In this newsletter, the author highlights the release of the Llama 2 base and chat models, as well as CodeLlama, the latest additions to the open-source AI large language model (LLM) landscape. They discuss the leaked details of the GPT-4 model and its performance over time. The author also mentions OpenAI’s new finetuning API for the GPT-3.5-turbo model and addresses the ongoing debate around closed, proprietary AI systems versus open-source models. They highlight the contributions of the open-source community and discuss innovations such as Llama-Adapters, LoRA, QLoRA, and the NeurIPS LLM Efficiency Challenge. The author notes that Llama 2 is a promising model but emphasizes the need for finetuning to improve its performance on specific tasks. They also mention the release of Code Llama, specialized models for code-related tasks. The author concludes by discussing the implications of the finetuning service for ChatGPT and the challenges of adapting models to specialized domains.
https://magazine.sebastianraschka.com/p/ahead-of-ai-11-new-foundation-models