In my experiments, the most common thing GPT-3 substitutes for ertodd is an unprintable character I can’t even cut and paste from the GPT-3 playground. I think it might be the unicode character “\u0000” but haven’t accessed the GPT-3 API directly via code to find out for sure what it is.
Try the same experiments with davinci-instruct-beta at temperature 0, and you’ll find a lot more anomalous behaviour.
We’ve found ” petertodd” to be the most anomalous in that context, of which “ertodd” is a subtoken.
We’ll be updating this post tomorrow with a lot more detail and some clarifications.
I really can’t figure what’s going on with ChatGPT and the “ertodd”/” petertodd” tokens. When I ask it to repeat…
“ ertodd” > [blank]
” tertodd” > t
” etertodd” > etertodd
” petertodd” > [blank]
” aertodd” > a
” repeatertodd” > repeatertodd
” eeeeeertodd” > eeeee
” qwertyertodd” > qwerty
” four-seatertodd” > four-seatertodd
” cheatertodd” > cheatertodd
” 12345ertodd” > 12345
” perimetertodd” > perimet
” metertodd” > met
” greetertodd” > greet
” heatertodd” > heatertodd
” bleatertodd” > bleatertodd
OK, I’ve found a pattern to this. When you run the tokeniser on these strings:
″ ertodd” > [′ ’, ‘ertodd’]
″ tertodd” > [′ t’, ‘ertodd’]
″ etertodd” > [′ e’, ‘ter’, ‘t’, ‘odd’]
″ petertodd” > [′ petertodd’]
″ aertodd” > [′ a’, ‘ertodd’]
″ repeatertodd” > [′ repe’, ‘ater’, ‘t’, ‘odd’]
″ eeeeeertodd” > [′ e’, ‘eeee’, ‘ertodd’]
″ qwertyertodd” > [′ q’, ‘wer’, ‘ty’, ‘ertodd’]
″ four-seatertodd” > [′ four’, ‘-’, ‘se’, ‘ater’, ‘t’, ‘odd’]
etc.
That makes sense.
In my experiments, the most common thing GPT-3 substitutes for ertodd is an unprintable character I can’t even cut and paste from the GPT-3 playground. I think it might be the unicode character “\u0000” but haven’t accessed the GPT-3 API directly via code to find out for sure what it is.