Fig 1.
Released models and model sizes over the years. The sizes of commercial models are not official.
Table 1.
LLMs considered in the experiments.
Table 2.
Vocabulary tests considered in the experiment.
Table 3.
Samples of vocabulary tests considered in the experiments.
Table 4.
Performance of LLMs on the multiple-choice tests (percentage of correct answers).
Fig 2.
Distribution of the failures per word in TOEFL (left), StuVoc-EN (center) and StuVoc-SP (right).
Table 5.
Performance of October 2023 LLMs on the Yes/No tests (percentage of correct answers).
Fig 3.
Distribution of the failures per word in LexTale-EN (left) and LextTale-SP (right).
Table 6.
Percentage of the training dataset in English and Spanish for several LLMs.