Search papers, labs, and topics across Lattice.
Affiliation:
6
0
9
7
This work shows that specialist optimization implicitly selects from this latent trajectory space, and establishes a new view of specialist training: when gold reasoning is absent, tuning choices directly control the latent supervision passed to downstream models.
Machine translation benchmarks have functionally saturated, but pairing human-authored failure cases with deterministic verification rules reveals critical multimodal blind spots that automated metrics consistently miss.
The top systems in hallucination detection outperformed baselines by up to 40 points, revealing significant advancements in tackling errors in vision-language models.
LLMs trained on a synthetic corpus can outperform native data benchmarks while using significantly fewer tokens, challenging the assumption that more data always leads to better performance.
No single model can reliably assess translation quality across all languages, highlighting the need for a nuanced, direction-aware approach to multilingual data evaluation.
Domain match and language relatedness trump joint vocabularies for effective knowledge transfer in multilingual NMT.
