OpenAI released its first open-source language models, gpt-oss-120b and gpt-oss-20b, which appear to follow Microsoft's Phi model approach of training on synthetic data. While these models perform well on benchmarks, they may underperform in real-world applications due to limited out-of-domain knowledge. The author suggests

5m read timeFrom seangoedecke.com
Post cover image
Table of contents
Phi models and training on synthetic dataSynthetic data is safer

Sort: