OpenAI has just released updates for its language models GPT-3.5-turbo and GPT-4, which also power ChatGPT and Microsoft’s Bing. One significant improvement is the ability to process more text and the use of the embedding model which makes the process much cheaper. OpenAI is also emphasizing that security measures won’t be compromised in any way, with user-entered data continuing to be protected and not used to train the models.
Language models are like automatic sentence completion based on probabilities, making them almost useless for some tasks. To solve the issue, OpenAI now provide the possibility of integrating plugins. Developers can now teach the models to access certain functions via the interfaces or external tools. The models, as a result, can separate and perform different tasks. Another feature is the extraction of text such as names, birthdays, locations, and more, which developers can fetch to expand the scope of ChatGPT use.
GPT-4 model is now available and can process more extended text inputs. The new GPT 3.5 turbo 16k provides four times the context length at only double the price. The new model can process approximately 20 pages of text with a single query, making it a more efficient model. OpenAI is planning to completely eliminate the waiting list in the coming weeks. However, the updates only affect paying customers using the API. They are available now, and most models will automatically update on June 27th.
In conclusion, OpenAI’s language models are becoming more efficient and accessible for use. The integration of plugins and extraction of text makes it more versatile. The new models’ cost reductions are also justifiable based on the improved efficiency of the language models.
