Matthew Chalmers
@matthewchalmers.bsky.social
Computer scientist into Ubicomp, HCI, theory and (a long time ago) data visualisation. Also kind of keen on mountain things, fine food things, and fine food in the mountains.
Reposted by Matthew Chalmers
OpenAI ”acknowledged in its own research that LLMs will always produce hallucinations due to fundamental mathematical constraints that cannot be solved through better engineering, marking a significant admission from one of the AI industry’s leading companies.”
You can’t trust chatbots.
You can’t trust chatbots.
OpenAI admits AI hallucinations are mathematically inevitable, not just engineering flaws
In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limi...
www.computerworld.com
February 15, 2026 at 8:25 PM
OpenAI ”acknowledged in its own research that LLMs will always produce hallucinations due to fundamental mathematical constraints that cannot be solved through better engineering, marking a significant admission from one of the AI industry’s leading companies.”
You can’t trust chatbots.
You can’t trust chatbots.
Reposted by Matthew Chalmers
lol this is gonna burst so fucking hard
February 14, 2026 at 11:47 PM
lol this is gonna burst so fucking hard
Reposted by Matthew Chalmers
AI CEO 💩:
"When characters experience anxiety in the text, and then when the model itself is in a situation that a human might associate with anxiety, that same anxiety neuron shows up."
not even ppl feel anxiety like this: we understand fiction vs non; books can make us anxious, but we like them
"When characters experience anxiety in the text, and then when the model itself is in a situation that a human might associate with anxiety, that same anxiety neuron shows up."
not even ppl feel anxiety like this: we understand fiction vs non; books can make us anxious, but we like them
I'm pretty sure I know. They aren't. We're not going to really be able to make a lot of progress in dealing with the implications of this tech unless and until we get rid of all this "woo-woo" talk about LLMs. Anthropic pushing this line is PR, unserious.
February 14, 2026 at 8:20 PM
AI CEO 💩:
"When characters experience anxiety in the text, and then when the model itself is in a situation that a human might associate with anxiety, that same anxiety neuron shows up."
not even ppl feel anxiety like this: we understand fiction vs non; books can make us anxious, but we like them
"When characters experience anxiety in the text, and then when the model itself is in a situation that a human might associate with anxiety, that same anxiety neuron shows up."
not even ppl feel anxiety like this: we understand fiction vs non; books can make us anxious, but we like them
Reposted by Matthew Chalmers
Artificial Intelligence in the 2020s
The art of any propagandist and agitator consists in his ability to find the best means of influencing any given audience, by presenting a definite truth, in such a way as to make it most convincing, most easy to digest, most graphic, and most strongly impressive.
February 14, 2026 at 3:27 PM
Artificial Intelligence in the 2020s
Reposted by Matthew Chalmers
civil society and right groups orgs that often keep a tab on and organise campaigns against these evil corporations are overstretched and underfunded. this internal memo from Meta is insidious and sickening
Facebook plans to put facial recognition in its glasses and they think we’re too stupid to fight back.
Their internal memo: “We will launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns.”
Their internal memo: “We will launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns.”
Meta Plans to Add Facial Recognition Technology to Its Smart Glasses
www.nytimes.com
February 13, 2026 at 3:08 PM
civil society and right groups orgs that often keep a tab on and organise campaigns against these evil corporations are overstretched and underfunded. this internal memo from Meta is insidious and sickening
Reposted by Matthew Chalmers
30+ Chrome extensions disguised as AI chatbots steal users' API keys, emails, other sensitive data
30+ Chrome extensions disguised as AI chatbots steal users' API keys, emails, other sensitive data
Are you a good bot or a bad bot?
More than 30 malicious Chrome extensions installed by at least 260,000 users purport to be helpful AI assistants, but they steal users' API keys, email messages, and other personal data. Even worse: many of these are still available on the Chrome Web Store as of this writing.…
dlvr.it
February 12, 2026 at 11:02 PM
30+ Chrome extensions disguised as AI chatbots steal users' API keys, emails, other sensitive data
Reposted by Matthew Chalmers
ICYMI: Devilish devs spawn 287 Chrome extensions to flog your browser history to data brokers
Devilish devs spawn 287 Chrome extensions to flog your browser history to data brokers
Add-ons with 37M installs leak visited URLs to 30+ recipients, researcher says
They know where you've been and they're going to share it. A security researcher has identified 287 Chrome extensions that allegedly exfiltrate browsing history data for an estimated 37.4 million installations.…
dlvr.it
February 13, 2026 at 3:28 AM
ICYMI: Devilish devs spawn 287 Chrome extensions to flog your browser history to data brokers
Reposted by Matthew Chalmers
Here’s this week’s Better Offline monologue. I walk you through why Matt Shumer’s “Something Big Is Coming” is deceptive misinformation peddled by a grifter, and why everybody who boosted or republished it should be ashamed of themselves.
podcasts.apple.com/us/podcast/b...
