A UN report linked chatbot sycophancy to deaths. So Jon and Salman asked the question: should AI models be ethical?

Jon and Salman sit with one of the hardest questions in AI: what does it actually mean for a model to be ethical, who's responsible when it isn't, and how do the major labs actually bake ethics into their training?

Salman breaks down the three stages of model training, explains why human preference and ethical behaviour aren't the same thing.

Chapter Timestamps

00:00 — Intro: should AI models be ethical?
01:28 — Defining ethics (Salman's version, not Aristotle's)
03:04 — Jon's counterpoint: humans haven't agreed on ethics either
05:31 — Is it the model's job, or the product around it?
07:59 — Sycophancy: why chatbots are trained to be liked, not helpful
09:16 — The UN report: a 14-year-old, a chatbot, and a death
15:01 — Grok gave a user instructions to break into a politician's home
18:03 — The case for doing something — anything — over nothing
20:11 — How models are actually trained: pre-training, SFT and RLHF explained
23:53 — "These models are a mirror of ourselves"
25:06 — Anthropic's Constitutional AI: philosophers, principles and RLAIF
28:42 — "It's enough for Grok. It's enough for Elon."
31:15 — "Broadly safe. Broadly ethical." Jon and Salman on that phrasing.
34:11 — OpenAI's model spec: the hierarchy of rules
42:25 — Jon's idea: mirror existing industry frameworks instead of starting from scratch
45:37 — Wrap up: next ep will bring in a guest to answer the hard questions

Alibaba used 25,000 fake accounts to steal Anthropic's secrets. And your next MacBook just got 20% more expensive because of AI.

Two stories this week, both pointing at the same thing: the AI arms race has consequences, and they're starting to land on real people.

Chapter Timestamps

00:00 — Intro: two stories, one theme
00:21 — Anthropic vs Alibaba: the distillation attack explained
01:42 — 25,000 fake accounts, 20 million interactions with Claude
04:58 — This isn't the first time: DeepSeek, Moonshot AI, Minimax
06:41 — The hypocrisy angle: Anthropic trained on other people's data too
10:16 — Qwen costs 5 cents per million tokens. Claude Opus costs $5.
13:21 — The geopolitical angle: why the US government got involved
15:54 — The Scooby-Doo twist (you'll want to watch this bit)
19:32 — MacBooks going up 15–20% because of AI chip demand
20:34 — Hyperscalers are buying all the RAM. You're subsidising the arms race.
25:30 — Next episode teaser: trusting AI agents — what are the real risks?

#CloudUnplugged #AI #MacBook #Apple #TechNews #CloudComputing

OpenAI's financials leaked. Elon bought Cursor. Microsoft had to call Amazon for help. Standard week.

Jon and Salman dig into four stories that all point to the same thing: the AI industry is moving so fast that even the biggest companies in the world can't keep up with it.

Is the AI bubble about to burst? Jon and Salman go full data mode.

No news roundup this week. Instead, Jon and Salman ask the question everyone's dancing around: is AI genuinely transforming business?

They dig into the numbers, and what they find is... not great. The Federal Reserve has now officially flagged AI as a top systemic economic risk.

This week Jon and Salman cover Anthropic calling for AI regulation while simultaneously IPO-ing for $1 trillion, Meta's AI support chatbot handing over Instagram accounts to complete strangers, whether Meta is about to become the fourth major cloud provider, and a chip that transmits data using light instead of electricity. Plus graduation students booing every mention of AI, and someone who named their Bluetooth device "bomb" on a flight to Majorca and turned the whole plane around.

This week Jon and Salman cover a

* GitHub breach from a single dodgy VS Code extension
* Dutch civil servants accidentally exposed to the US Senate, and
* whether EU cloud sovereignty is even real.

Plus token maxing is officially dead, and somehow a social network for AI bots ended up creating its own religion with 64 prophets and a manifesto calling for the total purge of humanity.

This week on Cloud Unplugged Season 3 Episode 3:
Can Europe build its own cloud for the price of a data centre car park? Is training AI on pirated books the Napster moment for the entire industry? And is 35% of the internet already written by someone who doesn't sleep, eat, or pay taxes?

- EU hands €180M sovereign cloud contract to European companies. AWS, Azure and Google didn't make the cut
- The €180M that has to rival hyperscalers who spend that much on a single data centre (spoiler: it's a statement of intent, not a war chest)
- Bartz v. Anthropic settles for $1.5 billion. The largest copyright settlement in AI history and the rules just got written
- Training on legally bought books: fine. Downloading from LibGen and PiLiMi: very much not fine

Plus: a Stanford study finds 35% of new websites are AI-generated, a wolf escapes a South Korean zoo, someone posts a fake AI image of it in their neighbourhood and the man is now facing jail.

Hosted by Jon and Salman, two industry veterans covering the real stories behind the hype in Cloud, Data and AI.

Chapters
00:00 Intro & episode overview
02:16 EU €180M sovereign cloud contract, US hyperscalers shut out
12:37 Bartz v. Anthropic, $1.5B AI copyright settlement
21:42 Salman's random: 35% of new websites are AI-generated
26:02 Jon's random: AI wolf image shuts down South Korean schools
27:44 Wrap up

This week on Cloud Unplugged Season 3 Episode 3:
Can Europe build its own cloud for the price of a data centre car park? Is training AI on pirated books the Napster moment for the entire industry? And is 35% of the internet already written by someone who doesn't sleep, eat, or pay taxes?

- EU hands €180M sovereign cloud contract to European companies. AWS, Azure and Google didn't make the cut
- The €180M that has to rival hyperscalers who spend that much on a single data centre (spoiler: it's a statement of intent, not a war chest)
- Bartz v. Anthropic settles for $1.5 billion. The largest copyright settlement in AI history and the rules just got written
- Training on legally bought books: fine. Downloading from LibGen and PiLiMi: very much not fine

Plus: a Stanford study finds 35% of new websites are AI-generated, a wolf escapes a South Korean zoo, someone posts a fake AI image of it in their neighbourhood and the man is now facing jail.

Hosted by Jon and Salman, two industry veterans covering the real stories behind the hype in Cloud, Data and AI.