AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

Hannah is home from Italy for 2 nights before repacking her case and heading to London for two events this week. She will be speaking at AI Dev Con and Lead Dev about the shifting roles, responsibilities, skills and team composition in software development. Ahead of her talk, Tessl has also published Hannah’s opinions and observations on this topic on their blog.

Charles’ house is full of teenagers for half-term, not all of them his. It’s been fun, and he loved having three days of minimal screens over the bank holiday weekend, and catching up with some very good old friends. He also went to see Kaitlyn Aurelia Smith at the Barbican on Wednesday, where she was playing with the London Contemporary Orchestra. She’s best known for mellow, reflective music but the second half was an unexpected riot of heavier drums and breakbeats stemming from a need “to make something confrontational, almost like sonic combat”. Her new album, Ruin: It’s Not Just Music is released on 2nd October through Someone Special, a label she co-founded this year with James Daniel. You can listen to a title track of sorts, “Ruin”, via a video starring pole-dancing Pedro Esteveaqui.

We’re just a couple of weeks away from Agent Craft our one day unconference for Agent Builders. As part of our commitment to diversity, accessibility and inclusion we always offer financial support to people who would like to attend but can’t afford the ticket price. If you’re a student, career changer or unemployed you can apply for a free ticket. We try to help as many people as we can!

Have a great week!

Hannah & Charles

Try Leapter's Visual Language

You don't need to read code to start building with Leapter. Try Leapter today and let them know what you think at showcase.leapter.com

What’s Hannah reading this week?

I wasn’t expecting to write about the pope in this newsletter but Pope Leo XIV has opinions on AI, and they are worth hearing. This week in his first encyclical (that’s an important popey speech) he called for the disarmament of AI for the protection of humanity. He talked about the risk of monopolistic control, the danger of autonomous weapons and ecological impact.

“Disarming AI means freeing it from the mentality of “armed” competition, which today is not limited simply to the military context, but is also an economic and cognitive phenomenon. This entails a race for ever more powerful algorithms and larger datasets, driven by the desire to secure geopolitical or commercial dominance. To disarm means discrediting the assumption that technical power automatically confers the right to govern. To disarm does not mean rejecting technology, but preventing it from dominating humanity. It means freeing technology from monopolistic control and opening it to discussion and debate, therefore making it human-friendly and restoring it to the plurality of human cultures and ways of life. Our task today is not only ethical or technical. It is ecological in the deepest sense, for it concerns a new dimension of our common home. AI is already an environment in which we are immersed, as well as a force with which we must engage. For this reason, merely regulating it is insufficient; it must be disarmed, welcoming and accessible.”

Technology is constantly reshaping our world and touching our lives, and if we accept that AI is here we must wield it for the common good. I appreciated this choice of words: “AI is already an environment in which we are immersed” which mirrors my own acceptance of AI. I am constantly worried about the impact of this technology but opting out doesn’t feel realistic. Helping to steer the conversation, using the technology to do some good and bringing more people into the dialogue is the reason I continue with AI for the rest of us.

It was a long session but the Guardian helpfully provided these highlights in less than 2 minutes and the full document “Magifica Humanitas” is published online.

It was surreal to see Chris Olah, Co-Founder of Anthropic, sat alongside the pope for this big moment. He spoke about the incentives that drive a successful company, being transparent about how they sometimes conflict with the greater good and he welcomed the push and pull that comes with outsiders holding Anthropic to account. I felt pretty sad that Chris felt like his incentives needed to be kept in check and that a business would inevitably behave unethically in pursuit of profit. This is a choice, not a mandate.

Whichever way you look at it this was an incredible PR opportunity for Anthropic who are increasingly looking like “the good guys” of AI, or at least trying to be. Trying is the first step to becoming. The full transcript has been published on the Anthropic Blog.

In the world of open source technology we’re still navigating the impact of AI models on cyber-security. We know that models like Mythos can find and exploit vulnerabilities, what we don’t know is how we’re going to address this risk at the speed and scale required. In response, this week RedHat / IBM announced Project Lightwell, an investment of $5bn and 20,000 developers to secure the world of enterprise open source.

