I wouldn’t put too much weight into this.
Some pattern with Cambridge researchers I noticed, they deliberately pick study populations that will give them a more likely preferable outcome to their research and think nobody notices.
Look at the study population.
> mean age = 45.9 years; age range = 18–81 years). This sample was recruited to be representative of the population of the United States on the characteristics of gender, age, and race. Participants were paid $3 for a 15-minute study.
See how they picked gender, race, and age, instead of also diversifying by level of education.
That’s not an oversight,it’s part of the study design to create preferable outcomes to then generate more funds for their research with the attention they get from publishing.
Most people 45 and above will not have any education at all about AI. So a large part of the study population has no reference or internal tooling in their brains that allows them to recognise patterns used in LLMs in the first place.
On top of that, they only had 15 minutes of time and got 3$ for it.
What kind of study population do you think a 3$ reward attracts?
And I am not saying these people are stupid, but I say if you sell 15 minutes of your lifetime for 3$ , I highly doubt you have the funds and time to meaningfully educate yourself about AI and gather references about how AI uses language differently, because of how the technology is designed. But there are also a lot of people who do that will simply never join the study, because they have zero incentive to do so.
So, good for them they are able to trick the average HN Billionaire, but the study design is somehow flawed by not taking education into account.
May their funding never dry up from your taxes.