Salesforce's B2B AI Strategy: Security and Reliability are Paramount

The biggest pain point in the process of generating AI services for B2B customers is believed to be security issues, according to Salesforce.

At the annual customer conference Dreamforce, Salesforce announced two major AI updates: 1) a more secure and reliable chatbot called Einstein Copilot, and 2) a secure underlying architecture called Einstein Trust Layer, which prevents third-party developers from using customer data to train AI models.

During his keynote speech, CEO Marc Benioff directly criticized all AI chatbots in the market, saying, "These systems are good, but not great. They provide incorrect answers. You can call it hallucination, but I call it a lie. These LLMs are very persuasive frauds."

He also expressed concerns about AI developers using enterprise data to train models, which has made many B2B customers reluctant to use AI. He added, "Many CIOs say, 'I'm not ready to give my data to these LLMs to make their systems smarter.'"

His concerns are not unfounded. A study released by chip manufacturer AMD in August showed that more than half of the 5,000 IT decision-makers surveyed worldwide stated that their companies are not ready to adopt artificial intelligence systems due to concerns about data security. Samsung is one of the many companies that prohibit employees from using chatbots like ChatGPT because of concerns that sensitive information may be used to train large models.

With the "Trust Layer," customers can safely use large models without worrying about the risk of information leakage. As long as customers access the large models (such as OpenAI's GPT-4) through the Salesforce platform, the transmitted information will always remain confidential, and AI developers cannot retain or use this data for training.

Marc Benioff stated, "We don't look at your data, your data is not our product. We are here to make you better, more productive, and more successful."

Einstein Copilot is also different from other AI chatbots. Salesforce emphasizes that compared to other products, Einstein Copilot significantly reduces the frequency of fabrications and inaccuracies.

However, it should be noted that there is currently no known method to completely eliminate the issue of hallucinations in large models.

Einstein Copilot is currently in the testing phase, and Salesforce has not provided an expected release date. The "Trust Layer" will be fully launched on the Salesforce AI platform in October.