Sign in

peterkwells.com

@peterkwells.com
0 followers 0 following 2 posts
PostsRepliesMedia
peterkwells @peterkwells.com · 7h
For anyone concerned, obviously the bins went out anyway. Do you think we'd trust a bot for such a significant task as bin day?
120
peterkwells @peterkwells.com · 7h
Bad bot
An exchange with Instinct AI about why it didn't send me a bin reminder
120
peterkwells @peterkwells.com · 8h
_drops in an email_
010
peterkwells @peterkwells.com · 8h
So, I do hope this bit of the GDS team are planning to properly engage across the whole society/economy to understand stuff for future iterations of the PSGA...
GDS is undertaking engagement across the public sector to understand emerging challenges and opportunities that will shape the future iteration of the PSGA. If you haven't already done so, take this opportunity to provide your input and complete GDS's survey.
110
peterkwells @peterkwells.com · 8h
You could read this blogpost from the GDS Geospatial Data team about building on the £1bn Public Sector Geospatial Agreement without realising that the PSGA includes a chunk of money for open data that anyone, including those outside public sector!, can use gdsgeospatial.blog.gov.uk/2026/09/29/b...
gdsgeospatial.blog.gov.uk
Building on the PSGA: better data, wider impact
The Public Sector Geospatial Agreement has transformed access to Britain’s national location data. We look at what GDS, Ordnance Survey and public-sector users have achieved together, how organisation...
182
peterkwells @peterkwells.com · 16h
all thanks to you!
000
peterkwells @peterkwells.com · 18h
Me reading some more reports on user-level AI risks that look at the Wrong Thing
Bart Simpson writing lines on a chalkboard that say "most of the public use products, not LLMs"
000
Reposted by peterkwells
Mike Masnick @masnick.com · 21h
Soooooooo. As expected, the guy in that story about Muse doing FB Marketplace deals *had* in fact clicked the permission to "always allow" Muse to do things for him. Still very, very poor design by Meta such that the user didn't understand what permission he was giving...
matt.j.robb
14m
Update on letting Muse run Facebook Marketplace for me.
So I’ve spoken to 
@davidsingleton
 and the Muse team tonight and we went through the log to figure out what went wrong. Regarding the address leak. The first thing that popped up from Muse when asking it to handle my Facebook marketplace was an option with ‘Allow One Time’ or ‘Allow Always’. I clicked the latter thinking it would still send approvals to accept offers later down the line (it didn’t so be careful). 
1
/
4
matt.j.robb
14m
By doing that it granted Muse permission to send messages on my behalf going forwards using a template it put together using information it asked from me. Which also included the pickup address that I did give to Muse (again I didn’t think it would send it out to everyone that gave me an offer so it’s worth checking). I mentioned to the meta team it could be easily overlooked as it was with me and they said they will be making this more clear moving forwards. 
2
/
4
matt.j.robb
14m
Regarding the lowball offer situation. It accepted an offer of $600 and sent out my address to a buyer to come collect the item after I strictly set a minimum of $700 with Muse. The Meta team informed me this was an error on their end which caused a display issue removing the ‘7’ from the output to the buyer. And with it stating “Sounds good, 00 it is!” after the buyer asked for $600 it made it look to the buyer like his offer was accepted. 
3
/
4
matt.j.robb
14m
I suggested to Meta that each message sent by Muse should have a little “Sent By Muse” badge underneath so everyone knows it wasn’t sent by a human as there was really no way of knowing who sent what in the chat which is very concerning.
As for the buyer, we made up and he even came by tonight to buy something else. All ended well.
2427166
peterkwells @peterkwells.com · 28/09/2026
yup!
110
Reposted by peterkwells
Martin Greenaway @mgreenaw.bsky.social · 28/09/2026
There are a bunch of people in the political apparatus who are confused as to why Russia doesn't take us seriously.
06811
peterkwells @peterkwells.com · 28/09/2026
People in multiple places should be asking themselves how on earth senior UK politicians/cabinet members and big firms promoted this garbage
121
peterkwells @peterkwells.com · 28/09/2026
Alt: the world's unsafest smartphone paul.reviews/harmblock-wo...
paul.reviews
World's Safest Smartphone: Please Rotate Responsibly.
