What AI boosters in higher ed do not understand is that if you replace feedback to students with AI and encourage students to use AI for ‘research’, you’re poisoning the well. Even *if* it were good at either thing now (spoiler: it’s NOT), it depends on examples to copy. medium.com/@gillesdeper... - ThreadSky

willpooley.bsky.social • 85 days ago

What AI boosters in higher ed do not understand is that if you replace feedback to students with AI and encourage students to use AI for ‘research’, you’re poisoning the well.
Even *if* it were good at either thing now (spoiler: it’s NOT), it depends on examples to copy. https://medium.com/@gillesdeperetti/ouroboros-of-ai-the-peril-of-generative-models-feeding-on-their-creations-bda62e47ac5f

Comments

willpooley.bsky.social•85 days ago

if you are throwing yourself headlong into ‘generative AI can do parts of my job that I don’t like’ you may get what you wish for in the short/medium term (unemployed). But once they’ve sacked 60% of staff, where will the models draw new research/latest pedagogy from?

willpooley.bsky.social•85 days ago

Incredibly dystopian tbh that *this quickly* it has become clear that we might actually need a defense of the value of the human, let alone the ‘humanities’. Hurtling towards a near future where the robots mark the robots work and destroy the planet to do so. Where are you planning to be then?

willpooley.bsky.social•85 days ago

The other development I’m personally dreading* is when genAI is so embedded in student research processes that they *write like AI*. That it won’t even matter if they’re using AI to ‘cheat’ by our own definitions, because they are behaving like an AI. (*is there evidence this is already happening?)

willpooley.bsky.social•85 days ago

same btw goes for researchers: if we get comfortable using it as a natural crutch, at some point we cannot do *without*. And we don’t understand what we ourselves are doing. Hard to defend the idea of specialism or expertise then isn’t it

willpooley.bsky.social•85 days ago

apols I really don’t want to become Bluesky’s minor humanities AI prophet of doom, but it’s like watching mold spread in a damp building. the fix once it’s set in is going to be very hard, but the rot depended on environmental issues in the structures

willpooley.bsky.social•85 days ago

Really I need to blog about this again rather than clogging the TL, but now I am going to try and spend some of today doing my job.

willpooley.bsky.social•85 days ago

good morning!

dirtbagbistro.bsky.social•85 days ago

spavel.bsky.social•85 days ago

Biologists training AI in tax accounting sounds about right

dirtbagbistro.bsky.social•85 days ago

lol. Every time I come across these I wonder how easy it is to spoof credentials in the application process

keirharper-thorpe.bsky.social•85 days ago

Yes, this is interesting. It takes a pro-AI stance in arguing that the training needs to be improved with human-curated resources for the AI to learn on. However, it is useful for us to alert to students, who have a tendency to over-trust AI output, that the quality is liable to deteriorate further

keirharper-thorpe.bsky.social•85 days ago

in coming months and years. While this article points to how the 'bias' leads to decreasing fidelity in terms of images, surely we need to alert to how the bias is liable to increase in outputs, exacerbating the challenges we are already seeing in terms of gender and racial representation.

reallyquitetired.bsky.social•85 days ago

The fact that the companies that built the big genAI models didn't write in an iron clad way of distinguishing the content they produce shows how recklessly irresponsible they are and therefore how suspicious we should be of their products.

robskinner.bsky.social•85 days ago

Couldn’t access the Medium piece, but I think it’s worth reading the article in Nature that it cites: https://rdcu.be/d2E2F

robskinner.bsky.social•85 days ago

The doomy title speaks, though, to the maths at work here rather than apocalyptic visions of digital mould. In short, I think this strengthens the argument that we need to teach about (and with) AI in order to properly prepare our students for the work beyond the historiographical essay.

robskinner.bsky.social•85 days ago

The authors conclude that we need ‘community-wide coordination to ensure that different parties involved in LLM creation and deployment share the information needed to resolve questions of provenance.’ That sounds to me like … archivists, right?

Comments

Posting Rules

Reply