Totally agree we need to push back on the "uncontroversial thingness of AI" as Lucy Suchman puts it so poetically in doi.org/10.1177/2053... It isn't one disembodied thing, but many different artifacts from different companies each with specific goals and agendas. Context is everything.

I agree 💯 but it's even more nuanced than that. When people say "generative AI" they usually mean OpenAI or handful of other companies. LLMs come in all shapes and sizes from all kinds of groups, yet we usually use it as a synonym for ChatGPT.

For instance, there is a universe of small models that can be run on a laptop (w/o calling externally hosted services), e.g. ollama.com/libraryhackmd.io/@opensourcei...

I'm very particular about how students write code; concise & semantically meaningful ('self-documenting'). (The tidyverse team did a great job on this, even if existing users objected to the frequently revised syntax). LLMs can give a starting point but need coaxing & edits to get there...

I definitely appreciate these concerns but I think it's important we help students understand them better. Not all LLMs are the same in terms of energy use or copyright, etc. Nor is AI unique in the space of things we use with exploitative labor practices & environmental destruction

Will be curious to hear as well! Definitely skeptical of LLM writing, but we'll be exploring this in ESPM-157 this semester, probably more on assisting with code and some technical writing than writing-intensive elements.

meanwhile, what fraction of their course-work or assessments can be successfully completed by AI? I think there is room enough to critique both our instruction and the VC hooey.

Agreed, but I think it is too easy to blame this failing on new media for merely doing what they do and corporate fundraising doing what they do. Science owes them a better critique than "sure it's published in the 'best' journals but we all suspect something is off there"

but the code has been public since 2017, & the index much longer than that. If anything that only underscores our reluctance to go beyond sniff test and engage deeply and collectively in building robust analyses.

Right, very much so! The authors cite numerous examples, and rev 2 addresses this: "my read of the conservation community is that they tend to not trust the LPI and see it as badly exaggerated without knowing why"