use of their local models but it seems to have been withdrawn)

0

Tom Stoneham @tomstoneham@dair-community.social · May 07, 2026

📈 Quantity Without Quality: Will the use of genAI just lead to research 'slop'? 💩

📆 Thu 4th June
🕒 3-4pm
👇 Register here:
https://us02web.zoom.us/meeting/register/9WXUcvrIT3macDD4LubI9w#/registration

In this session in our series on #AI in universities, we will discuss every way in which AI finds its way into our #research right now - our own research being used to train AI, using AI in our own research, and reading research papers that used AI.

☀️ Students, academic and professional service staff all welcome!

#Academia #AISlop #PullThePlug

1

0

2

0

Tom Stoneham @tomstoneham@dair-community.social · May 05, 2026

Replying to @mattburgess@infosec.exchange

@mattburgess@infosec.exchange I love this project. Thank-you! (I am working on more theoretical aspects of digital sovereignty and a resource of examples is great.) But as far as I can see, you are not tracking use of Matrix as an alternative to Slack/Teams for messaging and collaboration. The best implementation so far is LuxChat, though it is also built into France's La Suite I have heard rumours others including the MoD. https://www.luxchat.lu/en/

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · May 04, 2026

Replying to @tomstoneham@dair-community.social

@westsibe@mastodon.social Thanks

0

Tom Stoneham @tomstoneham@dair-community.social · May 03, 2026

Replying to @david_chisnall@infosec.exchange

@david_chisnall@infosec.exchange Thanks. I hope so and have had that experience previously, though it is intermittent. Just need a plan B!

1

0

Tom Stoneham @tomstoneham@dair-community.social · May 03, 2026

Hi y'all - I am travelling in China this month and it would be handy to keep in touch with some peeps via #Signal.

If you are running a #SignalProxy, please DM.

(Please don't recommend VPNs unless you have recent, detailed 1st hand experience.)

1

4

1

0

Tom Stoneham @tomstoneham@dair-community.social · May 01, 2026

IDK, but when I saw the video my first reaction was shock at a police officer kicking someone lying on the ground during an arrest. That cannot be police procedure, surely?

Then Mark Rowley said something about fearing he had explosives. IDK, but my hunch is that Tasering someone carrying explosives might be pretty dangerous, and I can't see why subsequently kicking them on the ground helps

It is possible to think two things are true: they were brave and the Met uses unacceptable violence.

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 30, 2026

Replying to @2legged@mastodon.ie

@2legged@mastodon.ie Setting aside the content of the story for a second, Bedale is not remote. It is a mile and a half from a junction on the main road from London to Edinburgh via Newcastle. And yet The Guardian used a stock photo from a *London* restaurant taken 4 years ago rather than give the restaurant they are writing about a bit of publicity. Shocking London-centric bias.

0

2

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 28, 2026

Replying to @signalapp@mastodon.world

@signalapp@mastodon.world I am trying to work out why Registration Lock is not on by default. In particular, why would you be able to set a PIN and not set Registration Lock? I can imagine a use case where someone didn't want a PIN at all, though that shouldn't be the default in a secure messaging app. But why PIN and no lock?

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 27, 2026

Replying to @UlrikeHahn@fediscience.org

@UlrikeHahn@fediscience.org @mxp@mastodon.acm.org Ah. I was imagining a situation where you stretch then rotate, then reverse the stretch without first reversing the rotation. That would be what I had in mind for 2).

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 27, 2026

Replying to @UlrikeHahn@fediscience.org

@UlrikeHahn@fediscience.org @mxp@mastodon.acm.org I don't know much about image editing, so I assume you are right. I wonder if a methodology designed for text editing has been applied to images without checking it is still applicable?

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 27, 2026

Replying to @UlrikeHahn@fediscience.org

@UlrikeHahn@fediscience.org @mxp@mastodon.acm.org You are right - I was spiralling off to bigger issues! As I read it - and methodologically this is way out of my area - there are two assumptions here: 1) Effective delegation requires delegated actions to be reversible 2) A minimum requirement for professional editing software is that it tracks changes and these can be individually reversed in any order I am far from sure that 1) is true, but 2) does seem reasonable. But I am out of my AoC here.

