Last Translation Benchmark Is Live: Contribute 10 Examples, Get Co-authorship

zouharvi · x · 2026-09-04

Last Translation Benchmark is a live paper+dataset still open for contributions. It tackles saturated benchmarks and unreliable metrics by collecting inputs (text, image, audio, video) that provably break state-of-the-art translation models, each with an automatic pass/fail verification rule. Contributors with 10 approved submissions earn co-authorship on the live publication. LTBv1 already has 3456 examples (90MB) on arXiv and a Hugging Face leaderboard, with rolling future releases.

Original post →

More from Research

Research channel →