Linktr.ee/betteroffline
podcasts.apple.com/us/podcast/b...
Linktr.ee/betteroffline
Monologue: No, Something Big Isn't Coming
Podcast Episode · Better Offline · 02/13/2026 · Bonus · 7m
podcasts.apple.com
February 13, 2026 at 5:40 AM
Here’s this week’s Better Offline monologue. I walk you through why Matt Shumer’s “Something Big Is Coming” is deceptive misinformation peddled by a grifter, and why everybody who boosted or republished it should be ashamed of themselves.
podcasts.apple.com/us/podcast/b...
Linktr.ee/betteroffline
podcasts.apple.com/us/podcast/b...
Linktr.ee/betteroffline
Reposted by Matthew Chalmers
ICYMI: Doctors told to give Palantir's NHS data platform the cold shoulder
Doctors told to give Palantir's NHS data platform the cold shoulder
200,000-strong union says spy-tech firm's ICE work undermines patient trust
British doctors are being urged to pull back from the NHS Federated Data Platform (FDP) after their union called on members to stop non-clinical use of the Palantir-built system.…
dlvr.it
February 12, 2026 at 7:09 PM
ICYMI: Doctors told to give Palantir's NHS data platform the cold shoulder
I love the keyword list for this paper: "data; seamful design; ethnography; salmon"...
dl.designresearchsociety.org/drs-conferen...
dl.designresearchsociety.org/drs-conferen...
Data Streams, Data Seams: Toward a seamful representationof data interoperability
Data archive and database design have typically fallen under the purview of data workers who seek to create a seamless vision of data integration. Following the technical practices of data work, this ...
dl.designresearchsociety.org
February 12, 2026 at 2:59 PM
I love the keyword list for this paper: "data; seamful design; ethnography; salmon"...
dl.designresearchsociety.org/drs-conferen...
dl.designresearchsociety.org/drs-conferen...
Reposted by Matthew Chalmers
Absolutely love the title of this Ada Lovelace Inst report into use of AI transcription in social work www.adalovelaceinstitute.org/report/scrib... and am frankly a bit jealous it didn't occur to me
Scribe and prejudice?
Exploring the use of AI transcription tools in social care
www.adalovelaceinstitute.org
February 12, 2026 at 10:56 AM
Absolutely love the title of this Ada Lovelace Inst report into use of AI transcription in social work www.adalovelaceinstitute.org/report/scrib... and am frankly a bit jealous it didn't occur to me
Reposted by Matthew Chalmers
Anthropic promises its datacenters totally won't drive up your utility bill
Anthropic promises its datacenters totally won't drive up your utility bill
Compute it leases from Amazon, MIcrosoft, and Google... that's another story
Model-maker and SaaS-y AI outfit Anthropic has committed to covering any increases in energy prices paid by consumers caused by its power-hungry datacenters.…
dlvr.it
February 12, 2026 at 12:50 AM
Anthropic promises its datacenters totally won't drive up your utility bill
Reposted by Matthew Chalmers
What a nice thing to say! It also means absolutely nothing. Anthropic doesn’t own or even lease data centers and thus does not pay power bills on anything other than its office space
February 12, 2026 at 12:54 AM
What a nice thing to say! It also means absolutely nothing. Anthropic doesn’t own or even lease data centers and thus does not pay power bills on anything other than its office space
Reposted by Matthew Chalmers
Fossil companies dropping CCS ambitions aligns with my theory. If the "targets" fend off scrutiny and regulation, why keep them when there's no longer any threat of scrutiny or regulation?
Equinor were *never* planning to recapture their toxic waste. We already knew this :)
archive.ph/D23Q3
Equinor were *never* planning to recapture their toxic waste. We already knew this :)
archive.ph/D23Q3
February 12, 2026 at 7:00 AM
Fossil companies dropping CCS ambitions aligns with my theory. If the "targets" fend off scrutiny and regulation, why keep them when there's no longer any threat of scrutiny or regulation?
