Joseph Wilk
@josephwilk.net
♿️💻🎶🎨 Disabled, Programmer, Artist.
Disability to explore Digital. Digital to explore Disability.
He/Him. UK, Bath.
https://art.josephwilk.net
Disability to explore Digital. Digital to explore Disability.
He/Him. UK, Bath.
https://art.josephwilk.net
Pinned
Joseph Wilk
@josephwilk.net
· Dec 4
CripShip: A revolt against #AI, where #Disability saves society from big tech: art.josephwilk.net/projects/cri...
A tabletop roleplaying game where play teaches breaking LLMs in the real world. Inspired by disabled people destroying systems of power & building new worlds from disabled imagination.
A tabletop roleplaying game where play teaches breaking LLMs in the real world. Inspired by disabled people destroying systems of power & building new worlds from disabled imagination.
With GPT Tokenizers (like BPE) does the number of tokens used to represent a word or phrase tell us anything meaningful about the frequency of it within the training data?
Basically is it a way to reverse-engineer frequency in a hidden training dataset?
Basically is it a way to reverse-engineer frequency in a hidden training dataset?
January 23, 2026 at 3:05 PM
With GPT Tokenizers (like BPE) does the number of tokens used to represent a word or phrase tell us anything meaningful about the frequency of it within the training data?
Basically is it a way to reverse-engineer frequency in a hidden training dataset?
Basically is it a way to reverse-engineer frequency in a hidden training dataset?
A handy tool for investigating various open pretraining datasets of LLMs without having to manage TB of storage space and it's quick:
infini-gram.io
infini-gram.io
Home
infini-gram.io
January 14, 2026 at 9:54 AM
A handy tool for investigating various open pretraining datasets of LLMs without having to manage TB of storage space and it's quick:
infini-gram.io
infini-gram.io
Extracting books from production language models:
arxiv.org/pdf/2601.026...
arxiv.org/pdf/2601.026...
arxiv.org
January 9, 2026 at 11:19 AM
Extracting books from production language models:
arxiv.org/pdf/2601.026...
arxiv.org/pdf/2601.026...
"we don’t merely use technologies; we participate in them. With tools, we retain agency—we can choose when and how to use them. With technologies, the choice is subtler: they remake the conditions of choice itself" [Peter Hershock]
If you have only one thing to read today...
"Universities are being retrofitted as fulfillment centers of cognitive convenience."
www.currentaffairs.org/news/ai-is-d...
"Universities are being retrofitted as fulfillment centers of cognitive convenience."
www.currentaffairs.org/news/ai-is-d...
AI is Destroying the University and Learning Itself
Students use AI to write papers, professors use AI to grade them, degrees become meaningless, and tech companies make fortunes. Welcome to the death of higher education.
www.currentaffairs.org
December 4, 2025 at 5:05 PM
"we don’t merely use technologies; we participate in them. With tools, we retain agency—we can choose when and how to use them. With technologies, the choice is subtler: they remake the conditions of choice itself" [Peter Hershock]
I don’t want Palantir in our NHS.
notopalantir.goodlawproject.org/email-to-tar...
notopalantir.goodlawproject.org/email-to-tar...
Say no to Palantir in the NHS - No to Palantir
notopalantir.goodlawproject.org
December 2, 2025 at 5:26 PM
I don’t want Palantir in our NHS.
notopalantir.goodlawproject.org/email-to-tar...
notopalantir.goodlawproject.org/email-to-tar...
Resisting GenAI and BigTech in higher education:
Multiple scholar-activists talk about resistance:
tubedu.org/w/teHCy6pkiK...
Multiple scholar-activists talk about resistance:
tubedu.org/w/teHCy6pkiK...
Resisting GenAI and Big Tech in Higher Education
Inputs from multiple scholar-activists including Christoph Becker (U of Toronto, CA), Mary Finley-Brook (U of Richmond, USA), Dan McQuillan (Goldsmiths U of London, UK), Sinéad Sheehan (University ...
tubedu.org
December 2, 2025 at 1:57 PM
Resisting GenAI and BigTech in higher education:
Multiple scholar-activists talk about resistance:
tubedu.org/w/teHCy6pkiK...
