Constant

Hi, Tudinh here.

I’ve always thought Einstein’s theory of relativity has a misleading name, because it was built around one thing that isn’t relative. Light always moves at the same speed, however fast you go, so he rebuilt everything else around it. Even time turns out to stretch, while the speed of light stays constant.

Einstein, the man who stretched time, often turned to timeless music

When everything around us keeps changing, our instinct is to stay flexible, yet there is something deeply reassuring when a thing is kept fixed. With so much shifting, fixed points matter more than they used to. For a brand, what stays constant might be a typeface or a promise, but choosing what gets fixed is harder now that everything else can be rebuilt more quickly.

Rebuilt by machines

Last month OpenAI’s machines took 88 hours to claim a solution to Navier–Stokes, one of the most important problems in maths, which had been open for nearly a century. Now OpenAI has claimed 372 more in a single day, among them 25 that Paul Erdős left unsolved. It may be one of the biggest days maths has ever had (though unsurprisingly not all of them hold up to the scrutiny of the maths community).

Many of us were drilled at school that showing your working in maths mattered more than the answer. A good proof is elegant, and once you grasp it you see why the answer is true. Machines show their working too, but they tend to bolt steps together until something works, leaving proofs that check out but no one can follow. A guest essay on Fields medallist Terence Tao’s blog calls this a golden age if machines help us understand more. It is a dark one if we rely on proofs nobody understands.

Paul Erdős, who set the problems, with a 10-year-old Terence Tao, 1985. Photo: the Tao family, CC BY-SA 2.0

I find myself wondering who will train the next Tao, and whether it will be a machine. Mathematicians learn the way he did, from each other. If machines do the proving, fewer will work on proofs together or be left to teach the next one. If people keep learning from each other alongside the machines, there is an excitement in the possibilities this opens up for us all. Advances in knowledge and discovery could reach daily life and the technologies we use, improving how we live.

Redesigning Dezeen around its readers

Nearly twenty years after Marcus Fairs founded it, Dezeen had built up a loyal readership and tens of thousands of articles, adding services one at a time. That growth had also left the website fragmented and harder to maintain every year.

For its first major redesign in ten years, we kept what readers knew and rebuilt the system underneath. Search and recommendations, once rented, now run on Dezeen’s own infrastructure with no licence to renew. All of it runs on 100% renewable energy. Choices like these are what make a rebuild better, not just newer.

“The ambition has been to create a richer and more versatile site that retains the spirit of the Dezeen that is known and loved. Marcus’s vision when he started Dezeen almost 20 years ago was to set the global architecture and design agenda … The new website will enable us to continue that mission for the years ahead”

Benedict Hobson and Wai Shin Li, co-CEOs, Dezeen

Read the case study

Good decks are hard, for people and machines

A deck is a string of choices, what to say, in what order and what to leave out. By the time it has been passed around, the fonts and the photos have usually changed. Press the AI button to enhance a slide and it rarely does, because it doesn’t know your brand or what you’re trying to say.

A brand’s fixed points, its typeface and its promise, usually live in a guidelines PDF that gets read once, if at all. We built AI Brand Engine to run them whenever a machine makes something, like a deck. A machine checks every rule, but unlike those proofs, anyone can follow the reasons.

Before it goes out, the engine flags what breaks the rules

AI Brand Engine pulls pictures straight from Air, one of the picture libraries it works with

When I asked for a shorter version of one of our decks, it cut the slides that no longer fitted and sent the new deck to Google Slides. It knew why each one was there. It uses the words and pictures a team already has, wherever they’re kept, and works with whichever AI the team already uses.

From the team

• Nobody owns the topic yet — everyone has the same AI tools now. What they can’t copy is your choice of a few topics to own, said clearly on your site and backed up elsewhere. Barry Cumberlidge, Chief Client Officer
• Internal work has no client — internal projects have no client waiting on them, so we borrow a deadline, or make a promise in public and keep it. Matthew Froggatt, Executive Chair
• What does collaboration look like at the proposal stage? — the best agency partners ask how it will be built when the brief lands, not once the creative is set. Roger Whittlesea, Director of Collaborative Partnerships

When machines can rebuild almost anything, is changing it still the hard part? More and more, I think the hard part is deciding what to keep, and then keeping it, constantly.

Until next month,

Tudinh Duong

Founding Partner & CEO

Made by ON

Ps. When Einstein was proved right in 1919, the New York Times put its golf correspondent on the story. Shut out of the meeting, he rang the astronomer Arthur Eddington, then asked him to explain it again in simpler words. The headline read “Lights All Askew in the Heavens”.