A recent arXiv preprint extends earlier findings on decoding language from brain signals. The authors previously demonstrated that a decoder could recover open-vocabulary word-level information from 19-channel EEG recorded during silent reading, by aligning short EEG windows with hidden-state embeddings of the presented word.

The new work, titled "Deep LLM layers expose passage-level context in EEG during silent reading," appears to build on that result. Based on the abstract, the study investigates how deeper layers of large language models correspond to EEG activity, moving beyond single-word alignment to broader contextual information.

Because the abstract is truncated, the full methodology and quantitative results are not yet available. The source is a single preprint on arXiv, so the findings have not undergone peer review. Still, the reported direction suggests that EEG may carry signals of not just what word is read, but how its meaning is shaped by the surrounding passage.