I did that last summer, I compared the performance of different english word embedding models, as far as I remember the best ones were GloVe and a few knowledge graph word embeddings.

None of them were better than a human at giving hints for 3+ words though

I tried this many years ago (before LLMs) with hundreds of real human games and it was never that good.

I love this.

Imagine the energy savings if more people didn’t just automatically reach for LLMs for their pet projects.