“Accuracy measurements using the BLEU algorithm – a way to evaluate the similarity of an original text to a machine-translated output by giving it a score between 0 and 1 – put the new tech at about 0.4.”
UTS researchers put an EEG cap on 29 people, had them silently read text and decoded about 0.4 BLEU of it. The model gets verbs and swaps nouns for synonyms, so “the author” comes out as “the man.” The work is a preprint with no peer review yet. The headline says mind reading. The data says a noisy guess at what you were reading.