May 25, 2022

As we begin to see AI in business, our interaction with machines is starting to change. Companies like Salesforce are exploring new ways to use AI to impact customers more directly.

While using AI is certainly helpful in finding customers who are most likely to leave or buy, it is certainly a step in the process and just the beginning of how AI will work for us in the future.

Salesforce’s AI journey began in 2016 when the company launched Einstein, its AI platform. In fact, Einstein was never conceived as a standalone product, but rather as a collection of intelligence, with the ability to touch every aspect of the Salesforce stack. The original team that brought it to life is largely gone, but work continues.

A year ago, the company named former Stanford professor Silvio Savaris as its chief scientist. One of the reasons he was willing to leave the academy was his ability to conduct cutting-edge research with vast datasets, a large staff, and the resources of a company like Salesforce.

He said he wants to continue the research he has been doing over the past two decades to make the skills available to people with no special training. “One of the main areas I’m pushing here is to use AI in new ways to empower people in business, and I’m thrilled to be able to unleash that power with so many simple options that anyone can use.” Savares explained.

To achieve this overarching goal, one of the key initiatives he and his research team of 100 have pursued is an approach to voice programming that the company has called CodeGen. The idea is to let people describe what they want to do in simple, colloquial language, and have AI generate code based on natural language instructions.

But it’s not just about telling AI technology what you want; Savarese said it was more like a conversation. “CodeGen really offers a new way to develop software. Instead of writing code directly, users in a conversation simply describe the problem they are trying to solve in plain English. So part of the conversation is very, very important,” he explained.

What he means is that you can ask for anything and the AI ​​will ask for clarification and walk back and forth, like in the example in the Salesforce blog post explaining CodeGen:

An example of coding a Salesforce conversation using the CodeGen tool.

An example of coding a Salesforce conversation using the CodeGen tool. image credit: sales department

Although it is still in experimental development, they are making progress in creating models suitable for two different audiences. “The goal is to appeal to multiple users. One is the more experienced developers, who in this case CodeGen will help them handle the manual parts of coding and processing, the parts that aren’t interesting from a coding point of view. Second users are people who have no programming experience, so they have almost no programming experience, but CodeGen can still give them the ability to create software to solve real problems,” he said.

Salesforce is trying to achieve something with conversational coding that hasn’t been done before. While Microsoft is working on something similar to the GPT3 framework, Savarese calls it deep learning extensively and includes extremely complex models.

“This is a fundamental coding model, so CodeGen is built on a huge autoregressive model with 16 billion parameters that trains on a very large amount of data,” he said. Here he breaks down the use cases with model examples depending on whether the user is experienced or not a coder.

While the project is still in the proof-of-concept stage, the next step is to release it to the internal developer community at Salesforce, which will happen when Savaris presents its findings at an internal conference later this month.

If the project moves beyond the experimental stage, the idea would be to allow data scientists and business analysts to code on top of data with Tableau, a company that Salesforce bought in 2019 for about $16 billion, which is a more affordable commercial approach. .

Voice-enabled coding may be the first step here, as other features such as content creation, website layout, and other features can also be reduced to their simple description. “The inspiration comes from the need for an easier way to interact with AI systems and the ability to use language to improve communication to inform certain processes.”

Leave a Reply

Your email address will not be published.