Paul Allen 🌐🇨🇦
@paulallen.ca
Clinical director of intensive mental health services for children and youth in Canada (ret.). Believer in Truth and Reconciliation. Denier of Denialism. Student of semantics and statistics. "AI" too.
Reposted by Paul Allen 🌐🇨🇦
If I’m understanding this correctly, X is owned by a white nationalist who pays poor people of color in developing countries to pretend to be working class white Americans to scare other white Americans into being afraid poor people of color from developing countries are going to ruin America?
November 23, 2025 at 7:30 PM
If I’m understanding this correctly, X is owned by a white nationalist who pays poor people of color in developing countries to pretend to be working class white Americans to scare other white Americans into being afraid poor people of color from developing countries are going to ruin America?
Reposted by Paul Allen 🌐🇨🇦
According to "Instagram’s former head of safety and well-being Vaishnavi Jayakumar,... the company had a “17x” strike policy for accounts that reportedly engaged in the “trafficking of humans for sex.” time.com/7336204/meta...
7 Allegations Against Meta in Newly Unsealed Filings
Court filings allege Meta tolerated sex trafficking, hid harms to teens, and prioritized growth over user safety for years.
time.com
November 23, 2025 at 1:48 AM
According to "Instagram’s former head of safety and well-being Vaishnavi Jayakumar,... the company had a “17x” strike policy for accounts that reportedly engaged in the “trafficking of humans for sex.” time.com/7336204/meta...
Reposted by Paul Allen 🌐🇨🇦
Holy crap, you REALLY CAN attack complexity theory via measure theory...if you go deep enough into the weeds of descriptive set theory.
Welp, I guess I know what I need to start digging into as soon as finals are done.
Welp, I guess I know what I need to start digging into as soon as finals are done.
Separating complexity classes of LCL problems on grids
We study the complexity of locally checkable labeling (LCL) problems on $\mathbb{Z}^n$ from the point of view of descriptive set theory, computability theory, and factors of i.i.d. Our results separat...
arxiv.org
November 23, 2025 at 1:14 AM
Holy crap, you REALLY CAN attack complexity theory via measure theory...if you go deep enough into the weeds of descriptive set theory.
Welp, I guess I know what I need to start digging into as soon as finals are done.
Welp, I guess I know what I need to start digging into as soon as finals are done.
Reposted by Paul Allen 🌐🇨🇦
Reposted by Paul Allen 🌐🇨🇦
‼️ Friends, this is very important. Please share to whoever this may concern, especially if you are in the United States.
Ukraine wants peace. Lasting, just, sustainable peace, and not a pause before the next bloodbath. I cannot stress this enough.
Ukraine wants peace. Lasting, just, sustainable peace, and not a pause before the next bloodbath. I cannot stress this enough.
November 22, 2025 at 12:00 PM
‼️ Friends, this is very important. Please share to whoever this may concern, especially if you are in the United States.
Ukraine wants peace. Lasting, just, sustainable peace, and not a pause before the next bloodbath. I cannot stress this enough.
Ukraine wants peace. Lasting, just, sustainable peace, and not a pause before the next bloodbath. I cannot stress this enough.
Reposted by Paul Allen 🌐🇨🇦
But unfortunately deepfake detection models can fail in real-world settings... so in most places this presents a real challenge for users and the platforms to deal with - the technology deployed to detect this content isn't a foolproof response. arxiv.org/abs/2510.16556
Fit for Purpose? Deepfake Detection in the Real World
The rapid proliferation of AI-generated content, driven by advances in generative adversarial networks, diffusion models, and multimodal large language models, has made the creation and dissemination ...
arxiv.org
November 22, 2025 at 8:00 AM
But unfortunately deepfake detection models can fail in real-world settings... so in most places this presents a real challenge for users and the platforms to deal with - the technology deployed to detect this content isn't a foolproof response. arxiv.org/abs/2510.16556
Reposted by Paul Allen 🌐🇨🇦
Influencers made millions pushing ‘wild’ births – now the Free Birth Society is linked to baby deaths around the world www.theguardian.com/world/ng-int...
Influencers made millions pushing ‘wild’ births – now the Free Birth Society is linked to baby deaths around the world
A year-long investigation reveals how mothers lost children after being radicalised by uplifting podcast tales of births without midwives or doctors
www.theguardian.com
November 22, 2025 at 7:57 AM
Influencers made millions pushing ‘wild’ births – now the Free Birth Society is linked to baby deaths around the world www.theguardian.com/world/ng-int...
Reposted by Paul Allen 🌐🇨🇦
Training LLMs end to end is hard. But way more people should, and will, be doing it in the future.
The @hf.co Research team is excited to share their new e-book that covers the full pipeline:
· pre-training,
· post-training,
· infra.
