hey! i have been accepted into Anthropic hackathon this weekend. my plan is to attempt to reproduce
https://github.com/bbaaii/DreamDiffusion
but for text. ive been told that it is not possible to "get text out of eeg data" but i will still try.
i would love to get some feedback beforehand on my strategy:
record a dataset of me thinking about pre-made sentences, for example "i am walking the dog in the park", "apples are usually red but sometimes green", etc.
so i'll have a dataset of pairs of text <-> eeg
i'll then use
![[Pasted image 20230728144109.png]]
10 - 12 - record brain
1. read a sentence 2 seconds
2. focus on your breath 5 seconds
3. say the sentence in your mind 3 times slowly
4. repeat
12 - 2 - data
2 - 4 - model
4 - 6 model
10 - 12 - finish or b plan failure + presentation
- [[B plan]]: show the poetry thing or claude presenting my failed experiment
12 - 2 - improve a bit - presentation
2 - 4 - presentation
4 - 6 - presentation NOW