didau.bsky.social
Same old same old
Substack: daviddidau.substack.com/
376 posts
3,369 followers
102 following
Regular Contributor
Active Commenter
comment in response to
post
I like Deming.
comment in response to
post
This is typically good, but I wonder if a thing that we need to be open about is how *unlikely* it is most people will do hard work *before* they turn to AI. Most of us will take the easy route if it's available to us.
comment in response to
post
Ta
comment in response to
post
There’s a case for saying that but I’m not sure how you make it a reality. Especially as it’s already being embedded in every tech tool
comment in response to
post
Oh, Socrates was definitely right: reading rots the brain. There’s no doubt writing is a form of cognitive offloading. But - and this is I think the clincher - there’s no putting the genie back in the bottle.
comment in response to
post
For me, the first countdown is a model - "this is what I want to happen". The slower repetition, scanning for compliance, is the making it happen and the long......... pause before 1 ensures everyone is successful
comment in response to
post
It *might* stay the same but OK, I take your point
comment in response to
post
I'd say 2 is too many
comment in response to
post
Without an appreciation for the differential accuracy of estimates of status vs. change, schools will continue to conflate test scores with growth, and mock data with insight.
So yes: let’s not throw away the mock exam. But let’s also stop pretending it tells us more than it does.
comment in response to
post
The real issue isn’t the existence of noise; it’s the epistemic overconfidence with which noisy results are interpreted, often by people unaware of the statistical traps they’re falling into.
comment in response to
post
We read them for pressure, when they are measuring something entirely different and when we start forecasting storms based on those readings, we inevitably overreach.
comment in response to
post
So perhaps the point isn’t that mocks can’t yield useful data - clearly, when designed well and interpreted cautiously, they can. It’s that the function of mocks is misunderstood. They are often used as thermometers, when they are in fact barometers.
comment in response to
post
As you elegantly put it, frequent tests improve our grasp of position, but degrade our confidence in velocity. That might not be Heisenberg, but it’s certainly instructive.
comment in response to
post
That is where NNT's point has relevance: not in the absolute unreliability of freqnt tests but in the diminishing signl of diffrnces betwn them. The closer our obs, the more we’re trying to infer movement from tiny perturbations, & the more likely we are to be misled by normal fluctuations.
comment in response to
post
Worse still, as the gap between S₂ and S₁ narrows (e.g., under short intervals or diminishing returns), the signal becomes harder to detect amidst the amplified noise. So frequent testing, if the aim is to detect progress, gives the illusion of resolution while making the actual picture blurrier.
comment in response to
post
It’s not that the second mock is less accurate than the first - it’s that the difference between them is a noisier estimate of change than either mark is of skill.
comment in response to
post
As you say, when comparing two assessments to detect progress, we’re looking not at S but at ΔS = S₂ – S₁, and the difference in noise terms (R₂ – R₁) compounds the uncertainty. The mean might still tell us something, but the variance inflates.
comment in response to
post
But where things get interesting - and where the mock regime collapses under its own contradictions - is when we switch from measurement of status to inference of change. That’s where the maths gets much less forgiving. (Or at least, I think so - you're the expert...)
comment in response to
post
The result is greater reliability, at least in estimating position. This is how standardised testing works: by pooling enough items, we reduce the standard error of measurement. So far, so good.
comment in response to
post
Your example, that exam outcomes can be modelled as S+R, with S representing stable skill and R randomness, offers a more precise & more useful framing. When we average multiple assessments, the random components (assuming independence) begin to cancel out, reducing variance around the signal.
comment in response to
post
You’re right to call out a misreading of what frequency of measurement actually does. The Taleb quote while rhetorically satisfying encourages a kind of seductive fatalism as though increasing observation inevitably drowns us in noise. As you point out it rather depends on what we want to observe.
comment in response to
post
Ok. I shall endeavour to absorb that. Will come back to you…
comment in response to
post
Blister in the Sun. Excellent choice.
comment in response to
post
Agh. What about The Sun Rising by The Beloved for that blissed out, post rave summer morning?
comment in response to
post
Cruel summer was my #6
comment in response to
post
“And by the way, pass the king size rizzla”
comment in response to
post
Sunstroke: choon
comment in response to
post
1. Summertime Healing, Eusebe
2. Summer in the City, Lovin’ Spoonful
3. Under the Bridge, Chilli Peppers
4. Cannonball, The Breeders
5. Hot Stepper, Ini Kamoze
#fridayfive
comment in response to
post
I’ve not read it. I find Atkinson quite hit and miss. Loved Love After Love and A God in Ruins. Have given up on a few though
comment in response to
post
Can you do a short of the giggling?
comment in response to
post
That was a short extract from the latest podcast I do with @didau.bsky.social www.youtube.com/live/62GTaJZ...