Meanwhile on Mastodon
datasci.social/@dodecadron/...

I was going to say, but they don't use LLMs for trials surely... but then omg simulated data??

Hmmmm first thought: anyone can use the automatic plagiarism machine but not everyone can or needs to conduct a study or analyse a dataset. But maybe you have in mind practices relying on data, like what goes on in businesses or how people use tracking apps. They could use more rigorous inference?

All very nice but please consider how the message will be understood by those of us who weren’t born here. You risk replacing one exclusion with another.

Very pro just-vibe for titles 😊 Unless someone is running a study on title prevalence. Gender, ethnicity, etc. better as tick box plus option for freetext. Code up the freetext too.

"Isolation" at Gorse runs from 8.30am to 3.05pm and sees pupils having their lunch in the room and not being allowed to socialise with other students.

The Gorse Academies Trust’s current behaviour policy also sees pupils facing what it itself calls “Isolation” for “lines in their eyebrows,” uniform violations or forgetting their student planner twice.

You'll have to explain it a bit more and use emojis if there was a joke thanks

For trials we use permuted blocks or minimisation and siblings

The solution obvs is to add more detail for gender and gender modality then I'd be happy

Also I'd guess what happens is that someone suggests adding a small number of standard titles like Mx plus a free text option and then someone else says, well if we're having those "woke" titles then I want Captain etc. too.

Annoys me a bit because it's rare to get the same detail for gender.

Also now wondering how often empirical analyses of PRNGs are used to crack public key encryption.

I know there have been (potentially deliberate) weaknesses introduced, e.g. en.wikipedia.org/wiki/Dual_EC...

This is fun :) The longest I've found so far is 32 ones with seed 1002755089.

I've covered from 0 to 1398462874 and 1890340790 to 2147483647.

Incidentally, I wonder how common it is for people to use negative seeds.

A spot of parallelisation would speed it along.

Oh dear that's not a very helpful discipline difference. It'll explain why my googles on "completely randomised" have occasionally been very confusing. Imbens and Rubin (2015), for example, use it to mean fixing the number of units in each experimental condition in advance.

Fisher would have advised laying my the field in equally sized blocks, each divided into n plots of equal size corresponding to the n crop treatments under study. Within each block, the assignment of treatments to plots would be completely randomised: so no risk blocks all have the same treatment.