3 ms·
Another whoa for me >Finally, we highlight surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a la
by famouswaffles 3y ago
Another whoa for me
>Finally, we highlight surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person learning from the same content.
Results - https://imgur.com/a/qXcVNOM https://imgur.com/a/qXcVNOM
- usaar333 3y agoI think this somewhat is mostly due to the ability to handle high context lengths better. Note how Claude 2.1 already highly outperforms GPT-4 on this task.
- a_wild_dandan 3y agoGPT-4V turbo outperforms Claude on long contexts, IIRC. Unless that's mistaken, I'd suspect a different explanation for that task.