3 ms·

Did you see the graph benchmark? I found it quite interesting. It had to do a graph traversal on a natural text representation of a graph. Pretty much your prob

Did you see the graph benchmark? I found it quite interesting. It had to do a graph traversal on a natural text representation of a graph. Pretty much your problem.

stopachka 9mo ago
Oh, interesting!
stopachka 9mo ago
Update: I took a corpus of personal chat data (this way it wouldn't be seen in training), and tried asking it some paraphrased questions. It performed quite poorly.
- abraxas 9mo ago
  Which models did you try?
  - stopachka 9mo ago
    Claude Sonnet 4.6