Experimental browser for the Atmosphere
In a fact tracing task, we find that classical retrieval methods (e.g. BM25) are still much better for retrieving examples that *entail* factual predictions (factual "attribution"), but TDA methods retrieve examples that have greater *influence* on model predictions.
Dec 13, 2024, 6:57 PM
{ "uri": "at://did:plc:u4swdfntdfizkswnejceoi2y/app.bsky.feed.post/3ld7hmrdfjk2a", "cid": "bafyreifmhmvqnhb4icoy5tdf5vcan3hxbvailxfk7vbtqchqxb3odtlo24", "value": { "text": "In a fact tracing task, we find that classical retrieval methods (e.g. BM25) are still much better for retrieving examples that *entail* factual predictions (factual \"attribution\"), but TDA methods retrieve examples that have greater *influence* on model predictions.", "$type": "app.bsky.feed.post", "langs": [ "en" ], "reply": { "root": { "cid": "bafyreiclbc63wx6wfj2mnxra2deepcdlvsrvlsrow3rygn2miwgvdd6xqu", "uri": "at://did:plc:u4swdfntdfizkswnejceoi2y/app.bsky.feed.post/3ld7hmptt2s2a" }, "parent": { "cid": "bafyreib33zywbbqt5fa4nexx5cp7rudfqkytzxc527afzwvpl5jmo5zxym", "uri": "at://did:plc:u4swdfntdfizkswnejceoi2y/app.bsky.feed.post/3ld7hmqmdak2a" } }, "createdAt": "2024-12-13T18:57:01.184Z" } }