Developers can teach that ChatGPT will focus on tasks like tweaking code or to match the tone agreed with an individual company.
OpenAI has allowed companies to adapt the GPT-3.5 Turbo with their training data, providing targeted chatbots for specific tasks. For example to respond to customers using less precise formulations.
So far, customization has been limited to GPT-3 variants such as davinci-002 or bebbage-002.
GPT-3.5 Turbo is a package of model models made available by OpenAI earlier this year that means that they are ideal for non-chatbot use. The model is capable of processing 4000 tokens at the same time, double the capacity of previous versions. The company added that testers used 90.9 l instead of 3rd.
GPT-3.5 costs 0,008 for the equivalent of 1,000 training tokens, 1,020 for 1000 input tokens, and 2,020 for 1,000 output tokens.
The model has very high levels of data, as could be used as GPT-4, if they’re not up to September 2021. The same goes with company data. OpenAI says that this information won’t be used to train models outside the client company.
The company says that an additional tuned GPT-3.5 Turbo will be able to substantially exceed the GPT-4 capabilities. That latter (which accepts a photo and text) will open for customization this fall.
After the release of GPT-4 in March, OpenAI announced it will implement a phased update of existing models, instead of creating new ones from scratch. At the recent Economic Times meeting, President Sam Altman confirmed that OpenAI isn’t training GPT-5, because there is a lot of preparatory work to do before launching a new language model.