Education's Verification Problem
Education cannot improve without knowing what works
May 13, 2026 · 1 hr 36 min · 12 segments
In this long-awaited sequel (eight years after her last appearance), Craig welcomes back Becky Allen — education researcher, co-founder of Teacher Tap, co-author of *The Teacher Gap*, and now a…
Becky AllenGuest
Craig BartonHost

When ChatGPT arrived, I kind of knew what I was seeing in the sense that a very long time ago when I was a PhD student, some of my contemporary PhD students were working on what I think was the precursor to all of this, which was image classification and neural nets, and they were working within health research.

And so because I'd taken a lot of time to try and make sense statistically of their work, as a result, I kind of had half an eye on the scientific side news that was going on around the translation of that from image generation into text creation.

I was incredibly excited and I think anybody who has known me for the past few years was known that I have been an optimist about not only the state of the models, but also like how good the models were going to get.

of us developing technology that will allow us to do the kind of cognitive work that I do and you do and that we think we are uniquely able to do, but maybe we're not.

I think the other thing that meant I had a really good experience early on was that early on these models were immediately pretty good for coding.

So I could immediately just read and understand parts of the teacher tap code base that had been written by our tech team and that I had no capability to be able to read and understand them.

And then from there, I was very quickly writing in like coding languages that I'd never learned in order to just like hook up our data, you know, into automations and other things because I couldn't be bothered to wait for a techie person to do it for me.

So I had all those kind of like reinforcing good experiences of AI, perhaps in that age in the early models where actually it wasn't great for loads of things early on because it didn't because it couldn't write that well and it hallucinated madly.

But now, you know, how do I use it today? Well, I use it for almost everything that I'm doing.

So I now... almost exclusively don't type back and forth with it unless I'm in a private place.


When ChatGPT arrived, I kind of knew what I was seeing in the sense that a very long time ago when I was a PhD student, some of my contemporary PhD students were working on what I think was the precursor to all of this, which was image classification and neural nets, and they were working within health research.

And so because I'd taken a lot of time to try and make sense statistically of their work, as a result, I kind of had half an eye on the scientific side news that was going on around the translation of that from image generation into text creation.

I was incredibly excited and I think anybody who has known me for the past few years was known that I have been an optimist about not only the state of the models, but also like how good the models were going to get.

of us developing technology that will allow us to do the kind of cognitive work that I do and you do and that we think we are uniquely able to do, but maybe we're not.

I think the other thing that meant I had a really good experience early on was that early on these models were immediately pretty good for coding.

So I could immediately just read and understand parts of the teacher tap code base that had been written by our tech team and that I had no capability to be able to read and understand them.

And then from there, I was very quickly writing in like coding languages that I'd never learned in order to just like hook up our data, you know, into automations and other things because I couldn't be bothered to wait for a techie person to do it for me.

So I had all those kind of like reinforcing good experiences of AI, perhaps in that age in the early models where actually it wasn't great for loads of things early on because it didn't because it couldn't write that well and it hallucinated madly.

But now, you know, how do I use it today? Well, I use it for almost everything that I'm doing.

So I now... almost exclusively don't type back and forth with it unless I'm in a private place.
Every episode on Radar is fully transcribed, speaker-labeled, and rich with metadata. Here is a taste of this one. Try Radar for free to see the rest.
3 of 11
Education's Verification Problem
Education cannot improve without knowing what works
The Rise Of Micro-Schools
AI could make smaller schools more viable
Fewer People, Same Professions
Jobs may survive while teams become smaller
+8 more clips · 15 min 12 sec of audio in all
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
All 11 clips — the highlight moments, each cut as its own audio, with a title and a speaker
All 12 segments — the transcript broken into labeled sections, every ad read marked
All 22 topics — jump to every other episode discussing the same subject
Every related episode — other shows Radar links to this one
No account is needed to search Radar.