Salesforce Einstein GPT for Service

A demo video of Salesforce Einstein GPT for Service

[ZacharyJTan]

0:29 Einstein GPT called the customer by the wrong name, Alan, instead of Arthur lol

[hartnady]

What is the back-end LLM? Either Salesforce has been secretly building it's own transformer based LLM in the background along side the Convolutional neural network that is Einstein Discovery and keeping it a big secret, investing millions of dollars into GPU server farms to train the model to set 8 billion weights and biases and then spending 40+ weeks of human reinforcement feedback training to perfect it, as OpenAI, Anthropic, HuggingFace, and the rest of them have been doing (albeit publicly) OR it is using an undisclosed open-source LLM (like Meta's) and hopefully running it on its own Cloud infrastructure so as to keep straight with the core marketing message that "your data isn't leaving Salesforce". Why is it so impossibly difficult to get transparency on the infrastructure and systems Einstein GPT runs on? Why is the technical stack so hush hush and why are there no developer resources available for a Product that is "GA" (generally available) and costs $50/user/month. Access to OpenAI's GPT APIs costs nothing in terms of a monthly subscription (you are only billed for the tokens you consume). What gives? The marketing and promotional material seems very smoke and mirrors. Salesforce, please be transparent in terms of what LLM is being used on the back-end for Einstein GPT for Sales & Service. CodeGen is has been transparently declared as the back-end LLM for Einstein GPT for VS Code, but nothing has been said about Einstein GPT for Sales and Service. Why?