Pelonomi Moiloa is the CEO and co-founder of Lelapa AI, a South African startup building artificial intelligence tools tailored to local languages. Lelapa aims to provide an alternative to the AI infrastructure being developed in Silicon Valley.
This interview has been edited for clarity and brevity.
Why is it important for developing countries to have their own language models?
Language models created elsewhere lack an understanding of the local context. They can harbor perspectives that can be damaging to the communities that use them. If there is anyone who should benefit from profit generated from a language — a cultural heirloom — it should be the people to whom that language belongs.
How challenging is a lack of data for African languages?
It is a big issue, but not as big as we think. We just have not gotten to the point where we can deal with it yet. Machine learning in the language space is an extremely inefficient process, and there is a lot of room for machines to be smarter and to learn from less.
Why has Lelapa placed resource efficiency at the core of its mission?
We believe that the current landscape of developing AI is an unsustainable one that drives resource exploitation. We do not want to contribute to that.
Creating models that require fewer resources to train democratizes access for users and developers, which allows the benefits of the technology to be shared more broadly, rather than only being available to a select few who can afford the exorbitant costs.
If you could change one thing about how ChatGPT deals with African languages, what would that be?
For ChatGPT to partner with local partners in order to expand their offerings to local contexts. This is something we are working on.
