OpenAI has recently announced updates to both their LLM and embedding models. The LLM version has been adjusted to version 0125, and they have also reduced the price for certain components.
The previous embedding model used by OpenAI was the Ada model, which has been in use for a long time. It featured a vector size of up to 1536 dimensions. However, in the text-embedding-3 model, vector sizes ranging from 256, 512, 1024, 1536, to 3072 dimensions are available. Even the smallest vector size performs better than the original Ada model.
The GPT-3.5 Turbo model has been updated to version 0125, fixing a bug where functions could only be called in languages other than English. On the other hand, the GPT-4 model has addressed the issue of being “lazy” and not working as instructed. It has also solved the problem of responding in languages other than English.
In addition to these updates, there are also minor announcements. GPT-4 Vision will be made available for general use in the coming months. The models for detecting unsafe content will be improved, and there will be the addition of a limited usage API key generation feature.
TLDR: OpenAI has updated their LLM and embedding models, reducing prices and offering higher-performing options. GPT-3.5 Turbo and GPT-4 have been updated with bug fixes and improvements. GPT-4 Vision, enhanced content detection models, and a limited usage API key generation feature are also coming soon. Usage statistics on the dashboard will now be displayed based on API keys.
Leave a Comment