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 27, 2026

Replying to @UlrikeHahn@fediscience.org

@UlrikeHahn@fediscience.org @mxp@mastodon.acm.org Good point. I see this as one of the tensions in the technology. Current 'general purpose' models need very careful prompting to do anything worthwhile, but the vision is to have models which will produce human-like results from the sorts of vague, context-sensitive, and mutual knowldge presupposing instructions we give human assistants. So no one can decide which prompting styles are relevant for evaluation of capabilities.

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 25, 2026

Replying to @mu@mastodon.nz

@mu @bruces Yeah, I get that. But it still rankles - just like only counting US bodies in a war they started. It is a country that has enabled evil corporations with global power to rape the rest of the world and then they start worrying when it hurts them. I'm British. We did that. And are deeply ashamed. Where is the shame in this article?

2

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 25, 2026

Replying to @bruces@mastodon.social

@bruces Count how many times this article talks about the effects on 'regular Americans' as if the other 7.5 billion people in the world also being screwed over by these companies are simply irrelevant. That #NeoColonialism is the deeper problem.

2

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 22, 2026

Replying to @jwildeboer@social.wildeboer.net

@jwildeboer@social.wildeboer.net I hope this is true, not because I like Apple or AI, but because where Apple leads, others follow (and decentralised, specific-purpose, smaller models is the only hope for planet and people).

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 10, 2026

Replying to @element@mastodon.matrix.org

@element@mastodon.matrix.org Also a good alternative to Discord for social and community groups. #SelfHost on a #RaspberryPi with accounts limited to people you know or are vouched for and skip all the age-verification nonsense.

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 10, 2026

Replying to @greenpeace@mastodon.social

@greenpeace Fixed your #Alt text for you: A large wind turbine rising above trees in the evening light with a clear blue sky. Caption reads: This machine stops wars. In smaller print: (It's a big fan of peace) #Ableism

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 09, 2026

Replying to @neil@mastodon.neilzone.co.uk

@neil I would say that there is a certain amount of #AppleEnvy at #Microsoft and #Google right now. A #WalledGarden allows higher #Rent and cheapens compliance with legislation that attempts to restrict what users can do. Open source - and thus backdoor-free - VPNs and e2ee are probably being targeted as preemptive obedience to future laws.

15

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 04, 2026

Replying to @ChrisMayLA6@zirk.us

@ChrisMayLA6@zirk.us

An NHS spokesperson said: “All suppliers, including Palantir, and their staff operate ... under NHS control and governed by strict contractual confidentiality obligations.”

= #ReguloSolutionism in action.

Contracts are the first line of defence but with a company as large and powerful as #Palantir they are no guarantee. As Eric Schmidt said:

if [you made money], then you'd hire a whole bunch of lawyers to go clean the mess up, right?

Contracts/law is just a cost of business

1

0

Tom Stoneham @tomstoneham@dair-community.social · Apr 04, 2026

Replying to @ChrisMayLA6@zirk.us

@ChrisMayLA6@zirk.us Do you have a link to the FT article?

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · Mar 21, 2026

Replying to @reb_bamf@mastodon.social

@reb_bamf@mastodon.social The ultimate victory of capital over labour.

0

Tom Stoneham @tomstoneham@dair-community.social · Feb 28, 2026

Replying to @axbom@axbom.me

@axbom@axbom.me I suspect that this is in part due to a deliberate *commercial* choice by #OpenAI False positives will annoy users who waste their time (and often money) going to hospital unnecessarily and disincline them to use the platform again. False positives will also annoy hospitals, especially in countries like the UK where they are very resource constrained. (1/2)

0

Tom Stoneham @tomstoneham@dair-community.social · Feb 04, 2026

Replying to @abbas_dp@techhub.social

@delta@chaos.social @abbas_dp@techhub.social Amazing work! It would be great to see your server installation guide, if it simplifies the 'official' process. I don't read Persian but maybe @Mehrad@fosstodon.org could help translate? With the Signal servers under pressure (technical or political?) yesterday, we need more decentralised options here too.

