Salesforce’s new AI Cloud has puzzled many about what makes it different from the competition, what’s new in the service, and whether you should consider subscribing. Analysts predict that only a few people may be interested in the expensive new product.
Salesforce AI Cloud will use the company’s previously announced Slack GPT, Tableau GPT, Apex GPT, MuleSoft GPT, Flow GPT, Service GPT, Marketing GPT, Commerce GPT, as well as the new Einstein Trust layer and to train large language models. Prompt Engineering Tool. (LLM).
According to CEO Marc Benioff, the new bundle focused on generative AI applications integrates Salesforce’s existing technology stack already offered through products such as Einstein GPT, Data Cloud, Tableau, Flow and MuleSoft. is said to be based on
AI Cloud does not have a unified interface
Salesforce promotes AI Cloud as a bundled product via the $360,000-a-year Cloud Starter Pack, which includes a free AI-ready evaluation by Salesforce Professional Services, but the company says it doesn’t have a single, unified interface. said.
“A lot of the AI cloud really shows up as an assistant across the different cloud offerings we have, such as Sales GPT for Sales Cloud, Marketing GPT for Marketing Cloud, and Service GPT for Service Cloud,” said Jayesh, senior vice president. Govindarajan said. President of Data Science and Engineering at Salesforce.
“However, if you are an independent service vendor and want to create your own prompts, AI Cloud offers a separate local console with a low-code experience for training language models at scale,” Govindarajan said. said, adding that Salesforce has already signed on to work with service partners such as Accenture, Deloitte and PwC.
According to Forrester principal analyst Liz Herbert, AI Cloud’s starter pack is just one of the product pricing models the company is pushing, with most companies already having a CRM in place. “Many Salesforce customers won’t use that model or be forced to buy it” and other software suites installed.
“When you look at how a typical customer already subscribes to Salesforce, it’s very unlikely that bundling will become a popular purchase method,” Herbert said, adding that Salesforce wants Salesforce. He added that the pricing element for businesses was also less specific. Use only one or more products under AI Cloud. Herbert said the pricing structure will adjust to usage of AI Cloud’s various services over time.
Salesforce AI Cloud Offers LLM Choices
According to the company, Salesforce’s AI Cloud architecture is built to support multiple LLMs and their training.
The AI cloud architecture is supported by the company’s public cloud infrastructure architecture, Hyperforce, which in turn supports a data cloud layer, followed by a layer that hosts multiple LLMs (proprietary or third-party), which The company’s new Einstein resides on top. Trust Layer said:
The Salesforce Trusted AI Cloud architecture consists of multiple layers.
Salesforce
According to Salesforce, the trust layer separates Einstein GPT-based generative AI engines, prompt engineering tools and prediction engines such as Slack GPT, Tableau GPT, Marketing GPT and Commerce GPT.
AI Cloud’s current LLM offering includes third-party models such as Amazon Web Services (AWS), Anthropic, and Cohere.
The company also partnered with OpenAI to use its API to access models such as GPT-4, which powers ChatGPT.
Salesforce also offers its own LLMs such as CodeGen, COdeT5+, and CodeTF for generative AI implementations in applications, especially those aimed at increasing productivity, bridging talent gaps, and reducing technology adoption costs. increase.
A new division called Salesforce AI Research has been created to evolve new LLMs and further develop existing LLMs to develop new applications for AI.
Companies can also have their own LLMs trained on their own domain-specific data, according to Salesforce.
“These models, whether they run via Amazon SageMaker or Google’s Vertex AI, are directly connected to the AI cloud through the Einstein GPT Trust Layer. It may stay within the trust boundary,” the company said.
This isn’t the first time someone has provided these features. Several other prominent technology vendors, such as AWS and IBM, already offer similar services in the form of Amazon Bedrock and IBM Watsonx.
How do multiple LLMs work together?
Salesforce’s AI Cloud “chooses the right LLM for the right task,” the company says, in contrast to the practice of some major software vendors who claim to combine different LLMs to deliver generative AI applications. .
“The right LLM decision is based on what the system has seen before,” said Govindarajan, adding that the LLM recommendation engine learns from data across all Salesforce deployments.
Govindarajan gave the example of Marketing GPT, which creates landing pages, and said performance data is also stored within Salesforce’s Marketing Cloud, so the system tracks the performance of landing pages.
“If a landing page works really well, another user who wants to create a similar landing page will be suggested the same LLM that was used to create that landing page,” says Govindarajan. .
According to Salesforce, this same logic applies to multiple flavors of Einstein GPT across different Salesforce Cloud products that continuously train your assistant.
“Generative AI assistants such as Slack GPT and Tableau GPT track whether users ultimately follow suggestions.It also tracks edits and changes to generated results. and become more personalized to the user,” said Govindarajan.
Einstein GPT trust layer taps IAM
According to Forrester’s Herbert, Salesforce’s Einstein GPT Trust layer is neither new nor differentiated, and follows the same principles that most other vendors offer when it comes to providing security. . According to Salesforce’s Govindarajan, like most other vendors, Salesforce uses identity and access management (IAM) permissions within the enterprise to protect data privacy and security.
Salesforce highlighted the security features of the Einstein GPT Trust architecture.
Salesforce
“When a user runs a query on the system, the technology stack behind these generative AI assistants searches a database for the attributes they need and has access to, and then the stack builds a semantic search model based on the knowledge graph. The entire company and the employee requesting the query,” said Govindarajan.
“The captured vectorized data is kept within the company’s servers, masked, and then fed to a large language model to produce a result or response. We copy or pull in identity and access management policies that already exist in ,” added Govindarajan.
The architecture of the Einstein Trust layer is built in such a way that once a query is processed, no information contained in the prompt or query is retained, the company said, adding that generated results are checked for toxicity and an audit trail. A log of all prompts is maintained.
availability
AI Cloud services will be available in various stages, with Einstein GPT and Service GPT expected to be generally available in June, according to the company.
Their Commerce GPT will be generally available in July. Other products in the bundle will be piloted for much of 2023 before being generally available next year.
It will be interesting to see how customers react to Salesforce’s new offerings. Salesforce is more of a bundle of existing products than an entirely new platform. But analysts believe the Salesforce AI Cloud will help further strengthen confidence in businesses’ use of generative AI.
“Salesforce recognizes that there are barriers to AI adoption, especially trust. With so many LLMs out there, it can be confusing which one to deploy, so with Salesforce’s AI Cloud, users will be able to plug in the LLM of their choice,” said SanjMo’s principal analyst. Sanjeev Mohan of the list said.
