Polly Thompson's Avatar

Polly Thompson

@pollyrt

"surprisingly people focussed for a tech person" | "stronger than you look" | "so embarrassing"

286
Followers
260
Following
152
Posts
20.08.2023
Joined
Posts Following

Latest posts by Polly Thompson @pollyrt

Alt text as a literary art form. Superb!

01.03.2026 13:50 ๐Ÿ‘ 17 ๐Ÿ” 4 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0
Preview
axe-con Digital Accessibility Conference | Deque Join us to learn how to build, test, monitor, and maintain digital accessibility. Hosted on February 24-25, 2026, axe-con is completely free and virtual.

Little reminder that axe-con, organised by @deque.com is this week!!
It's a free online, two days (February 24 25) event, with a very nice line up of talks about accessibility and inclusive design.
Want to learn about accessibility? Register, check them live, or catch the replays!

#Accessibility

23.02.2026 15:30 ๐Ÿ‘ 14 ๐Ÿ” 13 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 2
Preview
What if boldness were an explicit value of the civil service? Iโ€™ve been thinking a lot recently about what it takes to get difficult and meaningful things done. Why itโ€™s hard, exhausting and incrediblyโ€ฆ

I've been thinking a lot about boldness this week, so revisited this classic post from @janethughes.bsky.social. I can't it's 10 years old! medium.com/public-innov...

13.02.2026 10:35 ๐Ÿ‘ 5 ๐Ÿ” 0 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0
Preview
TBM 405: Hope, Context, and Control This essay is about hope, fear, AI, and the tension between control and collective sensemaking.

Techno-authoritarian thinkers elevate centralized, algorithmic control above humanistic complexity. People who bought into these areas (with their wallets and political donations) are on product podcasts talking about how great everything is.

cutlefish.substack.com/p/tbm-405-ho...

07.02.2026 20:10 ๐Ÿ‘ 7 ๐Ÿ” 3 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0
Preview
gradient.horse Draw a horse, watch it run!

omg everybody go draw a horse this is what the internet was made for

gradient.horse

09.02.2026 23:11 ๐Ÿ‘ 7009 ๐Ÿ” 3609 ๐Ÿ’ฌ 35 ๐Ÿ“Œ 154
From the NYT Evening newsletter, headline reads "A.I. bots are often wrong, but they're set to reshape medicine". Highlighted text reads: "A new study published today found that A.I. chatbots were no better than Google at guiding users towards correct or helpful health advice. The technology sometimes presented false information or dramatically changed its advice depending on slight changes in the wording of the question. But A.I. has already proved quite good at other tasks, like at reading scans and images - better than many doctor, in fact."

From the NYT Evening newsletter, headline reads "A.I. bots are often wrong, but they're set to reshape medicine". Highlighted text reads: "A new study published today found that A.I. chatbots were no better than Google at guiding users towards correct or helpful health advice. The technology sometimes presented false information or dramatically changed its advice depending on slight changes in the wording of the question. But A.I. has already proved quite good at other tasks, like at reading scans and images - better than many doctor, in fact."

This is a framing error, NYT. It's different technologies. Chatbots are not the same as the machine learning involved in reading scans (better defined inputs, clearer evaluation of outputs). Calling both AI is like conflating bombs dropped from planes with cars, because they both involve explosions.

10.02.2026 04:39 ๐Ÿ‘ 136 ๐Ÿ” 31 ๐Ÿ’ฌ 5 ๐Ÿ“Œ 5

Reminds me of discourse in which women are called "females". I always think it implies an appeal to biology= science= objectivity.

09.02.2026 14:45 ๐Ÿ‘ 6 ๐Ÿ” 0 ๐Ÿ’ฌ 0 ๐Ÿ“Œ 0

LinkedIn needs to add a ๐Ÿคฎ response option.

09.02.2026 12:57 ๐Ÿ‘ 4 ๐Ÿ” 0 ๐Ÿ’ฌ 0 ๐Ÿ“Œ 0

Yeah Iโ€™m looking for my dog (girlfriend who has a restraining order against me and she has a dog) and hereโ€™s the picture of the dog (she walks this dog and I can use it to figure out her schedule) thanks for sending all footage my way

09.02.2026 04:13 ๐Ÿ‘ 5396 ๐Ÿ” 1422 ๐Ÿ’ฌ 55 ๐Ÿ“Œ 39

I'm convinced AI is our generation's radium - a discovery with genuinely useful applications in specific, controlled circumstances that we stupidly put in everything from kid's toys to toothpaste until we realised the harm far too late where future generations will ask if we were out of our minds.

08.02.2026 22:23 ๐Ÿ‘ 17926 ๐Ÿ” 5442 ๐Ÿ’ฌ 245 ๐Ÿ“Œ 258
Preview
endless.downward.spiral - is this the beginning of the end of What3Words? Long-time readers know that I am not a fan of What Three Words. I think it is a closed, proprietary, and user-unfriendly attempt to enclose the commons. I consider that it has some dangerous failure m...

endless.downward.spiral - is this the beginning of the end of What3Words?
shkspr.mobi/blog/2025/02...

