1. can I train GPT with my own data?
Yes, you can train GPT with queries and the expected responses. An example from our domain would be to train GPT with product information and technical data for use in customer service as first level support or as a chatbot for your own website.
2. can I host chat GPT myself?
Since the beginning of this year, Microsoft has provided customers with the ability to host OpenAI tools such as ChatGPT, Dall-E 2 and GPT-3.5 on their own Microsoft Azure servers. This means that companies have full control over their voice models and do not have to rely on OpenAI's infrastructure.
3. Will my data be used to train others' language models?
No, when you have your own instance in Azure, the data remains under your full control. The data is not shared with OpenAI. In Azure, organizations can run their data and applications in a secure and controlled environment.
4. What is the cost of hosting in Azure GPT?
The "Azure OpenAI Service" can be run in Europe (Amsterdam), among other locations. The prices can be found here: Azure website
In our examples, we used the "Davinci" model. You pay $0.02 per 1000 tokens, where one token is roughly equivalent to one syllable.
5. How is it different from other translation tools like DeepL?
Our experience shows that DeepL is still qualitatively better than the GPT language model in the area of pure translation. What makes GPT better is the possibility to set tonalities. This allows you to create and translate texts specifically for certain target audiences and countries. For example, one can have a text translated specifically for the American market or with a focus on certain features. With DeepL Write, DeepL is also trying to move in this direction. If you like, you can try it out for yourself at DeepL Write.