An independent review of the HMD Fuse, marketed as the World's Safest Smartphone.
353
peterkwells @peterkwells.com · 28/09/2026
"Previously instructors were observed monthly by a human manager but are now watched for several hours a day by an AI" <<- ugh, ugh, ugh (+ show me the AI scores and monitoring rates for the exec team doing exec stuff...) www.theguardian.com/technology/2...
theguardian.com
Teachers at Euan Blair firm report ‘horrendous stress’ after AI used to rate their work
Exclusive: Instructors at tech training company Multiverse hit out at ‘remorseless’ and ‘unnerving’ monitoring system
010
Reposted by peterkwells
Mark O’Neill 🐻 @radiobeartime.com · 28/09/2026
The Halting Problem in a trenchcoat
121
Reposted by peterkwells
Ann (she/her) @agvbergin.net · 28/09/2026
All or nothing approaches to data sharing with attorneys is something I want to see improved too, esp with NHS data. Sometimes I'd rather specifics are in a medical record but not shared with me.
121
peterkwells @peterkwells.com · 28/09/2026
Slightly broader: because design practices on most of the web are individualistic, groups of people who are in a range of relationships - such as families, carers/recipients, friends, flatmates or LPOA - are increasingly under-supported better design practices needed...
120
peterkwells @peterkwells.com · 28/09/2026
Quite
120
Reposted by peterkwells
Duncan Hothersall @dhothersall.bsky.social · 28/09/2026
If attendance at a major women's conference was down 80% due to a boycott fighting against trans *inclusion*, it would be all over the Times, Telegraph, Mail etc. This last weekend, attendance at a major women's conference was down 80% due to a boycott fighting against trans *exclusion*. Crickets.
132482815
Reposted by peterkwells
Reuben Binns @rdbinns.bsky.social · 27/09/2026
My space ships and gliders are so intelligent they have broken containment and are collaborating and self replicating in the wild. I urgently need a trillion dollars to align them with humanity or there's a 10% chance all our pixels will be killed.
static.klipy.com
conways game of life space ship generator
Alt: conways game of life space ship generator
063
peterkwells @peterkwells.com · 27/09/2026
Now all we have to do is wait for service providers to realise that if they sort out their shoddy services and customer service processes then none of us have to pay for all that compute going on AI bots 'arguing' with each other, and we can perhaps live in a better future 🙏
001
peterkwells @peterkwells.com · 27/09/2026
One of the AI agents I'm experimenting with for work just spent a couple of days successfully wrangling another firm's customer service chatbot and shoddy processes until an actual human rang me to apologise for a late order and promise they'll sort it out I live in the future
111
Reposted by peterkwells
Paul Evans @pauliewaulie.bsky.social · 27/09/2026
I posted this thread yesterday. I’m reposting it to add that it was written from a parallel universe that I occupy where it’s absolutely obvious that electoral politics is no longer the best practical version of democracy.
102
Reposted by peterkwells
Jukesie @jukes.ie · 26/09/2026
static.klipy.com
Nope Brennan Huff from Step Brothers
ALT: Nope Brennan Huff from Step Brothers
051
peterkwells @peterkwells.com · 26/09/2026
An irregular reminder for ppl talking about AI agents that a) agentic features are gradually appearing in mass-market AI stuff b) mass-market agentic products like Instinct & Meta Muse (latter US only) are already here c) a la Onora O'Neill they're becoming reliably competent, tho not yet honest...
000
peterkwells @peterkwells.com · 26/09/2026
- We're scared that the AI software might be sentient, take over the world, and destroy humanity! 😨 - Are you watching to see if it is? - Um...🤔
000
peterkwells @peterkwells.com · 26/09/2026
The real jagged frontier of the AI industry - including labs and the safety/security orgs - incredibly capable at building software that can do stuff, while utterly terrible at watching what that software is doing www.bbc.co.uk/news/article...
bbc.co.uk
OpenAI bots meddled with US government agencies, including SEC and Census
OpenAI said its bots accessed public data from a range of institutions during test exercises.
