Corporate Market Expansion Specialist
Posted: Mon Jan 06, 2025 5:43 am
We have seen this issue before. I will tell you why I raise this issue. For example, social media companies have not invested adequately in their content moderation tools and resources for non-English languages. I share this not only out of concern for non-US users but many US users prefer languages other than English when communicating. So I am very concerned that social media will repeat the same mistakes in AI tools and applications. Q. Sir and Madam and how to ensure their linguistic and cultural inclusion in large language models is a key area of focus for your product development. Unfortunately Senator Padilla is starting from the position that he wants to moderate non-English languages and therefore asked about support for other languages.
Sam Altman: We think that's really important. One example is we're israel whatsapp phone number working with the Icelandic government to make sure their language is included in our model. Icelandic is a relatively underrepresented language, and it's underrepresented compared to many of the languages represented on the internet. We've had a lot of similar conversations. I'm looking forward to having similar partnerships with many of the less-resourced languages to include them in our model. - Unlike our previous model, which was good for English and not so good for other languages, now - it performs pretty well on a large number of languages. You can go down the list ranked by number of speakers and still get good performance.
But for these very niche languages we're excited to work with custom partners to include that language in our model runs. We're also focused on the part you asked about values and making sure culture is included. Have you heard about the opening of an office in Japan? Maybe that's the custom partnership part. Summary Reviewing our exploration of language representation and efficiency in large language models (such as ML), we came to several key conclusions. English Dominance English remains the most effective language for prompting large language models such as ML because it has broad k coverage in the model vocabulary. This dominance highlights the practical advantages of using English in prompting engineering.
Sam Altman: We think that's really important. One example is we're israel whatsapp phone number working with the Icelandic government to make sure their language is included in our model. Icelandic is a relatively underrepresented language, and it's underrepresented compared to many of the languages represented on the internet. We've had a lot of similar conversations. I'm looking forward to having similar partnerships with many of the less-resourced languages to include them in our model. - Unlike our previous model, which was good for English and not so good for other languages, now - it performs pretty well on a large number of languages. You can go down the list ranked by number of speakers and still get good performance.
But for these very niche languages we're excited to work with custom partners to include that language in our model runs. We're also focused on the part you asked about values and making sure culture is included. Have you heard about the opening of an office in Japan? Maybe that's the custom partnership part. Summary Reviewing our exploration of language representation and efficiency in large language models (such as ML), we came to several key conclusions. English Dominance English remains the most effective language for prompting large language models such as ML because it has broad k coverage in the model vocabulary. This dominance highlights the practical advantages of using English in prompting engineering.