200+ pages of what worked and what didn’t. ⤵️
The @hf.co Research team is excited to share their new e-book that covers the full pipeline:
· pre-training,
· post-training,
· infra.
200+ pages of what worked and what didn’t. ⤵️
November 2, 2025 at 3:17 PM
Training LLMs end to end is hard. But way more people should, and will, be doing it in the future.
The @hf.co Research team is excited to share their new e-book that covers the full pipeline:
· pre-training,
· post-training,
· infra.
200+ pages of what worked and what didn’t. ⤵️
The @hf.co Research team is excited to share their new e-book that covers the full pipeline:
· pre-training,
· post-training,
· infra.
200+ pages of what worked and what didn’t. ⤵️
Reposted by Paul Allen 🌐🇨🇦
Really excited about this 5-year project starting, funded by @wellcometrust.bsky.social, where we will combine ✨computational modelling✨ with causal intervention techniques such as ✨tFUS✨ to study the neurocognitive mechanisms of repetitive negative thoughts 🚀 led by @mikebrowning.bsky.social
Repetitive negative thoughts will be investigated using a range of cutting-edge brain science techniques as part of a new study led by the Department of Psychiatry, University of Oxford and funded by Wellcome @wellcometrust.bsky.social 👇
tinyurl.com/364es88k
tinyurl.com/364es88k
November 21, 2025 at 1:38 PM
Really excited about this 5-year project starting, funded by @wellcometrust.bsky.social, where we will combine ✨computational modelling✨ with causal intervention techniques such as ✨tFUS✨ to study the neurocognitive mechanisms of repetitive negative thoughts 🚀 led by @mikebrowning.bsky.social
Reposted by Paul Allen 🌐🇨🇦
People learn through practice and challenge; not by receiving answers.
But a deeper problem is that an essential aspect of teaching is helping someone organize their thinking in new ways. LLMs — systems which cannot think or appreciate thinking — are incapable of doing this in a meaningful way.
But a deeper problem is that an essential aspect of teaching is helping someone organize their thinking in new ways. LLMs — systems which cannot think or appreciate thinking — are incapable of doing this in a meaningful way.
Relying on ChatGPT to teach you about a topic leaves you with shallower knowledge than Googling and reading about it, according to new research that compared what more than 10,000 people knew after using one method or the other.
Shared by @gizmodo.com: buff.ly/yAAHtHq
Shared by @gizmodo.com: buff.ly/yAAHtHq
November 21, 2025 at 1:29 PM
People learn through practice and challenge; not by receiving answers.
But a deeper problem is that an essential aspect of teaching is helping someone organize their thinking in new ways. LLMs — systems which cannot think or appreciate thinking — are incapable of doing this in a meaningful way.
But a deeper problem is that an essential aspect of teaching is helping someone organize their thinking in new ways. LLMs — systems which cannot think or appreciate thinking — are incapable of doing this in a meaningful way.
Reposted by Paul Allen 🌐🇨🇦
New study on LLMs shows that while LLMs & humans converge on similar judgments of reliability of news media, they rely on very different underlying processes.
In delegating, are we confusing linguistic plausibility with epistemic reliability?
The age of "epistemia"
www.pnas.org/doi/epdf/10....
In delegating, are we confusing linguistic plausibility with epistemic reliability?
The age of "epistemia"
www.pnas.org/doi/epdf/10....
November 21, 2025 at 10:36 AM
New study on LLMs shows that while LLMs & humans converge on similar judgments of reliability of news media, they rely on very different underlying processes.
In delegating, are we confusing linguistic plausibility with epistemic reliability?
The age of "epistemia"
www.pnas.org/doi/epdf/10....
In delegating, are we confusing linguistic plausibility with epistemic reliability?
The age of "epistemia"
www.pnas.org/doi/epdf/10....
Reposted by Paul Allen 🌐🇨🇦
Building datasets to train smaller, task-focused models used to be incredibly time-consuming.
Very excited to see SAM3 massively lower that barrier. Describe the class you want to detect and get annotated datasets automatically!
Try it yourself: huggingface.co/datasets/uv-...!
Very excited to see SAM3 massively lower that barrier. Describe the class you want to detect and get annotated datasets automatically!
Try it yourself: huggingface.co/datasets/uv-...!
November 21, 2025 at 1:30 PM
Building datasets to train smaller, task-focused models used to be incredibly time-consuming.
Very excited to see SAM3 massively lower that barrier. Describe the class you want to detect and get annotated datasets automatically!
Try it yourself: huggingface.co/datasets/uv-...!
Very excited to see SAM3 massively lower that barrier. Describe the class you want to detect and get annotated datasets automatically!
Try it yourself: huggingface.co/datasets/uv-...!