08.02.2026 09:54 ๐Ÿ‘ 13 ๐Ÿ” 8 ๐Ÿ’ฌ 3 ๐Ÿ“Œ 0
Benjamin De Kraker O
@BenjaminDEKR
Follow
OpenClaw is interesting, but will also drain your wallet if you aren't careful.
Last night around midnight I loaded my Anthropic API account with $20, then went to bed.
When I woke up, my Anthropic balance was $O.
Opus was checking "is it daytime yet?" every 30 minutes, paying $0.75 each time to conclude "no, it's still night."
Doing literally nothing, OpenClaw spent the entire balance. How?

Benjamin De Kraker O @BenjaminDEKR Follow OpenClaw is interesting, but will also drain your wallet if you aren't careful. Last night around midnight I loaded my Anthropic API account with $20, then went to bed. When I woke up, my Anthropic balance was $O. Opus was checking "is it daytime yet?" every 30 minutes, paying $0.75 each time to conclude "no, it's still night." Doing literally nothing, OpenClaw spent the entire balance. How?

The "Heartbeat" cron job, even though literally the only thing I had going was one silly reminder, ("remind me tomorrow to get milk")
1. Sent ~120,000 tokens of context to Opus
4.5
2. Opus read HEARTBEAT md, thought about reminders
3. Replied "HEARTBEAT_OK"
4. Cost: ~$0.75 per heartbeat (cache writes)
The damage:
- Overnight = ~25+ heartbeats
- 25 ร— $0.75 = ~$18.75 just from heartbeats alone
- Plus regular conversation = ~$20 total
The absurdity: Opus was essentially checking
"is it daytime yet?" every 30 minutes, paying $0.75 each time to conclude "no, it's still night."

The "Heartbeat" cron job, even though literally the only thing I had going was one silly reminder, ("remind me tomorrow to get milk") 1. Sent ~120,000 tokens of context to Opus 4.5 2. Opus read HEARTBEAT md, thought about reminders 3. Replied "HEARTBEAT_OK" 4. Cost: ~$0.75 per heartbeat (cache writes) The damage: - Overnight = ~25+ heartbeats - 25 ร— $0.75 = ~$18.75 just from heartbeats alone - Plus regular conversation = ~$20 total The absurdity: Opus was essentially checking "is it daytime yet?" every 30 minutes, paying $0.75 each time to conclude "no, it's still night."

Iโ€™m starting to think the people who are excited about โ€œAI agentsโ€ have literally never used a computer in their lives

01.02.2026 02:56 ๐Ÿ‘ 2700 ๐Ÿ” 519 ๐Ÿ’ฌ 13 ๐Ÿ“Œ 149
Preview
Between 2017 and 2021 One Team Gov held space weekly for people working across government to come together, connect, reflect and support each others work. Some local meetups like in Manchester haveโ€ฆ |... Between 2017 and 2021 One Team Gov held space weekly for people working across government to come together, connect, reflect and support each others work. Some local meetups like in Manchester have be...

In 2026 we are reinstating the regular One Team Gov breakfast club in London and also running a virtual one for anybody who doesn't have a meetup near them.

If you're interested you can sign up for both using the link (and please retweet/skeet!)

www.linkedin.com/posts/onetea...

11.12.2025 13:57 ๐Ÿ‘ 28 ๐Ÿ” 22 ๐Ÿ’ฌ 2 ๐Ÿ“Œ 2

Data protection, as a concept, grew out of post-war legal scholarship in Germany which reflected on the use of data about people by the Nazi Govt, and formulated a set of limits on data use which recognised that to collect and use data about people was to impact the peopleโ€™s lives.

01.02.2026 08:45 ๐Ÿ‘ 197 ๐Ÿ” 101 ๐Ÿ’ฌ 2 ๐Ÿ“Œ 3
Video thumbnail

โค๏ธ

30.01.2026 19:57 ๐Ÿ‘ 28073 ๐Ÿ” 9225 ๐Ÿ’ฌ 362 ๐Ÿ“Œ 760

Reading a thread in the other place about resistance, and a sentence sticks with me:

"It is ok to be discouraged, but don't discourage others."

That applies to so much more than just resistance.

29.01.2026 07:36 ๐Ÿ‘ 8 ๐Ÿ” 1 ๐Ÿ’ฌ 0 ๐Ÿ“Œ 0

Bluesky hivemind - views please!

If you were writing an academic-adjacent piece and needed to include clear but accessible stuff on AI / algorithmic bias and its societal impacts, what are your go-to texts?

I have lots but want to make sure I'm not missing a classic!