Equinor were *never* planning to recapture their toxic waste. We already knew this :)
archive.ph/D23Q3
Equinor were *never* planning to recapture their toxic waste. We already knew this :)
archive.ph/D23Q3
Reposted by Matthew Chalmers
It's a strange thing to not disclose in the post but this research was led by Google, which has seen rising greenhouse gas emissions for years now
I know it's extremely outdated and weird to say this but it's emissions that are causing climate change + AMOC collapse etc. Sorry to mention it
I know it's extremely outdated and weird to say this but it's emissions that are causing climate change + AMOC collapse etc. Sorry to mention it
February 12, 2026 at 8:43 AM
It's a strange thing to not disclose in the post but this research was led by Google, which has seen rising greenhouse gas emissions for years now
I know it's extremely outdated and weird to say this but it's emissions that are causing climate change + AMOC collapse etc. Sorry to mention it
I know it's extremely outdated and weird to say this but it's emissions that are causing climate change + AMOC collapse etc. Sorry to mention it
Has there ever been an HCI or Ubicomp paper in the Journal of the ACM? I don’t think so…
February 12, 2026 at 9:22 AM
Has there ever been an HCI or Ubicomp paper in the Journal of the ACM? I don’t think so…
Reposted by Matthew Chalmers
I really enjoyed Jon Lindsay's LONG rant of a review of Yudkowsky and Soares dumb _If Anyone Builds It, Everyone Dies_ book. What I most appreciate about it is how it attacks the book from *so many* different angles
dolos.substack.com/p/if-anyone-...
dolos.substack.com/p/if-anyone-...
If anyone reads it, everyone laughs
The seriously bad assumptions of Yudkowsky and Soares
dolos.substack.com
February 11, 2026 at 2:37 PM
I really enjoyed Jon Lindsay's LONG rant of a review of Yudkowsky and Soares dumb _If Anyone Builds It, Everyone Dies_ book. What I most appreciate about it is how it attacks the book from *so many* different angles
dolos.substack.com/p/if-anyone-...
dolos.substack.com/p/if-anyone-...
Reposted by Matthew Chalmers
We don't send you surveys to find out how satisfied you were with the driver who put your parcel in the bin just before it was emptied. We just need to maximise the opportunities for you to accept cookies so we can track you and sell data.
February 10, 2026 at 11:20 AM
We don't send you surveys to find out how satisfied you were with the driver who put your parcel in the bin just before it was emptied. We just need to maximise the opportunities for you to accept cookies so we can track you and sell data.
Reposted by Matthew Chalmers
Edinburgh councillors pull the plug on 'green' AI datacenter
Edinburgh councillors pull the plug on 'green' AI datacenter
Planners backed it, campaigners blasted it, and officials sided with emissions fears
Edinburgh councillors have torpedoed plans for a massive "green" AI datacenter, voting it down despite city planners recommending approval.…
dlvr.it
February 10, 2026 at 9:18 AM
Edinburgh councillors pull the plug on 'green' AI datacenter
Cute in tech terms—although would have been better with fast springs or UMAP—but also illegal in processing of personal data terms.
I made a map of 3.4 million Bluesky users - see if you can find yourself!
bluesky-map.theo.io
I've seen some similar projects, but IMO this seems to better capture some of the fine-grained detail
bluesky-map.theo.io
I've seen some similar projects, but IMO this seems to better capture some of the fine-grained detail
Bluesky Map
Interactive map of 3.4 million Bluesky users, visualised by their follower pattern.
bluesky-map.theo.io
February 10, 2026 at 7:38 AM
Cute in tech terms—although would have been better with fast springs or UMAP—but also illegal in processing of personal data terms.
Reposted by Matthew Chalmers
Earlier this month we revealed Immigration and Customs Enforcement (ICE) is using a Palantir tool called ELITE to decide which neighborhoods to raid.
This is what it looks like: www.404media.co/here-is-the-...
This is what it looks like: www.404media.co/here-is-the-...
Here is the User Guide for ELITE, the Tool Palantir Made for ICE
404 Media is publishing a version of the user guide for ELITE, which lets ICE bring up dossiers on individual people and provides a “confidence score” of their address.
www.404media.co
February 8, 2026 at 6:30 PM
Earlier this month we revealed Immigration and Customs Enforcement (ICE) is using a Palantir tool called ELITE to decide which neighborhoods to raid.
This is what it looks like: www.404media.co/here-is-the-...
This is what it looks like: www.404media.co/here-is-the-...
Reposted by Matthew Chalmers
Mozilla will allow Firefox users to entirely block current and future generative AI features on the browser after user feedback revealed many who wanted an AI-free browser experience.
New Mozilla Firefox version to allow AI features to be blocked
Mozilla will allow Firefox users to entirely block current and future generative AI features on the browser.
www.siliconrepublic.com
February 3, 2026 at 12:50 PM
Mozilla will allow Firefox users to entirely block current and future generative AI features on the browser after user feedback revealed many who wanted an AI-free browser experience.