0xbt

Another plan, if character-level types are nevertheless infeasible, is to check out to manually encode the know-how of phonetics, at the very least, in some way a single way could be to knowledge-increase inputs by working with linguistics libraries to transform random texts to International Phonetic Alphabet (which GPT-3 by now understands to some extent). I have not been able to take a look at regardless of whether GPT-3 will rhyme fluently given a correct encoding I have tried out out a selection of formatting approaches, utilizing the International Phonetic Alphabet to encode rhyme-pairs at the starting or end of lines, annotated in traces, space-divided, and non-IPA-encoded, but even though GPT-3 knows the IPA for much more English words and phrases than I would’ve envisioned, none of the encodings present a breakthrough in effectiveness like with arithmetic/anagrams/acrostics. Thus considerably, the BPE encoding appears to sabotage efficiency on rhyming, alliteration, punning, anagrams or permutations or ROT13 encodings, acrostics, arithmetic, and Melanie Mitchell’s Copycat-model letter analogies (GPT-3 fails without having spaces on "abc : abcd :: ijk : ijl" but succeeds when space-separated, though it does not remedy all letter analogies and may well or may possibly not improve with priming applying Mitchell’s very own article as the prompt examine with a 5-year-outdated kid).