3 days ago
Hi HN! The idea: you see START and END words, and guess the path that AI chose to connect them. Example: OCEAN → ??? → ??? → ANTENNA Answer: OCEAN → WAVE → SIGNAL → ANTENNA. "Wave" bridges physical (ocean) to abstract (radio), then to antenna. These cross-domain jumps are where it gets tricky. Built with Next.js. Would love feedback on the difficulty curve.
2 days ago
That's quite fun, I wish it had more information about which model took that path and the inference / sampling parameters.
2 days ago
Thanks! Quick overview: Paths are deterministic, not LLM-generated. I use OpenAI text-embedding-3-large to build a word graph with K-nearest neighbors, then BFS finds the shortest path. No sampling involved. The explanations shown in-game are generated afterward by GPT-5 to explain the semantic jumps. Planning to write up the full architecture in a blog post - will share here when it's ready.
2 days ago
Oh that makes a lot of sense, I'm glad it works that way actually - the explanations afterwards left me wondering if it was truly explaining the connections or if it was inferring what they would be (leading to a problem a bit like how "thinking" doesn't actually show the real connections to get to an answer) I'm glad it's not doing that. Neat game and learning opportunity. (Sorry for not wording that very well - long day!)