Cynical Hannah (yes there is one) thinks this is just the practical response to all the contractual customer SLAs they already have in place that promise vulnerability remediation within a set timeframe. I’ve worked on enterprise agreements like this and they come with some hefty financial penalties. That’s pretty darn motivating!

Chainguard have also announced their own investment of $50m and 100 engineers to “save open source from Mythos”. Referring to it as “The Hardest Fork” CEO Dan Lorenc writes about the options available to companies who are dependent on the open source software that is under threat. One option is assuming responsibility for the security of that open source by forking and maintaining your own version and Chainguard are planning to leverage AI tools to make this happen, should that become the only path available.

I can’t help but think all this money could and should have some kind of positive impact for maintainers. Neither company mentions how much of this funding will be made available to help create sustainable remuneration for unpaid maintainers. This xkcd comic comes to mind.

Netflix received some negative press this week when they shared the new poster for Enola Holmes 3. It’s bad, like really bad. But that's not what people are debating. People are debating whether it’s AI generated bad or incompetent human bad? Does it matter? And what does it mean for creatives if we all accept this level of slop from a company worth $435bn?

It reminded me of the recent embarrassment at Manchester Airport where a prominent billboard featured an AI generated image of a man with 3 hands. “Manchester Airport makes $1.3 billion a year. You’re telling me they couldn’t afford a photographer and some models for this?” asks Thom Rylance of The Lottery Winners. He's not wrong.

While the pope warns of AI as “a force with which we must engage” popular dating app Bumble has decided that its AI assistant can help you find a partner far more efficiently than swiping through pictures of eligible singles. I have done a lot of dating in my time, I have used many dating apps and I honestly can’t see this working. Online dating already creates a warped perception that there’s an unlimited number of potential partners and that your criteria for a partner are the most important thing. A check box exercise. These apps make you focus on the fulfilment of your own needs before they allow you to connect with someone just as they are, with no expectations.

A dating profile with a sprinkle of originality and humour would have me throwing out all my dating criteria, my curiosity piqued by a tiny glimmer of their uniqueness. How can an AI assistant replicate that? Will it simply queue up a list of cookie cutter singles who check all the boxes? I’m not sure how this UX will manifest in the app but prioritising efficiency over serendipity doesn’t feel right to me personally.

What's Charles reading this week?

The New Stack has published the last article in my series with Kin Lane and Naftiko. I've written all six of these, a rare privilege, and it has been fantastic to spend time with Kin to really understand his thinking and how he sees generative/agentic AI from the perspective of an API expert.

In this last article we looked at business observability — what it means, and how to do it. Sometimes sponsored posts are a bit too vendor-y for my personal taste, but I very much feel we produced something helpful and worthwhile here. The thread running through the whole series is that the organisations best placed to do well with GenAI are the ones that did the boring-but-important work years ago and have well-written documentation, clear API contracts and consistent naming.

The BBC’s Joshua Nevett reports that the UK Home Office is rolling out AI facial recognition technology at the border to detect adult migrants claiming to be children, after data showed that 43% of the 6,400+ people who claimed to be children at the border last year were deemed to be adults. A £322,000 contract has been awarded to an IT firm to develop and test the system, which estimates age from photographs, with a planned live trial at Dover's Western Jet Foil processing centre in 2026 and full deployment by mid-2027.

I’m very alarmed by this, because the decisions really matter, and the broader picture from AI research isn't encouraging. Facial analysis systems have well-documented problems with accuracy gaps across ethnicities — they tend to perform worse on darker-skinned faces, which is a significant concern given the population it would be used on. Age estimation from photos is also a genuinely hard problem even for humans, let alone algorithms, because people of the same age can look very different depending on health, stress, sun exposure, and a host of other factors. Someone who has spent years in difficult conditions may look considerably older than they are. As a side note I want to recommend a book here called My Fourth Time, We Drowned by Sally Hayden, which explores the human side of state efforts to control and contain mobility, and is one of the most remarkable pieces of journalism I have ever read.

We didn’t cover GoogleIO last week, mainly because I hadn’t had time to catch up with it all. Having done so, I don’t honestly have much to say about it other than Google has a huge number of products and surfaces, and it’s adding AI features to all of them. But it is worth high- (or possibly low-) lighting that Google is continuing to push search from being about navigation and discovery to being about summary and extraction.

