I've noticed that while Claude 4 does a decent job, it still fails at the simple task of writing poems with 11 syllables per line, which I call the "Hendecasyllabic Test." I'm curious about when AI models will be able to consistently handle this kind of request. Even the so-called o-series models, which have done well in languages like English and Spanish, still struggle with it in other languages. I suspect we might see this capability emerge before we reach AGI, but I'm really interested in any guesses as to when that might happen. Additionally, even though multimodal AIs exist with voice capabilities, they still can't mark pauses between syllables effectively. This seems like a simple task for humans, as we can even count syllables in languages we don't speak. If AIs can't master this, it suggests that AGI might still be a bit further off, or at least that this particular skill is a challenge.
2 Answers
Honestly, I’m totally with you on this! I wouldn’t even manage that, and I’m basically AGI... or maybe just 'GI' 😂 But seriously, it's impressive that humans can count syllables so easily compared to AI.
I think Anthropic might be the first to crack this! They’re doing some interesting stuff with their models, especially around rhyme and completing poems from the end. It’ll be cool to see if they can figure out how to handle syllable counts too.
That’s true, but their current models mainly focus on text and images. Since syllable counting involves phonetics, it might be harder for them without an audio aspect.
Surprising, right? I mean, can't you just adjust the words in your own sentences until you hit that syllable count? It feels like common sense for us!