https://openai.com/research/gpt-4
argument for being given access
My specific interest is in the field of poetry. Presently I have opened a discussion with Alex Rudnicky about prosody and Whisper, because I feel I may have something to contribute in terms of human breath-emulation, but as my idea of a learning focus for pauses based on duration-strong punctuation notation would have equal effect on both speech learning/recognition and learning from written poetry, if I am not mistaken.
I am a poet through 40 years. Music of Pauses as a notational system works for poets, who are musically inclined, as reading poetry this way respects all pauses as different durations, thus making music of such text – whether it is intended or not. It differs a lot from poets and readers, who only concern themselves with mental information – and therefore have no regard for indicated line changes. For me and “mine” pauses emulates the body, because it is extremely conscious of breath – how much breath is actually possible to house in an unbroken line. A very concrete example: When I had quit smoking, I wrote a poem containing a very long line that I could read in one inhalation. When I returned to that line many years later, now an active smoker again, I could not read it aloud in one inhalation. But besides such very obvious “who’s got the biggest lungs” competition, experienced use of breath gives the poet an awareness of tiny changes made possible in the experience of the text by breaking lines in unexpected places, thus adding 1 or 2 extra layers of interpretation to the text.
One day, when ChatGPT 8 knows similis and metaphors and subtext very well, having a system of pauses not based on analysis of textual parts, but indicated by the poem as a way to emulate the poet’s body and importance put to breathing – used much to dramatic effect or sparingly to maybe even greater effect – it might be able to analyse poems on a whole new level. And maybe even good novel prose too.
best, kk