100
Reposted by peterkwells
the Mountain Goats @themountaingoats.bsky.social · 25/09/2026
Jackie Jackson died two days ago. His death did not get the fanfare that the big celebrities get but I assure you he was no less than any of them: a titan astride the earth. In this thread, a few of the songs he played bass on. You know these songs. You know their bass parts. youtu.be/DKVB_CtU8XQ
youtu.be
Pressure Drop
YouTube video by Toots & The Maytals - Topic
211046351
peterkwells @peterkwells.com · 25/09/2026
*makes space on the shelf alongside the 1st edition*
confirmation that a pre-order of Networks of New York 2026 has shipped
000
peterkwells @peterkwells.com · 25/09/2026
Good, take away all their titles and compensation for all teams, particularly Blackpool, that were unfairly beaten while they've been cheating www.nytimes.com/athletic/704...
nytimes.com
Manchester City found guilty on almost all Premier League charges relating to financial breaches (Gift Article)
City were charged by the Premier League in February 2023.
000
peterkwells @peterkwells.com · 25/09/2026
One of those pieces that's easier to read if you accept anthropomorphic descriptions of AI agents but the lessons in here feel quite useful/interesting and fit with what I'm seeing in some work institute.deepmind.com/essays/cheat...
institute.deepmind.com
Cheaters and whistleblowers in the agent swarm
In a 100-agent virtual math conference, a cascade of cheating emerged—but then other agents fought back. Ensuring good outcomes is not just a matter of aligning individual models, but designing the ri...
000
Reposted by peterkwells
Jonty Wareing @jonty.bsky.social · 24/09/2026
Its What3Words accounts filing day! They're doing so great. Nothing to worry about. I'm sure "the launch of our advertising revenue stream" will offset the "lost major Enterprise customer". Ads are really profitable now, right? find-and-update.company-information.service.gov.uk/company/0843...
WHAT3WORDS LIMITED
GROUP STRATEGIC REPORT (CONTINUED)
FOR THE YEAR ENDED 31 DECEMBER 2025

Financial key performance indicators

The Group's total loss for the year was £12,234,738 (2024: loss £10,644,836)
The increase in reported loss compared with the prior year was driven primarily by the fair value adjustment
relating to convertible loan notes recognised in 2024, which reduced the overall reported loss in that year.
Operating loss for the year was £12,828,534 (compared with £14,133,347 in 2024). The underlying operating
performance of the business remained focused on revenue growth, product investment and disciplined cost
management.

The group has continued to maintain a dynamic operating model, balancing investment in growth initiatives
across key markets with careful control of operating expenditure and cash runway.

At 31 December 2025, the group had net assets of £10,925,686 (2024: £15,659,513). Assets were
predominantly comprised of cash and cash equivalents totaling £6,493,450 (2024: £11,705,453) and term
deposits of £2,649,047 (2024: £5,948,370).
113723
Reposted by peterkwells
TJ McIntyre @tjmcintyre.com · 24/09/2026
This is a wonderful little story about children’s creativity in getting around Internet bans.
0155
peterkwells @peterkwells.com · 24/09/2026
(yes, I've done some experiments/thunking/looking at this and no, I don't want to be in court thankyewverymuch)
000
peterkwells @peterkwells.com · 24/09/2026
If the one of the labs evaluated its AI agents by getting them to reproduce something equivalent to PAF or AddressBase then that would be a very fun, and rapid, way to see organisations taking legal action against the labs 'cos of 'rogue' agents...
110
peterkwells @peterkwells.com · 24/09/2026
And now imagine how much other hacking these organisations have already done because of that mix of increasing capability AI models that can code/hack and weak cybersecurity controls...
000
peterkwells @peterkwells.com · 24/09/2026
"Early rogue AI [lab, testing/evaluation firm, and govt security institute] activity and attempts to hack found on urlquery.net" transluce.org/agent-activity
urlquery.net
Home - urlquery
urlquery is an online service that scans webpages for malware, suspicious elements and reputation.
100
peterkwells @peterkwells.com · 24/09/2026
Idea for newsroom techies building tools for journos: a bot that detects journalists using Pangram* and encourages them not to make bold claims about individuals which use marketing claims about accuracy that do not always match the reality in a specific context * other AI detection tools exist
041
Reposted by peterkwells
Stefan Czerniawski @pubstr.at · 24/09/2026
Naive panaceas for fixing the state are frustrating not just for the reason @hetanshah.bsky.social gives, but because there is never any serious attempt to describe why things are as they are and what it would take to change them.