We’ve talked about this in the newsletter previously, but what this represents is a direct challenge to the implicit social contract of search and, essentially, the free internet. It used to be that publishers let Google index their content (and make money off it), in return for Google sending them traffic. As we shift to Google Zero — i.e. no traffic from search at all, how does the web work? Of course, if Google doesn't do this OpenAI will, so they kind of have to. But I sure wish someone with very deep pockets, like Apple, would make a good search engine again that didn’t rely on subscriptions.

Another subject we’ve talked about quite a bit is that generative AI has unfortunately been sold as a way to enable very rich bosses to make even more money to spend on (to pick something at random) exploding rockets by making lots of other people redundant. From a marketing perspective this is a lazy, terrible strategy. The latest person to find this out was Standard Chartered’s CEO Bill Winters who, whilst discussing how automation was likely to lead to thousands of job cuts at the bank at a recent conference, said it wasn't about cost cutting but "replacing, in some cases, lower value, human capital, with the financial capital and the investment capital that we're putting in". A phrase so exquisitely, perfectly designed to hit the proverbial fan I felt genuinely sorry for Standard Chartered’s PR team. Also I want “Lower Value Human Capital” on a t-shirt.

As anti-AI sentiment continues to grow, Wired reporter Daniel Boguslaw has obtained over 1,000 pages of documents via FOIA requests. He reports that in the US the Department of Homeland Security, FBI, and the post 9/11 intelligence-sharing hubs known as fusion centers are increasingly monitoring and categorising people critical of technology, particularly AI and data centres, as potential extremists. “This new effort follows President Donald Trump's National Security Presidential Memo 7, which instructs the Department of Justice to target anyone holding ‘anti-American’, ‘anti-Christian’, and ‘anti-capitalism’ beliefs. Earlier this month, Trump's counterterrorism czar, Sebastian Gorka, released a public counterterrorism strategy claiming that left-wing extremists are one of the three top counterterrorism priorities facing the United States,” he writes.

Staying with Trump, his long-awaited AI executive order, meant to ensure that world-ending chatbots don't fall into the wrong hands, collapsed spectacularly when David Sacks, a venture capitalist who is one of a seemingly limitless number of people who has the president's personal number, rang Trump the morning of the signing and whispered the magic words, "It'll slow down innovation and China will win". That was enough evidence for Trump, who promptly scrapped the whole thing, leaving his own staff standing around a signing ceremony with no order to sign — whilst the invited tech CEOs hadn't even bothered to show up. Never mind that Sacks had been briefed on the order all week and said nothing until a last-minute call that, per a White House official, blindsided everyone "his own staff included". Imagine if the US tried a wild experiment: electing a group of reasonable, thoughtful people to actually sit down, debate, and pass laws through a clear and logical process. We could call it a 'legislature’.

SpaceX filed for an IPO this week. The S1 contains a laugh out loud graph of the sort Jeff Bezos specialises in. It comes in a discussion of SpaceX’s total addressable market:

SpaceX may well get its desired valuation even though the numbers are obviously absurd. The company is seeking a $2 trillion valuation, which is 107 times last year’s $18.67 billion revenue. For context Palantir, the most expensive software stock, trades at around 40 times next year’s revenue. SpaceX also posted $4.9 billion in losses last year, with growth slowing from 35% to 33%. This slowdown happened due to the addition of xAI (and thus also X), which tipped the company from a small profit to that massive loss, thanks to $5.1 billion in AI R&D expense. That R&D, as regular readers will know, went towards building a model that is in 5th place, whose entire founding team recently left the company, and whose USP is that it is particularly good at generating non-consensual deep fake porn including of children. But sure, it's a $26.5 trillion AI opportunity!

SpaceX’s Starlink is, I think, brilliant tech, but the firm has yet to demonstrate it can make a viable, fully-reusable rocket. It is possible that they could get to that valuation eventually, but I would advise against anyone older than about 50 investing, which takes me out but possibly not Hannah. I wouldn’t invest anyway, for ethical/moral reasons.