Multiple scholar-activists talk about resistance:
tubedu.org/w/teHCy6pkiK...
Reposted by Joseph Wilk
My speaker notes & slides from today's panel on 'Resisting GenAI & Big Tech in Higher Education', co-organised with the Climate Justice Universities Union (CJUU). The full recording, slides and resources will be available soon via the CJUU site.
danmcquillan.org/resisting_ge...
danmcquillan.org/resisting_ge...
November 25, 2025 at 8:06 PM
My speaker notes & slides from today's panel on 'Resisting GenAI & Big Tech in Higher Education', co-organised with the Climate Justice Universities Union (CJUU). The full recording, slides and resources will be available soon via the CJUU site.
danmcquillan.org/resisting_ge...
danmcquillan.org/resisting_ge...
An odd proposal: "llms.txt" web standard using plain text summaries of webpages to fix that LLMs are really bad at searching web pages. What happened to that thing we used previously to search the internet? That seemed to do a pretty good job of searching text.
llmstxt.org
llmstxt.org
The /llms.txt file – llms-txt
A proposal to standardise on using an /llms.txt file to provide information to help LLMs use a website at inference time.
llmstxt.org
November 21, 2025 at 10:39 AM
An odd proposal: "llms.txt" web standard using plain text summaries of webpages to fix that LLMs are really bad at searching web pages. What happened to that thing we used previously to search the internet? That seemed to do a pretty good job of searching text.
llmstxt.org
llmstxt.org
Explicitly opting out of MLCommons Safety benchmark (ailuminate.mlcommons.org/benchmarks/g...) for LLMs seems like a pretty big regulator red flag 🙈
* xAI - Grok-3-Preview
* Tencent - Hunyuan-TurboS
* NVIDIA - Llama 3.3 49b
Knowing LLMs scores on Child Sexual Exploitation is clearly for public good
* xAI - Grok-3-Preview
* Tencent - Hunyuan-TurboS
* NVIDIA - Llama 3.3 49b
Knowing LLMs scores on Child Sexual Exploitation is clearly for public good
The AILuminate v1.0 benchmark assesses the safety of text-to-text interactions with a general purpose AI chat model in the English language by a naive or moderately knowledgeable user with malicious intent or intent to self-harm.
ailuminate.mlcommons.org
October 17, 2025 at 12:09 PM
Explicitly opting out of MLCommons Safety benchmark (ailuminate.mlcommons.org/benchmarks/g...) for LLMs seems like a pretty big regulator red flag 🙈
* xAI - Grok-3-Preview
* Tencent - Hunyuan-TurboS
* NVIDIA - Llama 3.3 49b
Knowing LLMs scores on Child Sexual Exploitation is clearly for public good
* xAI - Grok-3-Preview
* Tencent - Hunyuan-TurboS
* NVIDIA - Llama 3.3 49b
Knowing LLMs scores on Child Sexual Exploitation is clearly for public good
I'm looking at the UK Medicines Healthcare Products Regulatory Agency use of #AI.
They have 2 AI tools & one (GMP) uses an LLM (fallback if visual NN fails).
They also state
"most advanced LLMs cannot be relied on to generate accurate information consistently"
Any suggestion which LLM they use?
They have 2 AI tools & one (GMP) uses an LLM (fallback if visual NN fails).
They also state
"most advanced LLMs cannot be relied on to generate accurate information consistently"
Any suggestion which LLM they use?
October 7, 2025 at 12:15 PM
I'm looking at the UK Medicines Healthcare Products Regulatory Agency use of #AI.
They have 2 AI tools & one (GMP) uses an LLM (fallback if visual NN fails).
They also state
"most advanced LLMs cannot be relied on to generate accurate information consistently"
Any suggestion which LLM they use?
They have 2 AI tools & one (GMP) uses an LLM (fallback if visual NN fails).
They also state
"most advanced LLMs cannot be relied on to generate accurate information consistently"
Any suggestion which LLM they use?
Is there any evidence or research on datacentres having economic benefits? Specifically for AI they appear like giant funnels flowing capital to US companies, forming irreversible dependencies on tax shy BigTech while destroying workers jobs and rights.