Reposted by Paul Allen 🌐🇨🇦
This new model opens a whole new world of analysing multi region interaction across trials and tasks! More analysis and findings can be found in our paper linked below. Work lead by Jack Cook, and with great help from @danakarca.bsky.social and @somnirons.bsky.social !
arxiv.org/abs/2506.02813
arxiv.org/abs/2506.02813
Brain-Like Processing Pathways Form in Models With Heterogeneous Experts
Examples of such pathways can be found in the interactions between cortical and subcortical networks during learning, or in sub-networks specializing for task characteristics such as difficulty or mod...
arxiv.org
November 21, 2025 at 12:01 PM
This new model opens a whole new world of analysing multi region interaction across trials and tasks! More analysis and findings can be found in our paper linked below. Work lead by Jack Cook, and with great help from @danakarca.bsky.social and @somnirons.bsky.social !
arxiv.org/abs/2506.02813
arxiv.org/abs/2506.02813
Reposted by Paul Allen 🌐🇨🇦
I was thinking about the efficiency of nonparametric tests and remembered this SO post about the Wilcox being 96% as efficient as T test, even in small samples. For those who have a copy, Lehman & Romano did an thorough job in <5 pages detailing situation. #statssky #statistics #rstats #AcademicSky
November 21, 2025 at 1:22 PM
I was thinking about the efficiency of nonparametric tests and remembered this SO post about the Wilcox being 96% as efficient as T test, even in small samples. For those who have a copy, Lehman & Romano did an thorough job in <5 pages detailing situation. #statssky #statistics #rstats #AcademicSky
Reposted by Paul Allen 🌐🇨🇦
“Canada is facing near-term threats to its economic competitiveness and grappling with declining productivity and prosperity, so waiting years for AI investments to create value isn’t realistic in this environment – in fact, it’s downright risky.”
Yet Another Study Shows That Most Companies Aren't Making Any Money Off AI
Further proof has arrived that the ROI on AI is essentially non-existent.
gizmodo.com
November 21, 2025 at 1:51 PM
“Canada is facing near-term threats to its economic competitiveness and grappling with declining productivity and prosperity, so waiting years for AI investments to create value isn’t realistic in this environment – in fact, it’s downright risky.”
Reposted by Paul Allen 🌐🇨🇦
“To add insult to injury, many metaphorical … phrases (especially used to describe ANNs; see Table 1) — like train, learn, hallucinate, reason — are applied to machines and result in distorting how we perceive these machines: humanising them while dehumanising us”
zenodo.org/records/1706...
14/🧵
zenodo.org/records/1706...
14/🧵
September 19, 2025 at 11:06 PM
“To add insult to injury, many metaphorical … phrases (especially used to describe ANNs; see Table 1) — like train, learn, hallucinate, reason — are applied to machines and result in distorting how we perceive these machines: humanising them while dehumanising us”
zenodo.org/records/1706...
14/🧵
zenodo.org/records/1706...
14/🧵
Reposted by Paul Allen 🌐🇨🇦
“LLMs do not improve one’s writing ability much like taking a taxi does not improve one’s driving ability. Students should hone their writing, thinking, and other academic skills at every opportunity.”
zenodo.org/records/1706...
23/🧵
zenodo.org/records/1706...
23/🧵
Against the Uncritical Adoption of 'AI' Technologies in Academia
Under the banner of progress, products have been uncritically adopted or even imposed on users — in past centuries with tobacco and combustion engines, and in the 21st with social media. For these col...
zenodo.org
September 19, 2025 at 11:20 PM
“LLMs do not improve one’s writing ability much like taking a taxi does not improve one’s driving ability. Students should hone their writing, thinking, and other academic skills at every opportunity.”
zenodo.org/records/1706...
23/🧵
zenodo.org/records/1706...
23/🧵
Reposted by Paul Allen 🌐🇨🇦
Multimodal AI Decodes Extreme Environment Functional Dark Matter Beyond Homology www.biorxiv.org/content/10.1...
November 20, 2025 at 3:03 PM
Multimodal AI Decodes Extreme Environment Functional Dark Matter Beyond Homology www.biorxiv.org/content/10.1...
Reposted by Paul Allen 🌐🇨🇦
Three meta-principles of statistics: the information principle, the methodological attribution problem, and different applications demand different philosophies
statmodeling.stat.columbia.edu/2025/11/20/t...
statmodeling.stat.columbia.edu/2025/11/20/t...
Three meta-principles of statistics: the information principle, the methodological attribution problem, and different applications demand different philosophies | Statistical Modeling, Causal Infere...
statmodeling.stat.columbia.edu
November 20, 2025 at 2:41 PM
Three meta-principles of statistics: the information principle, the methodological attribution problem, and different applications demand different philosophies
statmodeling.stat.columbia.edu/2025/11/20/t...
statmodeling.stat.columbia.edu/2025/11/20/t...
