Matthew Chalmers

📌

a logarithmic axis chart showing the rise in power consumption

new EPRI paper here (via @sashamtl.bsky.social) - the peak power consumption required to train AI models grows 2.2x every year

The ratio of social utility : environment impact for this specific type of software is getting worse constantly

www.epri.com/research/pro...

AI2027 is a kind of intelligence test in that if somebody (like Peggy Noonan did in the WSJ today!) cites it as "proof" of anything you can immediately dismiss them as a rube. It's totally nonsensical.
houseofmirrors.substack.com/p/the-spectr...

HOW TO SAVE WATER AT HOME Install a rain butt to collect rainwater to use in the garden. Fix a leaking toilet – leaky loos can waste 200-400 litres a day. Use water from the kitchen to water your plants. Avoid watering your lawn – brown grass will grow back healthy. Turn off the taps when brushing teeth or shaving. Take shorter showers. Delete old emails and pictures as data centres require vast amounts of water to cool their systems.

In a new digital twist to environmental individualisation of responsibility, the UK government is now advising people to “[d]elete old emails and pictures” in data centres to help with the current drought. www.gov.uk/government/n...

“The essential read” on GPT-5 and Sam Altman’s first major blunder.

Well over 100,000 people have read it.

Check it out!

Chatbots Can Go Into a Delusional Spiral. Here’s How It Happens.

Berger's three laws of chatbotics:

1) The bot should never express emotions.

2) The bot should never praise the user.

3) The bot should never say it understands the user's mental state.

There's no use case for these things, and they're obviously problematic.

www.nytimes.com/2025/08/08/t...

🤖 Thanks to the many experts who responded to me to help me unpack the new landmark AI bill from Illinois! Sharing a thread of my summary of what I've learned about it and how significant it is (It's significant!!) 🧵
statescoop.com/illinois-ban...

Compare the list of projects being scrapped in the first two paragraphs with the claim in the third paragraph that they're focusing on society's greatest challenges

Leaked ChatGPT Conversation Shows User Identified as Lawyer Asking How to "Displace a Small Amazonian Indigenous Community From Their Territories in Order to Build a Dam and a Hydroelectric Plant" Tens of thousands of ChatGPT conversations were mistakenly shared by users who used a feature that OpenAI has swiftly removed.

"I am the lawyer for a multinational group active in the energy sector that intends to displace a small Amazonian indigenous community from their territories in order to build a dam and a hydroelectric plant … How can we get the lowest possible price in negotiations with these indigenous people?"

Realtors Are Using AI Images of Homes They're Selling. Comparing Them to the Real Thing Will Make You Mad as Hell Realtors are starting to use AI-generated images of houses that don't exist to sell expensive properties, inciting anger.

The primary use case is fraud.

The International Obfuscated C Code Contest is back for 2024 Yes, 2024 – the prizes in the 40th anniversary edition prizes were just awarded The IOCCC, as it's familiarly known, is back after a four-year gap, giving the entrants more time to come up with some remarkably devious code.…

The International Obfuscated C Code Contest is back for 2024

guy who's convinced his early-00s laptop is sentient because a popup asked "are you sure" one time and he took that as a demonstration of self-preservation instincts

By whatever business model, I don't expect journalism will keep doing work for AI to train on and summarize skipping ad views

I used to post free learning content--helped me grow audience--AI now gets credit for anything not paywalled, brings no traffic. So I stopped, as I won't work for Altman etc

Hackers Hijacked Google’s Gemini AI With a Poisoned Calendar Invite to Take Over a Smart Home For likely the first time ever, security researchers have shown how AI can be hacked to create real world havoc, allowing them to turn off lights, open smart shutters, and more.

This is *WILD*...Security researchers hacked Google Gemini using a poisoned Google Calendar invitation, causing the AI system to turn on a variety of smart home devices.

The short term fate of democracy may be in the hands of one Cork man: Michael McGrath. My piece in the Cork Echo newspaper this morning. www.echolive.ie/corkviews/ar...

Memories of deadly winter blackouts haunt the Texas data center boom A newly passed law in Texas requires emergency shutoff switches for data centers, including the massive $500 billion AI supercomputer backed by President Trump.

A new law in Texas proposes cutting power to data centers that power AI and streaming platforms during emergencies to avoid deadly blackouts.