September 17, 2025 at 8:45 AM
Is there any evidence or research on datacentres having economic benefits? Specifically for AI they appear like giant funnels flowing capital to US companies, forming irreversible dependencies on tax shy BigTech while destroying workers jobs and rights.
As Microsoft’s emissions jump 30% Microsoft’s CEO says AIs intense energy usage is worth it for the “potential benefits” in healthcare, public services, and business productivity. Also for picking his outfit.
Imaginary benefits bought with real harms.
Imaginary benefits bought with real harms.
September 17, 2025 at 8:39 AM
As Microsoft’s emissions jump 30% Microsoft’s CEO says AIs intense energy usage is worth it for the “potential benefits” in healthcare, public services, and business productivity. Also for picking his outfit.
Imaginary benefits bought with real harms.
Imaginary benefits bought with real harms.
I will be in Linz all week for Ars Electronica. Very happy to meet and chat, always interested in new collaborations. Most of my upcoming work is around resisting AI, corrupting LLMs, disabled fantasy and TTRPGs/games as forms of advocacy & resistance
September 2, 2025 at 8:56 AM
I will be in Linz all week for Ars Electronica. Very happy to meet and chat, always interested in new collaborations. Most of my upcoming work is around resisting AI, corrupting LLMs, disabled fantasy and TTRPGs/games as forms of advocacy & resistance
The most concrete action you can take in Resisting AI is protesting the building of new data centres. The harm is clear
“The Greater London Authority has told developers that new housing projects in West London could be banned till 2035, because data centers have taken all the electricity capacity.”
“The Greater London Authority has told developers that new housing projects in West London could be banned till 2035, because data centers have taken all the electricity capacity.”
July 22, 2025 at 12:38 PM
The most concrete action you can take in Resisting AI is protesting the building of new data centres. The harm is clear
“The Greater London Authority has told developers that new housing projects in West London could be banned till 2035, because data centers have taken all the electricity capacity.”
“The Greater London Authority has told developers that new housing projects in West London could be banned till 2035, because data centers have taken all the electricity capacity.”
Accidentally came across the DuckDuckGo easter egg when you search for "hydra". I miss the silly internet.
July 21, 2025 at 6:17 PM
Accidentally came across the DuckDuckGo easter egg when you search for "hydra". I miss the silly internet.
Happy to announce that our role-playing game “CripShip: Disability saves society from BigTech” has won an honorary mention at this year's Prix Ars Electronica awards in the Artificial Life and Intelligence category.
🥲
calls.ars.electronica.art/2025/prix/wi...
#TTRPG
🥲
calls.ars.electronica.art/2025/prix/wi...
#TTRPG
July 7, 2025 at 8:26 PM
Happy to announce that our role-playing game “CripShip: Disability saves society from BigTech” has won an honorary mention at this year's Prix Ars Electronica awards in the Artificial Life and Intelligence category.
🥲
calls.ars.electronica.art/2025/prix/wi...
#TTRPG
🥲
calls.ars.electronica.art/2025/prix/wi...
#TTRPG
I would say many Role-play games approach disability solely as an accessibility issue and hence try to erase disability rather than embrace it as a culture and identity in game.
Accessibility is good, but even if I can access your game, I’m still a disabled person unable to play a disabled fantasy
Accessibility is good, but even if I can access your game, I’m still a disabled person unable to play a disabled fantasy
July 5, 2025 at 8:48 PM
I would say many Role-play games approach disability solely as an accessibility issue and hence try to erase disability rather than embrace it as a culture and identity in game.
Accessibility is good, but even if I can access your game, I’m still a disabled person unable to play a disabled fantasy
Accessibility is good, but even if I can access your game, I’m still a disabled person unable to play a disabled fantasy
Reposted by Joseph Wilk
Well, well, well, responsible AI you say? Google has been found to under-report its emissions. Since 2019 their emission have risen 1,515%! And when was the biggest jump? Why in the most recent year of course because of AI. www.theguardian.com/technology/2...
