Not at all, the impression I have from students I'm familiar with (Danish bus school, bachelor & master's level). A challenge, for sure, but think most (of our) students realize that using chatbots does not immediately lead to learning. Obviously it is our/my responsibility to teach appropriate use.

This is why it is important to know about how RAGs work - if your chatbot is based on a RAG, it will not only provide answers based on the source material. Hence, no hallucinations. Now, all AI systems are not 100% flawless, but neither is any human.

As indicated in the article, this claim by Chollet still seems relevant. Although, just like LLMs 'cracked' language, O1 might be able to 'crack' certain domains.

Just curious - is the reason why you categorize this as selection bias (and not, primarily, as collider bias), because that is just the way it is referred to, in your field? or is there a more particular reason.

The original post by chehendriksen.bsky.social‬.

You'd struggle to find any Danish academic that disagrees. Government decision. Cheaper to get people through to 50% teaching load / job market more quickly, I guess. (tbf, as you indicate, in practice the 30 ECTS points are probably more like 3-4 months).

In Denmark you need 30 ECTS in a 3 year PhD program (also 6 months of teaching activities)

is how often participants (e.g. at a university) talk about the treatment. So, if ones study runs over a few weeks, might the effect be different in the beginning of a study, compared to later. Maybe. Would hope someone has considered this (we are implementing a dummy, to control). 2/2

He. To some extent, yes. I mainly do lab (onsite, online) experiments so many challenges don't apply. Was interesting to see such a detailed report of how major the challenges are in field research (in terms of randomization, baselines, drop outs etc.). One thing I've wondered in social lab-stuff1/2

I didn't know the - very good - Krauss paper. Thank you for the tip.

Maybe - but at least we can then agree, that it aint easy :). (I have been involved in IRB related issues for quite a few years, btw.).

I am quite confident that if I had written that post, I would not have received an email from anyone :) - influencing an IRB can also be quite difficult, in particular if shared across disciplines.

Even if we argue that we knew it will work - finding out the magnitude is still interesting, is it not? Need empirics for that. In some fileds the variance of effect sizes will probably be bigger.

Let's say we had had a "Latour-style" ethnography of certain psych teams in the 2010s. Reasonable chance we more quickly could have realized the extent of flawed practices? I guess you are more focused on if Phil of Scie should do it. I care less about that, would just like this kind of fieldwork.

If one knows what people are doing (wrong), one is better able to show and explain how to do it right? In any case, I don't think Latour's work (e.g.) is about changing how people do science. It was about understanding how science actually proceeds - no normative implication needs to be drawn.