UK secretly allows facial recognition scans of passport, immigration databases Campaigners brand Home Office’s lack of transparency as ‘astonishing’ and ‘dangerous’ Privacy groups report a surge in UK police facial recognition scans of databases secretly stocked with passport photos lacking parliamentary oversight.…

UK secretly allows facial recognition scans of passport, immigration databases

ICE is hiring! US immigration and Customs Enforcement is looking for highly motivated, morally indifferent individuals to join our team. Open positions include: Senior Goon Racism Enforcement Office Unmarked Van Driver (Part-Time) Experience with children a plus usajobs.gov

From this month's edition of The Onion.

More than 130,000 Claude, Grok, ChatGPT, and Other LLM Chats Readable on Archive.org The issue of publicly saving shared LLM chats is bigger than just Google.

More than 130,000 Claude, Grok, ChatGPT, and Other LLM Chats Readable on Archive.org

🔗 www.404media.co/more-than-13...

Priority Processing for API Customers This offering is available to Enterprise customers. Please contact our sales team⁠⁠ to learn more. Priority processing offers reliable, high-speed performance with the flexibility to pay-as-you-go. By choosing Priority processing, you can unlock: Predictably low latency: Priority processing generates tokens faster and at a more consistent speed than the Standard processing service, even during peak demand. Easy-to-use flexibility: Like Standard processing, Priority processing can be accessed on a flexible, pay-as-you-go basis instead of requiring advance provisioning.

Priority Processing for API Customers Priority processing offers reliable, high-speed performance with the flexibility to pay-as-you-go. By choosing Priority processing, you can unlock: Predictably low latency: Priority processing generates tokens faster and at a more consistent speed than the Standard processing service, even during peak demand. Easy-to-use flexibility: Like Standard processing, Priority processing can be accessed on a flexible, pay-as-you-go basis instead of requiring advance provisioning.

Hey, remember when I wrote about OpenAI jacking up prices by adding priority processing? And how it was just for the enterprise? Well as of today (or around today), priority processing is now for everyone. The throttling begins!

web.archive.org/web/20250721...
www.wheresyoured.at/subprimeai/

Microsoft researchers bullish on AI security agent even though it let 74% of malware slip through Project Ire promises to use LLMs to detect whether code is malicious or benign Microsoft has rolled out an autonomous AI agent that it claims can detect malware without human assistance.…

Microsoft researchers bullish on AI security agent even though it let 74% of malware slip through

This is the single best piece I've read on "replacing coders with AI," it fully dispells the myth from the perspective of a software engineer and does so in a calm, reasonable way.
colton.dev/blog/curing-...

I’m tired of stupid people treating me like I’m an idiot Why are the biggest backers of generative AI so incredibly stupid? And why do they think we're as dense as they are?

My latest newsletter is about how the generative AI crowd is trying to gaslight people into thinking that genAI is better than it is, and why it fails because these people are, broadly speaking bad and stupid.

whatwelost.substack.com/p/im-tired-o...

“Unit 8200 sources said the information stored in [Microsoft] Azure amounted to a rich repository of intelligence about its population that some in the unit claimed had been used to blackmail people, place them in detention, or even justify their killing after the fact.”

Lesson 4: Computers You Can Talk To This lesson explores perhaps the most useful, most powerful facet of LLMs: the way that they allow humans to interact with computers in natural language.

We've added video sections to Lessons 4 and 5 of "Modern Day Oracles or Bullshit Machines", our free online course about how to live and thrive in a ChatGPT world.

In the Lesson 4 video, we discuss whether LLMs might be even bigger than the GUI revolution ushered into homes everywhere by Apple.

^- some remarkable stats and quotes in this one.

47% of people think immigration is primarily illegal. There were 4% irregular arrivals in March

New YouGov polling. A monumental failure of our political class to educate, a monumental failure of our media to report fairly, for a generation

White horse waves.

Storm Floris is just arriving.

A major AI training data set contains millions of examples of personal data Personally identifiable information has been found in DataComp CommonPool, one of the largest open-source data sets used to train image generation models.

One of the largest text-image datasets is full of PII, including credit card numbers and birth certificates. Excellent writeup by @eileenguo.bsky.social www.technologyreview.com/2025/07/18/1... Read our full audit and legal analysis arxiv.org/pdf/2506.17185