It certainly seems to be the case that e.g. Claude isn't aware of its training set and therefore can't reason about its own capabilities, I'm not clear why it couldn't?

I think your wardley map explains why they don't want to. I think they've got to try being in the higher levels of value

I've found that the command blacklist works well - are you finding it ignores it?

I think the levels of tech debt might be lower in absolute terms, but higher relative to building. Many of the things that matter are about providing guide rails for the ai, many of which involve left shifting. Debt ridden code almost by definition lacks these.

The Cambridge-Manchester collab makes more sense to me than the Cambridge-Oxford corridor. But it's not new, so you can't announce it to make it look like you're doing something.

Fantastic article on sweating the (right) problem. A timely kick up the arse to some of my own approaches at the moment :-)
nilam.ca/2024/11/02/t...

That should really be "Harford's Palchinsky Principles".

Palchinsky, in the end, tried the new idea of political dissent at a scale which proved not to be survivable.

Mostly as a reminder to myself: Palchinsky's principles:
1/ Seek out new ideas and try new things
2/ Try new things at a scale where failure is survivable
3/ Seek feedback and learn from mistakes
Objectivity and honesty required for 3!