Week 169: Red Kites

A week of internal tech conferences, theatre databases, pretty countryside, Midsomer Murders, birds of prey, memories, galleries, roast dinners, and overhyped AIs.

Posted on by Rowan Manning.
Tagged with WorkWalkingFamilyFoodArtAI

  • This week was Engine Room, the FT’s internal tech conference. As always, it was a fun day and really well-organised. I shouldn’t pick favourites but I think Andy’s talk on his decade-long (!!) Theatrebase project was really excellent. I’m a big fan of a talk where somebody introduces you to a subject they’re passionate and knowledgeable about; this ticked all the boxes.

  • A Red Kite flying low, taken from beneath it. It has broad wings and a slightly forked red tail

    On Friday we headed over to Marlow to stay in a hotel. It’s very pretty and is also where a lot of Midsomer Murders was filmed. On Saturday we did a nice 12-mile walk across the countryside, looked at some of the filming locations, and saw several pairs of Red Kites. It was a lovely day, walking in the cold is my favourite and the ground was crunchy with ice.

    I love seeing Red Kites. Back when I was a kid my dad would frequently point out wildlife, particularly birds of prey. He loved Red Kites and I have really early memories of him talking to us about the conservation efforts after years of persecution. Whenever I see them I think of him.

  • On Sunday we met up with Luke and walked around the Courtauld Gallery for a while, marveling at the medieval animal depictions. Afterward, we wandered up to The Ship in Holborn for a lovely roast.

  • I started writing a blog post about generative AI. I put together an outline and then realised it was so large in scope that it’s probably more like a series and I need to rethink it. Just writing an outline was helpful, I can probably more readily talk about all my issues. So far the topics I want to cover are:

    • How bad LLMs are at writing code, particularly from a reliability and security perspective.

    • How showing someone your prompt is probably more valuable than showing them an AI-generated document.

    • How LLMs don’t know things, covering hallucinations.

    • How things will degrade over time as more of the training data consumed by newer LLMs will have been created by previous iterations, touching on how much junk is flooding the internet.

    • How using an LLM is stealing, probably more from a personal moral perspective but also covering the many lawsuits.

    Let me know if you’re interested (or not) in any one of these. I’m feeling spicy about it all at the moment which is a good motivator.