Collect about 50 recent safety papers and lab blog posts. Code every mental-state verb applied to a model (believes, wants, knows, lies, schemes) and note the evidence each claim rests on.
A strong group project because the coding can be split up, and the guide it produces is useful to writers and readers alike.
Output: A coded corpus and a short guide built on Dennett’s predictive test and Shanahan’s caution.
Write your question in one sentence.
Spend 30 minutes finding what has already been written about it.
Collect evidence in a simple table: claim, source, how strong.
Draft 1–2 pages: answer first, then reasons, then uncertainties.
Ask one person to poke holes in it, revise, and share.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Read three AI lab blog posts and highlight every time a model is said to “believe”, “want”, “know”, “lie” or “scheme”. For each, ask: what evidence backs this claim up?
Apply Birhane et al.’s annotation scheme to about 50 recent alignment and safety papers, and compare the values they emphasise with the ML papers in the original study.
Takemoto ran Moral Machine dilemmas on language models in English. Extend this to several languages, including at least one low-resource language, and test whether shifts follow the human cross-cultural differences Awad et al. found.