RTs welcome ๐Ÿ™‚

26.01.2026 11:29 ๐Ÿ‘ 8 ๐Ÿ” 8 ๐Ÿ’ฌ 5 ๐Ÿ“Œ 0

Thanks for giving us some practice at #ukgc26 @jacattell.bsky.social and @mikerose.co.uk
This would make a good sticker โค๏ธ

25.01.2026 16:32 ๐Ÿ‘ 2 ๐Ÿ” 0 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0

For fans of profound wisdoms gleaned from colexifications:

A lot of languages do not make a difference between "lazy" and "tired".

23.01.2026 09:45 ๐Ÿ‘ 36 ๐Ÿ” 7 ๐Ÿ’ฌ 2 ๐Ÿ“Œ 1

The idea that data is objective and good data practice means eliminating bias has it backwards. Data is subjective, and good data practice means tracing and identifying how the biases inherent to the data we construct serve or donโ€™t serve specific peoples and ends to which that data is put to use.

21.01.2026 03:11 ๐Ÿ‘ 43 ๐Ÿ” 8 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0
Preview
Power and participation in public tech This conference is for anyone interested in how to achieve better outcomes in the delivery of digital, data and AI work in the public sector through the involvement of the public, communities and workers. Come if youโ€™re a public servant trying to engage or involve the public in your work on technology. Come if youโ€™re working in civil society, in the union movement, or with grassroots groups, trying to be heard. Come to share what youโ€™re up to, and to learn from others; leave with new insights, ideas, and connections.

We are running a conference on 'Power and participation in public tech' and registration is now open to all.

๐Ÿ“† Friday 6th March
๐Ÿ•ฐ๏ธ 10:00 - 17:00
๐Ÿข Manchester, UK

๐Ÿ”— connectedbydata.org/events/2026...

(1/X)

20.01.2026 17:00 ๐Ÿ‘ 9 ๐Ÿ” 9 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 1

Sorry to hear that- that sounds tough x

20.01.2026 19:25 ๐Ÿ‘ 1 ๐Ÿ” 0 ๐Ÿ’ฌ 0 ๐Ÿ“Œ 0
Preview
TransformGov Talks: January 2026 ยท Luma Join us at 5.30pm for a 6pm start. Matt Wood-Hill, MHCLG, and Linda O'Halloran, TPXimpact on public service market-shaping Patrick Diamond, Professor in Publicโ€ฆ

๐Ÿšจ 6PM, WED 28 January ๐Ÿšจ

At TPXImpact or online

Matt Wood-Hill, MHCLG, and Linda Oโ€™Halloran, TPXimpact on public service market-shaping.

โ€‹Patrick Diamond, Professor in Public Policy, Queen Mary, University of London, on government effectiveness

luma.com/pc7ji8sg

Please share! Thanks ๐Ÿ™

19.01.2026 13:10 ๐Ÿ‘ 6 ๐Ÿ” 10 ๐Ÿ’ฌ 0 ๐Ÿ“Œ 3

Copilot doesn't really work for many of the use cases in LG for various reasons. There's definitely demand for something more specific - currently met by one product that has been widely adopted but is v expensive ๐Ÿ‘€

19.01.2026 13:31 ๐Ÿ‘ 0 ๐Ÿ” 0 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0

Also- hello! I was hoping you'd be at #ukgc26. Would be really lovely to catch up

19.01.2026 12:20 ๐Ÿ‘ 1 ๐Ÿ” 0 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0

Yes! We were in the pilot group of LAs for Minute, but it wasn't clear what was happening with it post -pilot. But the team heard my pitch and said hello ๐ŸŽ‰

19.01.2026 12:19 ๐Ÿ‘ 1 ๐Ÿ” 0 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0

My specific option is "Transcription and structured document drafting seems to be emerging as the "killer app" of gen AI in local gov. Could we/how could we build and operate a shared product to avoid suppliers creaming off too much of the potential saving/benefit?"
Does that sound ok? #ukgc26

17.01.2026 10:49 ๐Ÿ‘ 5 ๐Ÿ” 0 ๐Ÿ’ฌ 2 ๐Ÿ“Œ 0

That's a good point! Big groups need a bit of planning so I'm a bit late for that

17.01.2026 10:33 ๐Ÿ‘ 1 ๐Ÿ” 0 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 0

I'm thinking of pitching at #ukgc26
Something local gov. Is it better to go for general or specific topics/questions?

17.01.2026 10:31 ๐Ÿ‘ 4 ๐Ÿ” 0 ๐Ÿ’ฌ 3 ๐Ÿ“Œ 0

@ukgovcamp.com a gentle nudge to campers travelling from London: these travel problems are what colleagues in the regions face whenever called to meetings or office days in London. This is a great chance to understand their pain, and consider meeting start/finish times etc in future ! ๐Ÿ˜‰
#ukgc26

17.01.2026 10:02 ๐Ÿ‘ 34 ๐Ÿ” 5 ๐Ÿ’ฌ 1 ๐Ÿ“Œ 1