What a great piece of history here on this article! It's the perfect dichotomy of the practical vs theoretical. Church’s work is theoretically profound, its practical applications were realised later, contrasting with contemporaries whose work had immediate practical impact. I believe the same is happening right now in the world of AI / GenAI, when only later we'll realise that a lot of theoretical work is being deployed in a very practical way. For me it's hard to follow up on every paper and research, but it's been one amazing journey to get all the details. Yet, as Church believed, no algorithm can determine the truth of all mathematical statements, which is something that we see everyday when we challenge the limits of what ca be computed.

I'm biased here... Very! Firstly because the Dreamcast was a great console that I always wanted and back then was hard to get - having adult money is a bliss.... But then because it had a Hitachi CPU, and now I work for Hitachi... It's like a full circle. It also shows the engineering work around Linux to make it work for this console - which was the precursor of Linux on PlayStation. The best part is fitting Linux into the hardware constrains and then run it. For me this offers insights into vintage computing and the evolution of operating systems on gaming hardware.

Black Friday is just around the corner and if you work in Retail, I feel your pain! It's one of the busiest times of the year, anticipating the other time, Xmas. This Shopify globe of what people are buying is an amazing visual experience that now just got some improvements. The best part is when we understand the technical details behind it and what are they doing to improve the experience. I do enjoy learning all the infrequencies of the systems and how technical teams address problems and come up with solutions. More companies should do this, so that knowledge can be shared.

Lets finish on a strong note this week! Feature steering sounds like an elegant solution to a messy problem - adjusting AI’s biases without breaking it. But let’s be real, tampering with these systems feels a bit like hacking your own code after deployment: it works, until it doesn’t. Steering is a balancing act—enough to nudge outputs, not enough to wreck functionality. Off-target effects? They’re the plot twists of AI tweaking. And the “neutrality feature” sounds great on paper but feels like trying to make a compass point north while spinning a globe - it’s all connected in ways we don’t fully grasp. Ultimately, while it’s crucial to address AI bias, the broader challenge is keeping these models useful while avoiding a Pandora’s box of unintended outcomes. Maybe the sweet spot isn’t a setting but a philosophy: proceed carefully, and keep asking the hard questions.