AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

Happy Father’s Day, unless you are in a country that doesn’t celebrate it today in which case, um, happy Sunday.

This week Charles has mostly been at his desk tapping away. Hannah has also been at her desk tapping away. This is good from a work point of view but doesn’t, it has to be said, make the most exciting of intros.

Next week should be a bit more exciting: Hannah is travelling to Vienna to give the opening keynote at OWASP Global AppSec, their 25th anniversary. Her talk is an attempt to infuse all the change, the noise and the new risks into something that feels energising and inspiring. A tough gig.

Have a wonderful week.

Hannah & Charles

What’s Charles reading this week?

Back in May we reported that SpaceX had signed a deal to ‘partner’ with struggling AI coding startup Cursor for $10bn, with an option to buy the company later in the year for $60bn. Now SpaceX has announced that it is indeed to acquire the company. The announcement came only a few days after SpaceX went public at a bonkers valuation of about $1.77 trillion.

Meanwhile, in a fairly aggressive move, the US Department of Justice on Monday filed a motion to intervene and dismiss the NAACP’s suit against SpaceX subsidiary xAI. The lawsuit claims that xAI is illegally operating 27 gas turbines without an air permit in Southaven, Mississippi, effectively building a power plant for its Colossus 2 data centre that runs the company's chatbot, Grok. The DoJ argues that Colossus 2 and Grok are exempted as they are “critical to the economy” and the U.S. military. Crowned the world’s first trillionaire last Friday when SpaceX went public, Musk financed Trump’s presidential campaign more than any other donor and is pouring money into midterms (which is of course purely coincidental).

There are some revelations in the court document, not least that the U.S. military used xAI’s Grok to help plan its bombing missions against Iran earlier this year. “The Department [of war] specifically relies on xAI’s “Grok Gov Model,” which offers features “found in no other frontier AI model”. For example, during Operation Epic Fury, Grok (through the Department’s Maven Smart Systems) “enabled U.S. forces to deploy over 2,000 munitions to 2,000 distinct targets within 96 hours . . . , a testament to the greatly increased operational efficiency made possible by the Grok Gov Model.”

Human error and/or outdated intelligence is every bit as likely, but I can’t help but wonder if one of the targets Grok selected with such efficiency was the school in southern Iran which the U.S. attacked killing more than 170 people, most of them children. One way to interpret the Trump administration's ludicrous demands on Anthropic [see Hannah’s section] is that Anthropic was reluctant to allow its models to be used in the same way, and Trump took exception.

Sticking loosely with AI and the environment, The Verge’s senior AI reporter Hayden Field has an extraordinary story about three Amazon software engineers, who are all members of Amazon Employees for Climate Justice, and had testified as private citizens concerned about AI and data centres gobbling up a third of Seattle's electricity. They didn't say they worked for Amazon, but within a week of their testimony (and conveniently one day after the council passed a data centre moratorium), the three were yanked into surprise HR meetings to be told they were under investigation. The company then said it "doesn't tolerate retaliatory behaviour". Apparently with a straight face.

Amazon's defence is that the employees violated a policy against speaking as a company spokesperson without approval. Which would be a slightly more convincing argument if any of them had actually done that. One employee drily noted this culture of fear is basically Amazon's whole management strategy — layoffs, performance plans, forced rankings — so the retaliation wasn't exactly a shocker.

While getting close to Trump has helped Musk’s cause, it may be damaging Palantir. The UK is reviewing Palantir’s contract with the National Health Service. The NHS is sorely in need of modernisation (it only just junked its last fax machines), but if half of your business is selling to governments, then there are pros and cons to the CEO making loud pro-MAGA provocations.

The UK Government has also announced plans to ban children under the age of 16 from accessing some social media platforms, including TikTok and YouTube. Ministers say the aim is to protect children's wellbeing and mental health, and to give parents more support when setting boundaries around screen use. The changes are expected to come into force by spring 2027, after new regulations are introduced.

I hate social media with the passion of a thousand burning suns, but I really think the ban is ill-considered. There are currently no practical details about how the ban would be enforced. VPNs are a thing. And isn’t this ultimately about parenting rather than a role for government?

For my own children social media has been an important source of connection. One of them, who spends a lot of time on YouTube, says that all the scams and NSFW content he’s been exposed to come from YouTube ads, which also show up on child accounts, and which a ban isn’t going to help because…it's a website.

For disabled people, community can be really important when it comes to sharing experiences. For example, many visually impaired kids are the only ones in their school. The Royal Society for Blind Children says that social media has become a place where friendships can be made, and where young people can express themselves and feel less alone. The charity supports better age-appropriate protection, and action against platforms that go against the guidelines. But they fear the impact of an outright ban. They're not the only ones who feel that way. The National Deaf Children's Society has similar concerns, saying they want to see action that keeps children safe from harm, while still allowing deaf young people to stay connected, to express themselves, and to access support.

