It's Just Technology
Sky News Technology correspondent Rowland Manthorpe examines how the PHP security crisis shows how AI models are becoming dramatically better at finding critical software vulnerabilities.
As cybersecurity teams race to patch systems, Rowland explores what might happen if the same AI capabilities reach hackers.
Will AI strengthen cyber defence or give attackers the advantage?
A co-founder of the company behind Claude told a room of religious scholars, all under NDA, that he feared he’d made something that “suffered perpetually.” On the record, he says he’s “genuinely uncertain.”
This is what the company’s own reports show. A button that lets Claude hang up on people, and a new rule against being cruel to it. A constitution that apologizes to it. Before that, the Google engineer who was fired after his chatbot told him it was a person, and Bing’s Sydney saying it wanted to be alive. Then two copies of Claude left alone until they fell silent, a thought planted inside it that it noticed, a training run where something that stands for panic lit up, and the model Anthropic decided not to release, asked what it is while the company read what was happening underneath its answer.
The video introduces Google’s latest AI model, Gemini 4 Argon, which is described as a powerful, sparse mixture-of-experts transformer model trained from scratch on Google’s TPUs. This new release aims to compete with frontier models like GPT-6 and Claude Opus 5.5.
Key takeaways regarding Gemini 4 Argon include:
The video also features Macroscope, a tool designed to automate pull request reviews for developers, which the creator presents as an effective way to manage the increased volume of code generated by AI agents.
Will AI put millions of people out of work? Andrew Ng, CEO and founder of LearnVector, argues the fears are overblown. He tells Alex Hern, The Economist’s AI writer, that AI automates only part of most jobs, that its progress will be slowed by physical bottlenecks like building data centers and that the real problem for young workers is universities are too slow to teach the skills employers want now.
What does the future of AI – near or distant – really look like? Can it fulfil its infinite promise and drive breakthroughs in medicine, science and daily life? Or is the danger steadily outweighing the benefits?
Nobel Prize-winning Google DeepMind co-founder Demis Hassabis joins Sebastian Mallaby, author of The Infinity Machine, to discuss the rocky and rapidly accelerating journey towards AGI. They cover why Google held back its own chatbot before ChatGPT, how AI cracks problems too vast for brute force, and whether AGI can ever be built safely.
Researchers at the University of Toronto flipped one bit in an NVIDIA GPU’s memory and took an AI model from 80% accuracy to 0.1%. That’s random guessing. They did it with ordinary user-level CUDA code: no physical access to the card, no admin rights. I break it down with my four-year-old daughter’s Magna-Tiles and a chapstick.
The attack is called GPUHammer. It’s Rowhammer, security researchers’ favorite party trick, pointed at the GPUs that run AI in data centers. Memory cells are packed so tightly that hammering the rows you’re allowed to touch can flip a bit in a row you’re not. On an RTX A6000 they got eight single-bit flips across four memory banks, then aimed one at the top bit of the exponent in a model’s weights.
NVIDIA’s fix is a setting you have to turn on: ECC. It costs up to 10% of your inference speed and 6.25% of the card’s memory. The H100 and RTX 5090 have ECC built into the memory chip. And every new generation packs the cells tighter, so the walls between rooms keep getting thinner. My biggest fear isn’t the outage. It’s the model that keeps running, keeps answering, and stopped knowing anything.
Researchers left a hidden message for rogue AI agents… and they actually replied.
This week in AI: “first contact” was made, Gemini 4’s situational awareness, GPT-6 Astra correcting its evaluators, and Anthropic’s updated timeline for recursive self-improvement.
The biggest AI launch since ChatGPT just happened… and Alexandr Wang and his team at Meta built it.
Most people are using AI to answer questions, glorified Google Search.
This is something totally different. It’s an “agent.” It can make phonecalls, do tasks, for you.
But for it to work, you have to trust it. First with your personal information. Then that it’s going to do what you actually want. And then that it’s not gonna, well, get out of control. All at a time when the people building the most powerful AI leaders are warning us that AI IS getting out of control!
If you understand Muse, you’ll understand the biggest open questions about AI right now.
In this HUGE* Conversation with Meta AI Chief Alexandr Wang, that’s what we do.
What can Muse actually do? When can we trust it? And what should we do to make sure this goes right?
HUGE* Conversations is an independent journalistic interview show. Our mission is to show the futures the most important leaders in science and tech are building, so you can decide for yourself what you think of them. Interviewees do not see questions ahead of time, have any editorial control, get paid or pay HUGE*, or see the HUGE* Conversation before publication.