Salesforce joins the ranks of generative AI with Einstein GPT
The Californian publisher gives in to the temptation of generative AI. Einstein GPT combines Salesforce's proprietary AI for CRM with that of OpenAI. Its model is trained with public data, but companies can feed it with their own data.
Salesforce took advantage of its Trailblazer DX developer conference in San Francisco on 6 and 7 March to announce its generative AI. Einstein GPT is, as its name suggests, a combination of the editor's proprietary Einstein technology and ChatGPT-type models. The Californian company will enter into partnerships with various suppliers, starting with the best-known, OpenAI, the originator of ChatGPT, Dall-E and Whisper.
Salesforce has integrated Einstein GPT into its business applications for sales, service, marketing, sales and IT. The solution is already interfaced with Mulesoft, Tableau and Slack. For example, a sales team will use it to identify new prospects and relevant people to contact, whether or not they are already in the company's CRM. The AI will then suggest to the sales team that they start a conversation with their target by email, based on previous exchanges, among other things. Employees can directly edit these pre-written texts, but can also ask the AI to rewrite its messages in a lighter tone, for example. Unlike consumer AI such as chatGPT, interactions between "humans" and Salesforce's generative AI are integrated directly into the user interface of business applications,
A human in the loop
The trauma of cognitive bias in AI does not spare Salesforce. Clara Shih, corporate vice president and general manager of Service Cloud, was quick to quote the now hackneyed Spiderman line: "With great power comes great responsibilities". The Einstein GPT teams are working closely with the publisher's ethics department, without giving further details. Moreover, Jayesh Govindarajan, senior vice-president of AI and machine learning, has repeatedly stressed the importance of systematically keeping a "human in the loop". This is as much about ethics as it is about the ability to verify the ownership of the data used or the relevance of the results of queries sent to Einstein GPT.
Like OpenAI, Einstein GPT trains with public data. But not only that. It can also draw on databases that companies subscribe to. As Jayesh Govindarajan explained at the press conference, Einstein GPT will use "large language models (LLMs) such as GPT-3, applied to semi-public paid databases, such as weather for example". But above all, following on from Einstein, which has been available since 2016, it is a generative AI optimised for CRM that Salesforce has developed.
Training the model with the company's data
A company can therefore choose to train the AI with its own data from the Salesforce Data Cloud. This is a way of obtaining personalised content, adapted to the specific context of the company. "Customers can connect this data directly to OpenAI's advanced AI models or opt for another external model and use a natural language prompt directly from Salesforce CRM to generate content that can be adapted in real time to the increasingly changing behaviours of consumers," explains Salesforce in its press release.
Automatically supplementing developers' code
In addition to its business applications, Salesforce is also making its generative AI available to developers. According to the publisher, its AI will draw its knowledge from a company's "in-house" programming by training its model on the lines of code already written in the company. "The principle is the same as Google's auto-complete suggestions," noted Jayesh Govindarajan. Einstein GPT will offer the developer a portion of code to continue their programme.
The Californian also took the opportunity of its announcement to confirm the integration of ChatGPT into Slack (see below). "This app includes AI-based conversation summary creation, search tools on any and all, and message writing assistance." Finally, Salesforce is creating a new $250m Salesforce Ventures fund to support "generative AI startups and create a responsible ecosystem in the field".