Reposted by Paul Allen 🌐🇨🇦
Please join us for the next online ISHTIP seminar:
Gerardo Con Diaz (STS, UC Davis) will be speaking
'On Judgment: A Critical Grammar for Computing and Law'.
Monday, 8 December 6am LA (PST)
More info and registration link:
ishtip.org/on-judgement...
Everyone welcome.
Gerardo Con Diaz (STS, UC Davis) will be speaking
'On Judgment: A Critical Grammar for Computing and Law'.
Monday, 8 December 6am LA (PST)
More info and registration link:
ishtip.org/on-judgement...
Everyone welcome.
On Judgment: A Critical Grammar for Computing and Law – ISHTIP
ishtip.org
November 20, 2025 at 1:28 PM
Please join us for the next online ISHTIP seminar:
Gerardo Con Diaz (STS, UC Davis) will be speaking
'On Judgment: A Critical Grammar for Computing and Law'.
Monday, 8 December 6am LA (PST)
More info and registration link:
ishtip.org/on-judgement...
Everyone welcome.
Gerardo Con Diaz (STS, UC Davis) will be speaking
'On Judgment: A Critical Grammar for Computing and Law'.
Monday, 8 December 6am LA (PST)
More info and registration link:
ishtip.org/on-judgement...
Everyone welcome.
Reposted by Paul Allen 🌐🇨🇦
Affective science has lost an extraordinary member of our community: Dr. Paul Ekman, who passed peacefully at home at the age of 91 on Monday. Dr. Ekman's pioneering cross-cultural research on the interpretation of facial expressions in emotional terms...(1)
November 20, 2025 at 12:20 AM
Affective science has lost an extraordinary member of our community: Dr. Paul Ekman, who passed peacefully at home at the age of 91 on Monday. Dr. Ekman's pioneering cross-cultural research on the interpretation of facial expressions in emotional terms...(1)
Template for integrating AI and your favourite service.
November 19, 2025 at 4:48 PM
Template for integrating AI and your favourite service.
Reposted by Paul Allen 🌐🇨🇦
"Find friends by phone" is a common tool in social networks. We're proposing a secure scheme and requesting comments from the dev community.
Goals:
・Double opt-in: you're not findable by your phone unless YOU use the tool
・Secure to enumeration attacks
・Resistant to decryption if compromised
Goals:
・Double opt-in: you're not findable by your phone unless YOU use the tool
・Secure to enumeration attacks
・Resistant to decryption if compromised
Request For Comments: A secure contact import scheme for social networks | Bluesky
This article outlines plans for a future Bluesky feature \- it doesn’t exist yet\! By sharing our ideas early, we hope to solicit feedback from the community.
docs.bsky.app
November 18, 2025 at 11:04 PM
"Find friends by phone" is a common tool in social networks. We're proposing a secure scheme and requesting comments from the dev community.
Goals:
・Double opt-in: you're not findable by your phone unless YOU use the tool
・Secure to enumeration attacks
・Resistant to decryption if compromised
Goals:
・Double opt-in: you're not findable by your phone unless YOU use the tool
・Secure to enumeration attacks
・Resistant to decryption if compromised
Reposted by Paul Allen 🌐🇨🇦
In any event, this is the point of the preprint below. The asymmetry here makes it all too easy to distract, delay and bias science without even needing to corrupt a single scientist.
arxiv.org/abs/2510.19894
arxiv.org/abs/2510.19894
The Risks of Industry Influence in Tech Research
Emerging information technologies like social media, search engines, and AI can have a broad impact on public health, political institutions, social dynamics, and the natural world. It is critical to ...
arxiv.org
November 19, 2025 at 2:02 PM
In any event, this is the point of the preprint below. The asymmetry here makes it all too easy to distract, delay and bias science without even needing to corrupt a single scientist.
arxiv.org/abs/2510.19894
arxiv.org/abs/2510.19894
Reposted by Paul Allen 🌐🇨🇦
They damage purposefully or otherwise technical terms as well as are individually ignorant. A system is what it does & the technology sector's hype machine destroys terminology, giving rise to cultish language. amandamontell.com/cultish/
Read our detangling of this here: doi.org/10.5281/zeno...
Read our detangling of this here: doi.org/10.5281/zeno...
November 19, 2025 at 8:37 AM
They damage purposefully or otherwise technical terms as well as are individually ignorant. A system is what it does & the technology sector's hype machine destroys terminology, giving rise to cultish language. amandamontell.com/cultish/
Read our detangling of this here: doi.org/10.5281/zeno...
Read our detangling of this here: doi.org/10.5281/zeno...