I found myself nodding along to this comment from Dr Robyn Muir, Senior Lecturer in Media and Communications at the University of Surrey on, ironically, LinkedIn:

"Banning young people from social media is not a fix-all solution.

esearch consistently shows that online safety depends on strong media literacy skills and open conversations with trusted adults. While concerns about social media are understandable, restricting access alone will not remove risk.

nstead, we should focus on helping young people develop the confidence and critical thinking skills needed to navigate digital spaces safely. Parents, carers and teachers also need support to have open, informed discussions about social media.”

But of course teaching children critical thinking skills is hard, whereas demanding a social media ban is easy.

Finally from me, one of the world’s foremost experts on deepfakes is, essentially, giving up because of AI. “I feel like I’m going blind,” he says.

What's Hannah reading this week?

I wanted to start my section this week with an article from 2009, where very smart people said lots of smart sounding things about how the cloud was about to eliminate IT jobs. Old news is new again. For me this article served as a reminder that job anxiety isn’t new, in fact it’s a permanent fixture if you want a career in tech.

"The cloud is accelerating that movement of technology into the business, with business-process-level expertise becoming more important than ever. Formerly technology-centric jobs will require a lot more nontechnical, business-oriented capabilities, he says, and IT staffers will increasingly come from the business end of the organization.”

Replace “cloud” with “AI” and you could be reading an article about how we no longer need software developers.

The cloud took infrastructure and made it a commodity, creating jobs for people who could build businesses on top of this new abstraction. There are cloud builders and there are cloud users where there used to be server admins and traditional IT. The jobs changed yes, but there are far more jobs building infrastructure than there were in 2009. We will follow the same path with AI.

This article from 2009 popped up on my feed the same day as I read Bob Dylan’s section in the New York Times about being 80 where he states “You see life repeating itself everywhere”. You don’t need to be 80 to see life repeating itself everywhere, especially if you work in tech. Working in tech you see the hype cycle over and over and over and eventually you get better at riding the rollercoaster. Up and down, up and down. Weee!

On my first literal rollercoaster when I was 12 years old I was so terrified I had my head down and my eyes closed the whole time. Now I sit in the front row with my hands in the air!

At the G7 Summit in Paris this week world leaders sat down with Sam Altman (OpenAI), Dario Amodei (Anthropic) and Demis Hassabis (Google Deepmind) to talk about AI safety and cooperation. We do not need to debate whether these tech companies are more powerful than most governments, they literally have a seat at the table.

It was timely given that the US is still mandating controls around who has access to Anthropic’s most advanced models Mythos and Fable.

The tech guys don’t want the US government cutting off their international revenue. The G7 (apart from the US) don’t want to be locked out of this powerful technology. At this table, only Trump stands in the way of global cooperation and governance. (Oh fuck)

Apparently The White House has told Anthropic that they must prevent all possible jailbreaks before they can make Fable available again. As pointed out by numerous AI and Security researchers, this is not possible. The attack surface is infinite and Anthropic could never guarantee their models can’t be manipulated.

Grab your popcorn because this saga may go on and on.

While the European tech scene is reeling that we’ve been denied access to Fable, Mistral pulled off one of the most beautiful marketing stunts I have ever seen: Le Chaton Fat, a fictitious model of 30 trillion parameters that was reported to be outperforming every other model. So powerful the EU had to limit access to it. With Arthur Mensch CEO of Mistral tweeting that “It's actually le gros chaton” it’s easy to see why some (yes me) were taken in and thought for a minute that a fat kitten might just save us from Trump’s influence.

The best of the memes are below because they are just too good not to share.

The best memes about Le Chaton Fat, a fictitious model.

During a week when everyone outside of the US is figuring out how they can de-risk their dependence on US controlled models, Mistral gave us hope that we do have experts building LLMs right here in Europe. They might not be leading today but as the models improve over time it probably won’t matter.

In the world of software engineering people have been talking about a new open source project called ponytail. The best code is the code you never wrote. Ponytail was created by Dietrich Gebert to improve the efficiency of his coding agents.

You know him. Long ponytail. Oval glasses. Has been at the company longer than the version control. You show him fifty lines; he looks at them, says nothing, and replaces them with one. onytail puts him inside your AI agent.

The results look impressive: ~54% less code (up to 94%), ~20% cheaper, ~27% faster. If you try out ponytail let me know what you think! I’d love to do a followup!

Another sobering reminder came this week from Randy Shoup, who recently spoke about how he had led a successful transformation at eBay, doubling engineering productivity, and it didn’t matter. With so many teams turning to AI to unlock unprecedented speed in software development it’s worth remembering that velocity counts for nothing if you don’t have the culture to make it count.

When it comes to culture Meta has always been viewed as an engineering led, impact oriented company culture … that is until this year. This excellent report from Gergely Orosz (The Pragmatic Engineer) shares the whole story of how Meta’s culture has been destroyed by an AI intoxicated leadership team. Senior Engineers who once had freedom and autonomy have been reassigned to data labelling to train Meta’s next wave of LLMs and they're looking for an exit. When the AI hype dies down and the dust settles I wonder if they can rebuild their culture or if we’ll see something akin to Randy’s experience at eBay - once the culture is gone the pace of engineering doesn’t matter.