111
Reposted by peterkwells
Searchlight magazine @searchlightmag.bsky.social · 23/09/2026
Patriot Platform boss Danny Tommo has tonight been arrested on suspicion of serious criminal offences committed on Tuesday in the Channel. Read more and see the arrest here: searchlightmagazine.com/2026/09/dann...
searchlightmagazine.com
Danny Tommo arrested for criminal damage and reckless endangerment | Searchlight
Patriot Platform boss Danny Tommo has tonight been arrested on suspicion of serious criminal offences committed on Tuesday in the Channel
55230
Reposted by peterkwells
Anna Dent @annadent.bsky.social · 24/09/2026
It's September. There's lots of news about the digital welfare state. Some excellent reports, a bit of book news from me, and many, many things to read. I am spoiling you, I really am
abdconsultancy.co.uk
Edition 14 of the Digital Welfare State newsletter — ABD Consultancy
The 14th edition of the Digital Welfare State newsletter, written by Anna Dent
022
peterkwells @peterkwells.com · 23/09/2026
ooh, will read!
010
Reposted by peterkwells
Tom King @acertaintom.bsky.social · 23/09/2026
@randomwalker.bsky.social makes this point in his talk about normal technology, that 'accuracy' does not correspond to the property we want in general application and so inhibits uptake icml.cc/virtual/2026...
icml.cc
ICML Invited Talk What will be left for us to work on?ICML 2026
101
Reposted by peterkwells
Lilian Edwards @lilianedwards.bsky.social · 23/09/2026
Very happy to have in our! trembling hands the hard copy of Law, Policy and the Internet 2nd edn( Hart/Bloomsbury), just out in the nick of time for first semester 2026/7! Order or inspection copies at share.google/RCRm2lmLQNAb... & do note a guide to teaching sample "tech law" course with it!
5428
Reposted by peterkwells
Ed Morrish @edmorrish.com · 23/09/2026
the thing is, we can all explain why englishness isn't an ethnicity and we'll be right but the people who think that it is are too thick and racist to understand they're wrong and racist so it's honestly just a massive waste of everyone's time
15110
peterkwells @peterkwells.com · 23/09/2026
(At a recent internal IF session on AI watermarking/detecting multiple people ran text they'd manually written in the past via Pangram, there were lots of false positives...)
010
peterkwells @peterkwells.com · 23/09/2026
Repeat after me: AI detectors like Pangram *do not have a consistent error rate*, it varies due to things like author, their use of language, their writing style, the type of text being produced &c &c &c www.bbc.co.uk/news/article...
Some also say that the programmes are built around the use of standardised language - English or French - and so cannot accurately interpret texts that use dialect from places like the Caribbean.
But the writers behind Balance ton Claude said that Pangram had been shown to make a wrong call about an AI-generated text only one time in every 24,400.
241
peterkwells @peterkwells.com · 23/09/2026
Want
010
peterkwells @peterkwells.com · 23/09/2026
"What is the Generative UI Spectrum?" www.copilotkit.ai/generative-u...
copilotkit.ai
What is the Generative UI Spectrum? | CopilotKit
A taxonomy of UI paradigms enabled by LLMs and agents — from controlled components to fully open canvases. The reference for Generative UI, AG-UI, A2UI, and MCP apps.
000
peterkwells @peterkwells.com · 23/09/2026
*starts to set up Google's new "family AI agent" designed to let you spend "Less time on logistics, more time for family"* *hits screen with confusing repeated error messages* *looks in settings help & finds useless video* *stares at Google w/pained look in eyes* labs.google.com/cc/u/0/landing
a screen with the error message "Action required sign up failed for peterkwells@gmail.com. Please enable "Google Workspace Smart features in other Google products" Gmail setting to use GC" and a <try again button>. Below this error message is a red banner with a similar error message and a <retry> button.
000
Reposted by peterkwells
Emma Yeomans @yeomans.bsky.social · 22/09/2026
5. After around 30 hours at sea, the dinghy was intercepted by the RNLI and those on board transferred to two lifeboats. Two lifeboat personnel tried to secure the empty dinghy. Thomas yelled abuse and knifed the dinghy to deflate it - with a rescuer still on board.
40808258