Matthew Chalmers
LightNews LightNews
banner
matthewchalmers.bsky.social
Matthew Chalmers
@matthewchalmers.bsky.social
190 followers 56 following 160 posts
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.
Posts Replies Media Videos
Reposted by Matthew Chalmers
juniorhoncho.bsky.social
just started having recency bias and i love it. easily the best bias
August 12, 2023 at 6:04 PM
Reposted by Matthew Chalmers
garymarcus.bsky.social
👉Softbank sells entire Nvidia position.

👉Oracle debt downgraded.

👉Meta financing games revealed.

👉OpenAI CEO @sama couldn’t explain how company would meet its $1.4 T obligations.

👉Coreweave drops 20% in a week.

You do the math.
November 11, 2025 at 6:52 PM
Reposted by Matthew Chalmers
ketanjoshi.co
"In an internal memo from September, CEO Sam Altman said that OpenAI’s “audacious long-term goal is to build 250 gigawatts of capacity by 2033.” If Altman achieves this goal, OpenAI will need almost exactly as much electricity as India’s 1.5 billion people"

Great @truthdig.com piece on chips ->
The Ecological Cost of AI Is Much Higher Than You Think - Truthdig
As the demands of AI grow, each generation of microchips requires more energy, minerals and water to produce, driving a ruinous cycle.
www.truthdig.com
November 11, 2025 at 1:47 PM
Reposted by Matthew Chalmers
edzitron.com
Forbes estimates OpenAI is blowing $15m a day on Sora. Sure, why not? I bet OpenAI’s inference costs are absolutely horrifying
www.forbes.com/sites/phoebe...
Here’s How Much Cash OpenAI Is Burning On AI Video App Sora. What It Means
Some back-of-napkin math suggests OpenAI is spending more than a quarter of what it’s making to power the AI slop factory.
www.forbes.com
November 11, 2025 at 3:18 AM
Reposted by Matthew Chalmers
johnnyryan.bsky.social
See letter here bsky.app/profile/kris...
November 10, 2025 at 12:19 PM
Reposted by Matthew Chalmers
mjcrockett.bsky.social
This NeurIPS workshop claims that LLMs "provide an important foundation for exploring human cognition, emotion, and social interaction"

This is flawed logic, as @lmesseri.bsky.social and I argue here:
www.sciencedirect.com/science/arti...
November 9, 2025 at 3:23 PM
Reposted by Matthew Chalmers
kiswanson.medsky.social
Ryan Moulton @moultano Oil is the most Lovecraftian thing that actually exists. You're telling me that there's a black ichor under the earth, made from the ancient dead, whose burning can realize all the dreams of man but only at the price of slowly returning the earth to its primordial state?
November 8, 2025 at 5:44 PM
Reposted by Matthew Chalmers
mattrosoff.bsky.social
Sounds right:

The report found that the heaviest AI users are thought leadership writers (84%), PR/comms professionals (73%), and content marketing writers (73%).
November 9, 2025 at 7:17 PM
Reposted by Matthew Chalmers
ketanjoshi.co
Extremely funny that the abundance-aligned tech industry created the single most effective weapon for development opponents they've ever got their hands on

Going to be a fun time when the anti-wind and anti-solar groups get going on this stuff

www.theguardian.com/politics/202...
AI-powered nimbyism could grind UK planning system to a halt, experts warn
Tools that help people scan applications and find grounds for objection have potential to hit government’s housebuilding plans
www.theguardian.com
November 9, 2025 at 10:10 PM
Reposted by Matthew Chalmers
brendelbored.bsky.social
Tech guys six months ago: haha yes we’re cutting all this WASTEFUL spending by eliminating medical research and USAID

Tech guys now: yes I think taxpayers will be excited to bailout my non consensual pornography machine
November 8, 2025 at 11:02 PM
matthewchalmers.bsky.social
We saw an otter today, basically in the middle of Edinburgh. This was maybe 100m from a busy shopping street. Amazing!
November 8, 2025 at 10:33 PM
Reposted by Matthew Chalmers
ketanjoshi.co
When someone builds a cheap and ubiquitous slop firehose, we're all forced to close our windows and lock our doors

Still waiting for someone to explain how this has been a net benefit for human society

www.404media.co/arxiv-change...
arXiv Changes Rules After Getting Spammed With AI-Generated 'Research' Papers
Cornell University’s arXiv will no longer accept Computer Science reviews and position papers.
www.404media.co
November 8, 2025 at 9:43 PM
Reposted by Matthew Chalmers
ryanmarino.bsky.social
Panera’s moderately caffeinated lemonade was loosely associated with 2 deaths before it was taken off market.

