Livio Baldini Soares
LightNews LightNews
liviobs.bsky.social
Livio Baldini Soares
@liviobs.bsky.social
11 followers 64 following 3 posts
Research on Natural Language, LLMs, Knowledge Graphs and Retrieval. Previously, research on systems software. Currently @ Google Deepmind, NYC.
Posts Replies Media Videos
liviobs.bsky.social
RICHES supports multi-hop retrievals, interleaving thoughts to plan the next steps. We demonstrate strong performance on various ODQA tasks, including single & multi-hop question answering.
Tables 1 and Table 2 from the paper. Table 1 caption: Example of RICHES vs Dense Retrieval for single-hop QA. Only the retrieved text is shown for illustration. Table 2 caption: Example Iterative retrieval outputs from RICHES. Remarks are annotated as (# Comments).
November 12, 2024 at 6:19 PM
liviobs.bsky.social
RICHES retrieves documents by directly decoding their contents, constrained by the corpus. This unification unlocks adaptability to diverse tasks via prompting alone! No extra training needed.
Figure 2 from the paper depicting 4 different time stamps of the decoding process using a beam. The figure caption states: "Visualization of constrained beam for query: "when did marathon change its name to snickers?". The final RICHES output is "Marathon was renamed Snickers in 1990". Bold boxes track the progress of the top-beam sequence. Grey crossed out boxes are sequences that the LLM preferred, but were blocked by corpus constraints".
November 12, 2024 at 6:19 PM