CrapGPT and an historic mortgage crunch

Top of the Charts

Afternoon all,

Quite a week. Mortgages are up, Boris is out.

Basically Britain’s having a full on protestant work ethic/puritan phase – no-one with a mortgage can holiday/eat out/savour the small joys of life, but at least we can tell the kids that crime = punishment. Cromwell would be proud. Until he found out that Charles I wasn’t dead Boris was basically going to get away scot free with breaking all the rules for ex-ministers taking new appointments and earn squillions writing for the Mail.

If you can’t enjoy afford the finer things in life basics, at least enjoy this week’s reads. I would say enjoy our COTW too, but you won’t if you’ve got a mortgage/are the government planning an election in 2024 just as the mortgage crunch is set to get crunchiest…

Have a good weekend.

Chief Executive
Resolution Foundation

Callous competition. Being competitive often predicts achievement at school and work – you’ll have noticed. But a study gets into the nitty gritty, breaking down competitiveness into its different motivations (enjoying competition vs needing to win). Both were linked to career success and women tended to indicate a lower willingness to compete on both motivations. But guys might want to hold off champagne-wise. If your key motivation for competition is a desire to win (the category with the most significant skew male) you are likely to be less happy and more neurotic. And more Machiavellian. And more of a narcissist. So, being competitive is fine, but being a psychopath? Not so much.

Difficult direction. It’s London tech week, so the heatwave has been reinforced in central London by all the hot air being talked about AI. More useful was Noah Smith’s blog on calls to “direct” this technology, so as to influence the nature of its impact on workers/humans (ie so it destroys the right jobs/boring tasks). His basic view is that would be nice but it’s not possible in practice – we just don’t know enough about how technologies end up changing the jobs we do. Instead, the answer is to put in place robust institutions which can deal with the disruption (ie unions/welfare states).

Children of ChatGPT. While we’re on interesting AI takes, this paper (great summary) provides the answer to something I’ve been pondering. If we’ve trained the current generations of GPT (and other large language models) using online text written by humans, and that technology now gets used to create loads of new text for the internet, what happens when we train the coming generations of GPT on the text largely written by their predecessors? The content gets worse generation on generation, slowly filling the internet up with what the authors technically labels ‘blah’. The paper has the maths, but the intuition is this is the new version of that old problem when photocopies of photocopies got lower quality each time you take another copy (and no I don’t want to hear from those of you too young to remember such things). This is the AI version of inbreeding.

Forceful forces. Assaults on police officers are up in England and Wales – there were 41,000 in 2021-22, with around 12,000 of those resulting in injury. There’s also rightly a focus on the use of violence by the police. But the two aren’t entirely separate. New US research finds a link between officers being injured on duty, and their peers deciding to use force in the following days and weeks. An officer being injured increases the likelihood of officers that trained, but do not work, alongside them using force by 7 per cent. That use of force in turn (unsurprisingly) led to more civilian injuries. Notably, the effect is larger when considering officers of the same race. The point the authors are making is we want to reduce violence against and by the police. But it’s much easier to identify a vicious cycle, than end one.

Growing green. How should the EU respond to US green manufacturing subsidies asks a new note from our friends at the Centre for European Reform. Examining trade patters across electric cars, heat pumps, solar panels etc, it finds that the impact of gravity (i.e. how much distance impacts trade volumes) has been getting stronger over the past five years. Why? Because as many technologies mature, firms open production sites nearer customers/other parts of supply chains to save transport costs. Basically, the argument is that the EU market is big enough so that ‘on-shoring’ production into it will happen anyway in lots of these technologies, so there’s no need for big subsidies. What to do instead? Focus on the technologies where this won’t happen and you don’t want to rely on China (ie wind power – where China is in the middle of eating the EU’s lunch). Britain of course faces rather bigger questions trade strategy wise – like shall we have one (we think we should and this is what it should be).

Chart of the Week

…is on the looming mortgage crunch. We’ve got new analysis on the scale and timing of the crisis out tomorrow (have a stiff gin ready), but COTW is our go at putting this mortgage crunch into historical context. Specifically, we wanted to answer the common question: how this crunch compares with the big rate rises from 1989. After the most recent market panic on how high rates will go (nearly 6 per cent is the current bet), it’s now bigger is the answer. The increase in mortgage costs compared to last year would take up an extra 3 per cent of income for a typical household with a mortgage. The pain might be bigger, but it’s also more concentrated on a smaller share of the population: near 40 per cent of households had mortgages in the 90s but, with older people more likely to own outright and younger people less likely to own at all, that has dropped to below 30 per cent today. So a crunchier, but more concentrated, crunch is what’s coming.