Fig. 6: Global SOTA improvement map for NLP.

Vertical dashes represent ‘anchors’, i.e., first results establishing a new benchmark. Diamond-shaped icons represent gains in a SOTA trajectory. Icon colors represent the relative improvements in SOTA for a specific benchmark as described in Fig. 5. Each task may contain data on multiple benchmarks, which are superimposed. Benchmarks containing fewer than three results at different time points and AI tasks that would contain only a single icon are not displayed. Detailed information for each data point (such as benchmark names) can be viewed in the interactive online versions of these figures at https://openbiolink.github.io/ITOExplorer/. A similar plot for computer vision, as well as plots aggregated by high-level task classes, are available in the supplementary figures and interactive online material.