The really open NLP fightback

Large language models have become the property of big tech firms. Open science is starting to fight back.

Large language models like GPT-3 have shown advanced performance across a range of language tasks. Howver, the ironically named Open AI recently signed an exclusive deal with Microsoft over access to its GPT-3 model.Other tech fimrs are developing thier own closed models.

These models are very effective, but it is difficult for researchers to assess how they perform with respect to ethical considerations such as biases including racism, sexism and abusive ideas.

NLP startup Hugging Face is now coordinating a project to build a set of large multilingual language models that are fully open and available for researchers to study.

🛎️ Why this matters: Science and software work best when they are open.

📖 Read more (2538 words) 📖