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · Feb 02, 2026

Replying to @JulianOliver@mastodon.social

@JulianOliver@mastodon.social @rzeta0@mathstodon.xyz @analogfusion@mastodon.art #Proton will be there when Proton Meet comes out of beta. It is full suite (well, only docs and sheets ATM but mail, calendar, drive, password manager, authenticator, VPN) and all #e2ee. And easy import from Google. Only catch is that it is freemium and most people think technology should be free. @protonprivacy@mastodon.social

2

0

Tom Stoneham @tomstoneham@dair-community.social · Feb 02, 2026

Replying to @tomstoneham@dair-community.social

@DJGummikuh@mastodon.social LLMs are provably worse than that because they will completely ignore or actively change information which is unlikely. Classic example was a meeting notes app changing an extended leave request from a female member of staff (for special holiday) to a maternity leave request. We also see LLMs which aren't fine tuned for medical contexts not recording unusual patient symptoms or changing symptoms to align better with probable diagnosis.

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · Feb 01, 2026

Replying to @whitneymcn@mastodon.xyz

@whitneymcn@mastodon.xyz One thing no compression algorithm could do is spot when the original author has misjudged what is important. And LLMs are even worse because they flatten the unusual.

1

0

Tom Stoneham @tomstoneham@dair-community.social · Feb 01, 2026

I ought to write this up properly, but here's a tl;dr in advance: #AI does not *summarise* documents - it *shortens* them by using structural similarity to documents in its training data to guess/predict which sentences are important and which are not. If that is what you want, fine. But don't be fooled into thinking it is doing anything like human summarisers do. They work on the *meaning* of the document in front of them.

27

6

18

1

Tom Stoneham @tomstoneham@dair-community.social · Jan 23, 2026

Replying to @jwildeboer@social.wildeboer.net

@jwildeboer@social.wildeboer.net There is another political aspect of this. If US #TikTok algorithm is trained only on US data, then it creates an information bubble. Cuts the majority of US users off from the rest of the world. This is a soft/manipulative version of an 'internet block', a restriction on cross-border flows of information. (Of course Americans can still access that information other ways, but the overwhelming majority won't bother. And if #Meta preemptively obeys ...)

0

Tom Stoneham @tomstoneham@dair-community.social · Jan 14, 2026

Replying to @mattburgess@infosec.exchange

@mattburgess@infosec.exchange It seems like 'police intelligence' and 'artifical intelligence' have one thing in common, and it isn't intelligence. This is nice reporting based on Hansard: https://www.bbc.co.uk/news/live/c394zlr8e12t?post=asset:5da5244d-3aee-48b8-b2dd-92950c100780#post

1

0

1

0

Tom Stoneham @tomstoneham@dair-community.social · Aug 15, 2025

Replying to @mojoaxel@social.tchncs.de

@mojoaxel@social.tchncs.de @aboutsignal@mastodon.social I can confirm Llama 3.2 running locally with real-time data gives: "Signal: Signal is my go-to choice for private conversations. It's end-to-end encrypted, meaning only the sender and receiver can read the messages. It also has a clean interface and works on both iOS and Android." But also mentions Wire and Telegram Secret Chats

0

Tom Stoneham @tomstoneham@dair-community.social · May 31, 2025

Replying to @ArnimRanthoron@ruhr.social

@ArnimRanthoron@ruhr.social @jasongorman@mastodon.cloud We have good theories explaining how syntactic rules can implement semantic rules. In fact, nearly 200 years of formal logic. We also have good theories explaining why a next token generator does not implement linguistic understanding. But sadly these are not well understood outside linguistics and philosophy of language.

0

Tom Stoneham @tomstoneham@dair-community.social · Mar 05, 2025

Replying to @ConnectedByData@social.coop

@ConnectedByData@social.coop No thanks. https://www.joanwestenberg.com/big-tech-wants-you-trapped-the-open-web-sets-you-free-2/

0

Tom Stoneham

Posts

Remote instance

Media