The overwhelmingly cheap cost of storage has meant that we rarely delete things any more.

At work we come up with an idea, research it and present our findings as something that we could do. A few months pass, an evolution on that original idea comes along. We refresh our memory from the original idea, research what’s changed and then present it as a new idea, leaving the old one as it was.

Storage is cheap. The old idea is doing no harm. It’s just a harmless old document that no one would go and read.

Until we come to train LLMs on our company documents. All of a sudden, there’s important context missing. All those old (and good bad!) ideas and out of date reports all of a sudden are presented equally for an LLM to train against.

When we then come to interrogate this knowledge, how will we know what’s useful? How will we be able to trust it?

Today shows that you can still play this as a team sport, with honesty, fairness and transparency at the centre of every race and still come out on top as champions.

Papaya rules.

I realise I’m a couple of weeks late to the party with this, but the onboarding user experience with poke.com is brilliant.

It immediately feels familiar, being positioned as a contact through an existing chat interface, and it feels as though it “knows” you within a few minutes. Negotiating a monthly price is also novel, though I expect the vast majority of users end up paying the same amount.

However, I’ve played and enjoyed it for about a week and the novelty is starting to wear off. Sure, it’s immediately accessible and has helped draft a few emails, but in each case I’ve felt the need to adjust the draft before sending. Scheduled reminders and daily briefings are nice, but I’ve struggled to fit them into any sort of routine.

This is very much the state of AI more broadly today. Novel, intriguing, powerful. But we’re still figuring out how to integrate it properly. The real power is still to come. We’re still in the dial-up era.