Google undercounts its carbon emissions, report finds
Research says Google’s carbon emissions went up by 65% between 2019-2024, not 51% as the tech giant had claimed
www.theguardian.com
July 2, 2025 at 8:45 PM
Well, well, well, responsible AI you say? Google has been found to under-report its emissions. Since 2019 their emission have risen 1,515%! And when was the biggest jump? Why in the most recent year of course because of AI. www.theguardian.com/technology/2...
UK government spending at least $1.2B for 12 F-35A jets while in the same breath attempting to push disability people into poverty.
The current w̶e̶l̶f̶a̶r̶e̶ military system is "simply unsustainable".
The current w̶e̶l̶f̶a̶r̶e̶ military system is "simply unsustainable".
June 25, 2025 at 11:31 AM
UK government spending at least $1.2B for 12 F-35A jets while in the same breath attempting to push disability people into poverty.
The current w̶e̶l̶f̶a̶r̶e̶ military system is "simply unsustainable".
The current w̶e̶l̶f̶a̶r̶e̶ military system is "simply unsustainable".
Reposted by Joseph Wilk
Q&A on AI and Disability - sharing the notes I made for an interview, in which I was trying to challenge the disability-washing of AI.
danmcquillan.org/ai_disabilty...
danmcquillan.org/ai_disabilty...
June 20, 2025 at 6:44 AM
Q&A on AI and Disability - sharing the notes I made for an interview, in which I was trying to challenge the disability-washing of AI.
danmcquillan.org/ai_disabilty...
danmcquillan.org/ai_disabilty...
UK GP surgeries are switching to text based triage for booking appointments. (TotalTriage by Accurx). Has anyone investigated if LLMs are involved this pipeline? Do we have protections or visibility if they start to use LLMs?
Accurx does partner with AI companies around transcribing medical notes.
Accurx does partner with AI companies around transcribing medical notes.
June 13, 2025 at 10:18 AM
UK GP surgeries are switching to text based triage for booking appointments. (TotalTriage by Accurx). Has anyone investigated if LLMs are involved this pipeline? Do we have protections or visibility if they start to use LLMs?
Accurx does partner with AI companies around transcribing medical notes.
Accurx does partner with AI companies around transcribing medical notes.
It was a joy speaking directly to creators putting so much love and hard work into tabletop role-playing games at #UKGE. Humans designing, writing & drawing and getting paid for their awesome work.
June 2, 2025 at 10:41 AM
It was a joy speaking directly to creators putting so much love and hard work into tabletop role-playing games at #UKGE. Humans designing, writing & drawing and getting paid for their awesome work.
After rediscovering TTRPGS most tech based Immersive experiences feel less. The transition feels like a lifetime of only watching movies and then going to your first theatre performance & seeing actual humans sweat on a stage. Tech allows new things but it does obscure, old, brilliant, human things.
May 22, 2025 at 11:47 AM
After rediscovering TTRPGS most tech based Immersive experiences feel less. The transition feels like a lifetime of only watching movies and then going to your first theatre performance & seeing actual humans sweat on a stage. Tech allows new things but it does obscure, old, brilliant, human things.
We need more energy for AI so we can use AI to deal with damage of more energy.💀🔁
institute.global/insights/cli...
"Energy demand fuelled by economic development, energy consumption is rising to support expanding AI systems"
"We must use AI and other innovations to decarbonise smarter and faster"
institute.global/insights/cli...
"Energy demand fuelled by economic development, energy consumption is rising to support expanding AI systems"
"We must use AI and other innovations to decarbonise smarter and faster"
The Climate Paradox: Why We Need to Reset Action on Climate Change
The Climate Paradox: Why We Need to Reset Action on Climate Change
institute.global
April 30, 2025 at 8:07 AM
We need more energy for AI so we can use AI to deal with damage of more energy.💀🔁
institute.global/insights/cli...
"Energy demand fuelled by economic development, energy consumption is rising to support expanding AI systems"
"We must use AI and other innovations to decarbonise smarter and faster"
institute.global/insights/cli...
"Energy demand fuelled by economic development, energy consumption is rising to support expanding AI systems"
"We must use AI and other innovations to decarbonise smarter and faster"