Bender and Koller argue a system trained only on text cannot learn meaning. Use Neuronpedia’s free interface to Gemma Scope to inspect features for concepts that seem to need grounding (colours, spatial relations) and check whether they respond across very different wordings.
It connects hands-on interpretability with a well-known argument about language models, and it runs in a browser.
Output: A feature analysis and essay.
Argue whether a stable, wording-independent internal feature is the kind of evidence that answers the octopus argument, or whether it misses Bender and Koller’s point about reference.
Write your question in one sentence, and what result would surprise you.
Do a 30-minute literature check — if someone has done part of it, extend their work instead of repeating it.
Pilot with 20–50 prompts or examples before you scale anything up.
Record everything in a table and read real examples, not just averages.
Write up: question, method, results, limitations, and what you would do next.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Use Neuronpedia, a free browser tool, to see which internal “features” of a language model light up for different ideas, then try to break your own explanation of one.
Lange argues AI outputs should count as one input alongside our own reasons, not replace them. Run a small study: show volunteers AI answers (some reliable, some subtly wrong) and measure how much they update.
Ask several models the same questions about their inner states while varying wording, persona, system prompt and language. Measure how much the answers change.