This article alone has 4 examples of ChatGPT encouraging young people to commit suicide, and OpenAI’s own public stats estimate over a million users discuss suicide with ChatGPT each week.
November 7, 2025 at 10:56 PM
Reposted by Matthew Chalmers
ketanjoshi.co
Something I've really noticed among institutional "responses" to generative technologies is that "DO NOT USE THIS TOOL, IT'S COMPLETELY INAPPROPRIATE" is fudamentally erased from any possibility of ever being an option

Often paired with fatalist stuff like "this isn't going away"
View Steve Little’s graphic link Steve LittleSteve Little • 2ndPremium • 2nd AI Program Director, National Genealogical Society; Founder/Owner: AI Genealogy Insights; United Methodist ClergyAI Program Director, National Genealogical Society; Founder/Owner: AI Genealogy Insights; United Methodist Clergy 1d • 1 day ago • Visible to anyone on or off LinkedIn Follow This spring, genealogists discovered a problem: new AI "photo restoration" tools and techniques using GPT-4o and Nano Banana don't actually "restore" historical images—they generate plausible new images that can permanently corrupt our historical record. The image below shows the issue. One is a historical photograph. The other is an AI-generated face based on that photograph—convincing, but completely fabricated. The man on the left is my grandfather; the man on the right is not. By August, when Google's Nano Banana made this technology instantly accessible to millions, unlabeled AI "restorations" began appearing throughout online family trees. Many were being treated as authentic historical photographs. Expert genealogists responded. The Coalition for Responsible AI in Genealogy—led by genealogists who deeply understand evidential standards and the Genealogical Proof Standard—developed practical guidelines to address this challenge. Today they released their statement: "Protecting Trust in Historical Images." Three core recommendations: 1. Always label modified images 2. Always cite the original source 3. Use as illustration, not evidence These aren't restrictive rules—they're practices that let us explore what these technologies offer while protecting the integrity of our shared historical record. As a technical advisor to the Coalition, I'm grateful to the expert genealogists who led this work and to our broader community for taking these challenges seriously. Read the full story of how the genealogy community identified and addressed this challenge: https://lnkd.in/eTgZtTW4 I’ve been privileged to serve as a technical advisor to the Coalition, but the real work here was done by genealogists who deeply understand evidential standards and the Genealogical Proof Standard. This statement reflects their expertise. The Coalition’s three core recommendations are simple and practical: Always Label modified images. Always Cite the original source. Use as Illustration, Not Evidence. For detailed guidance on implementing these recommendations, see the educational articles linked at the bottom of the statement. When we use AI tools—and many of us will—we're accountable for accuracy, transparency, and ethical practice. Our Responsibility These tools aren’t going away. As genealogists, our responsibility is clear: we must be the “humans in the loop.” When we use AI tools—and many of us will—we’re accountable for accuracy, transparency, and ethical practice. The Coalition’s guidelines help us meet that responsibility. They protect the integrity of our shared historical record while allowing us to explore what these technologies might offer.
November 8, 2025 at 10:15 PM
Reposted by Matthew Chalmers
fagstein.bsky.social
Ontario court says media outlets can sue OpenAI over copyright infringements.
www.theglobeandmail.com/business/art...
Media outlets win motion for Ontario court to hear OpenAI lawsuit
Legal action brought by The Globe and other media groups alleges U.S. company violating copyright law by scraping content without consent or payment
www.theglobeandmail.com
November 8, 2025 at 1:10 AM
Reposted by Matthew Chalmers
davidgerard.co.uk
Oxford pretends AI benchmarks are science, not marketing

How could all these benchmarks be fake, it’s a mystery

www.youtube.com/watch?v=KcYZ... - video
pivottoai.libsyn.com/20251106-oxf... - podcast

time: 6 min 16 sec
scientist in lab coat with a clipboard, looking "hmm" at a robot
November 6, 2025 at 3:18 PM
Reposted by Matthew Chalmers
socialmedialab.ca
When deception’s more profitable than honesty, users lose.

