© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
GPT-4 Turbo was the key announcement of the OpenAI DevDay conference, which is now more capable and is also cheaper to run. Similarly, the company also announced GPT-4 custom models, which allow large enterprises to train and run their own generative AI models based on GPT-4.
OpenAI’s Custom Models program enables select organisations to work with a dedicated group of researchers from the OpenAI research team to train GPT-4 for a specific use case, which includes additional modifications to the model, specific pre-training, running custom RL-post training, and these organisations will have exclusive access to the custom model.
The custom-trained model will be compliant with the existing enterprise privacy policies and the specific data will not be used to train other models or will not be reused in any context.
According to OpenAI, this program is only available for domains with large data sets of at least billions of tokens, and to train a custom model based on GPT-4, it will take “several months.” and one needs to spend at least $2 to $3 million US dollars to start with.
Businesses and enterprises interested in training their own GPT model can apply for the same via OpenAI’s official website. Do note that this service is limited to select industries and while applying for the same, one also needs to mention details like the reason to train a custom model and the type of data set that will be used to train the model. Right now, this program is limited to select countries, including India.
