More powerful models are not always the better choice for engineering systems. Benjamin Nweke highlights the tradeoff between flexibility and consistency in LLM outputs.

Google just announced they are redesigning search around AI. Here's how to get actual search results again.

The Trump administration has made a dizzying array of force posture decisions this month, raising doubts about the United States’ deeper commitment to the NATO alliance, write Daniel Fried and Torrey Taussig. Read more ⤵️

Highly recommended.

Open source models can shield user data from being weaponized. "Personal data embedded within existing commercial foundation models thus positions AI as a link between commercial personal data and automated weapons’ target lists and surveillance capabilities." arxiv.org/abs/2410.14831

Not sure how long it has been around but it’s a really great ML resource huggingface.co/learn by @hf.co

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 story, in the end, is ours. snyder.substack.com/p/the-man-in...

Apple removed 1,213 apps from the App Store in 2025 at the request of the Russian government. meduza.io/en/news/2026...

What would make Bellingcat an organisation you’d like to support? We want to hear from you. Help us build the Bellingcat of tomorrow by filling out this quick survey: bellingcat.questionpro.eu/t/AB3vDHmZB3...

For those interested I previously wrote about how Anthropic and others co-opt safety and defense terminology to take over the arbitration of risk determinations, thus subverting democratic processes arxiv.org/abs/2504.15088

I joined Hari Sreenivasan on CNN International and PBS to discuss the use of AI in warfare and the impacts we're already seeing of this fallible technology being used in Iran, and how it ultimately obscures accountability. Full interview can be found at youtube.com/watch?v=w16f...

On @channel4news.bsky.social @heidykhlaaf.bsky.social explains the stakes of Anthropic’s Mythos and its impact on security. We lack the evidence and independent expertise needed to verify the claims around the security tool - especially problematic given the risks posed by LLMs. shorturl.at/kNtU9

"Generative AI has no shortage of ways that it might, be shaped into genuinely useful products, but this shaping needs to actually happen before the hyper-scalers earn the right to continually harass the psyche of billions of people with breathless pronouncements. " calnewport.com/who-asked-fo...

"I think we’ve entered a stage where we need to almost entirely discount any claims made by the AI companies themselves until we can independently verify what’s actually going on." calnewport.com/is-claude-my...

What if a patch or badge could reveal Russian intelligence structures? Watch the session to unpack what’s behind military insignia 🎥 www.disinfo.eu/23-april-dec... Next up 💥 Look what you made me do: How FIMI actors weaponise pop culture 🗓️ 30 April, 14:30 CEST us06web.zoom.us/webinar/regi...

"IFIT tested six leading AI models including ChatGPT, Deepseek, Grok, and others on three real-world prompt scenarios from Syria, Sudan, and Mexico. [...] From a total possible point value of 100/100, the average score across all six models was only 27 points." ifit-transitions.org/publications...

Coding aside, even the best AI systems struggle to be economically viable in the workplace. What happens then?

“As people working on AI, we know that these systems can centralize power and that they do make mistakes.” – Google Employees

Newspapers shouldn’t cover things Trump says as news. There should be a special rubric called "Trump said this" in the corner of the page or as a separate tab. This would be clarifying. www.theguardian.com/world/2026/a...

"But what is notable about Anthropic’s diligent efforts to proactively prevent public harm from the use of its new AI model is that there was not a single AI accountability organization on the list of experts Anthropic brought together."

🗣️"Rather, than lose the proverbial forest through the trees, researchers can ask more important questions, such as: why is a review needed, and where does it aim to take a field?" #AcWri #PhDchat

Archaeologists at Pompeii used AI for the first time to digitally reconstruct the face of a man killed during the AD 79 eruption of Mount Vesuvius. His remains were found near Porta Stabia, apparently fleeing toward the coast while using a terracotta mortar as a shield from falling volcanic debris.

We have a deal. theonion.com/at-long-last...

Get excited. www.nytimes.com/2026/04/20/b...

"The lawsuit targets the CDC DHS, ICE, IRS, and the Social Security Administration, all of which have failed to produce records in response to Freedom of Information Act requests seeking information about their use of Palantir tools and access to sensitive data."

I’ll be @debalie.bsky.social in Amsterdam tonight talking about AI at war. No shortage of material these days! Tickets here: debalie.nl/programma/ai...

General Purpose Technologies have downstream effects everywhere, good and bad. Those outcomes can be mitigated, or encouraged, with the right sorts of policy choices If the only options are being for or against a GPT, you end up having very few levers to pull to make it work out.

. @heidykhlaaf.bsky.social called for “independent, rigorous evaluations that objectively assess the risk of tools within existing cyber laws. Democratic institutions should determine how these models are released and deployed, not a private corporation with its own incentives." shorturl.at/4dkAm

“I’ve been using ChatGPT to help with cooking for a while now, so I didn’t think too much of it when the ingredients list included a bottle filled with gasoline and a cloth wick,” said the alleged attacker

How did Magyar win? A long running grassroots campaign in rural and small town Hungary; a refusal to be distracted by government propaganda; a focus on economic issues and corruption; a central promise to bring back democracy and the rule of law

We recently released a new online tool at Bellingcat that helps measure probable destruction in Iran and the surrounding gulf states. It is free to access here: www.bellingcat.com/resources/20...

The White House’s AI framework sidelines concerns about algorithmic bias and discrimination, casting efforts to mitigate those harms as ideological interference, argues Genevieve Smith. Treating AI as “neutral” doesn’t eliminate harm; it risks scaling inequality under the guise of objectivity.

For the Tech Policy Press podcast, Justin Hendrix spoke to Katrina Manson, author of the new book Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare. They discussed the book, the war in Iran, the Anthropic-Pentagon dispute, and the future of war in the AI age.

💥New | Before AI agents act for us, we need to know how AI searches for us ✍️ @jannajoceliomena.bsky.social Giulia Tucci & Aanila Kishwar #AIsearch #AgenticAI #Expertise