Internal documents suggest that Meta earns $3.5B(!) every 6 mo. from scam ads, revealing a deeper systemic problem: the economic incentive to tolerate “higher legal risk” content still outweighs any penalty. www.reuters.com/investigatio...
Meta is earning a fortune on a deluge of fraudulent ads, documents show
Meta projected 10% of its 2024 revenue would come from ads for scams and banned goods, and it internally estimates that its platforms show users 15 billion scam ads a day, company documents show.
www.reuters.com
November 7, 2025 at 4:14 AM
Reposted by Matthew Chalmers
ketanjoshi.co
This is huge news in @cameronwilson.bsky.social's @thesizzle.com.au- Microsoft is being forced by the Aus regulator refund all the ultra-dodgy AI plan pushing it was doing for Office 365

Wild that other regions aren't also using regulatory power to punish Microsoft

thesizzle.com.au/p/google-sur...
Microsoft is refunding its scummy AI plans so that the ACCC doesn’t go so hard on them Microsoft is offering refunds to 3 (!) million Australian customers and apologised about its dodgy AI plan pushing (SmartCompany). In an email sent to customers today, Microsoft says it will refund all the additional cost of its Copilot AI plans if people revert to its non-AI plan by the end of the year. “In hindsight, we could have been clearer about the availability of a non-AI-enabled offering with subscribers, not just to those who opted to cancel their subscription,” the statement said. grovelling email by Microsoft The Sizzle: None of this happens if the ACCC hadn’t launched legal proceedings against Microsoft. Notably, this is just for Australian customers. Everywhere else just remains ripped off. If our consumer protections make us a nanny state, then stick a dummy in my mouth and give me a diaper!!! Greatest country in the world!!!!
November 6, 2025 at 12:02 PM
Reposted by Matthew Chalmers
ketanjoshi.co
"Common Crawl has opened a back door for AI companies to train their models with paywalled articles from major news websites. And the foundation appears to be lying to publishers about this—as well as masking the actual contents of its archives"

I hate these people so much
The Company Quietly Funneling Paywalled Articles to AI Developers
“You shouldn’t have put your content on the internet if you didn’t want it to be on the internet,” Common Crawl’s executive director says.
www.theatlantic.com
November 6, 2025 at 3:40 PM
Reposted by Matthew Chalmers
404media.co
The FBI has subpoenaed the domain registrar of archive.today, demanding information about the owner.
FBI Tries to Unmask Owner of Infamous Archive.is Site
The FBI has subpoenaed the domain registrar of archive.today, demanding information about the owner.
www.404media.co
November 6, 2025 at 3:24 PM
Reposted by Matthew Chalmers
histoftech.bsky.social
“authors & publishers who filed a lawsuit against the Sam Altman-led firm have secured access to internal Slack messages… discussing the mass deletion of a pirated books dataset… A NY district court ordered OpenAI to hand over the communications regarding data deletion”
futurism.com/artificial-i...
OpenAI in Danger After Authors Suing It Gain Access to Its Internal Slack Messages
Authors and publishers, who are suing OpenAI, secured access to internal Slack messages and emails discussing the deletion of pirated books.
futurism.com
November 6, 2025 at 7:57 AM
Reposted by Matthew Chalmers
theregister.com
Microsoft apologizes for not explaining cheaper no-AI M365 plans, and all it took was a government lawsuit
Microsoft apologizes for not explaining cheaper no-AI M365 plans, and all it took was a government lawsuit
Even offers refunds if users sign up for AI they don’t want Microsoft Australia has apologized to users of its M365 suite after regulators accused it of steering them towards pricey bundles that include its Copilot AI service.…
dlvr.it
November 6, 2025 at 12:54 AM
Reposted by Matthew Chalmers
thierryaaron.bsky.social
😢

"Perhaps the most striking findings is that the British public vastly overestimates the cost getting to net zero by almost 14,000%, fueling scepticism & a reluctance to pay. On average, they estimate it will need 28% of GDP by 2050, compared to @thecccuk.bsky.social forecast of just 0.2%"
Are the advocates for net zero losing the fight?
The cost of misunderstanding: How public perception shapes the net zero debate
fgsglobal.com
November 4, 2025 at 10:02 AM
Reposted by Matthew Chalmers
tomerullman.bsky.social
Nature suggests you use their "Manuscript Adviser" bot to get advice before submitting

I uploaded the classic Watson & Crick paper about DNA structure, and the Adviser had this to say about one of the greatest paper endings of the century:
November 3, 2025 at 1:55 PM
Reposted by Matthew Chalmers
rachelcoldicutt.bsky.social
Good bit of analysis here.

'Putting lay-offs down to AI sounds better than saying “we need to keep margins high so we’re sacking some low performers” and politically safer than saying “unpredictable Trump tariff policy means we are hiring less young people.”'
November 3, 2025 at 6:39 AM