- teaching machines to think like humans
- all-in on LLMs, 400 people working on this
- Zhipu's GLM models vs OpenAI GPT models
- GLM (auto-regressive blank-filling) vs. GPT (generative pre-training)
- Tokens: a mix of English and Chinese models
- General Language Model (GLM)
- Training stability
- Tradeoff between stability and efficiency