The null hypothesis is that AI fails to do something. To falsify the null, we need the strongest, not weakest, attempt to prove AI success

The null is easy to not falsify: if you want AI to fail, it will fail with bad prompts or bad models or a failure to try to figure approaches that work for LLMs

The PCUK argument leans heavily on public opinion and says almost nothing about the statistical evidence. That's a spectacular sign of a bad argument.

The problem is many men over 50 will have something that looks like prostate cancer if you look hard enough, but in most cases the cancer won't progress, or will progress so slowly it would never cause harm in their lifetime, and it's hard to know which ones you can ignore.

"A poll by Prostate Cancer Research of 3,000 adults shows that 90 per cent believe it is important for the UK to introduce screening."

Is that what we do now? Clinicals trial are too hard. Let's just do an opinion poll instead.

% who say each 2025 Budget policy was the 'right thing to do' (1/3)

Increase gambling taxes: 82%
Freeze rail fares: 82%
Reducing energy bills by £150 a year by reducing green levies: 75%
Increasing minimum wage: 71%
Mansion tax: 67%
Decreasing biz rates for retail/hospitality/leisure: 64%

This is what he said about the Truss/Kwarteng budget: 'This was the best Budget I have ever heard a British Chancellor deliver, by a massive margin.'

Ironic that this government was elected with a landslide majority on a one word slogan of "change". Given how worried it is that delivering anything even close to that label would upset too many people.

There's no obvious big picture sitting above this budget beyond muddling through.