Generative AI experiment points to a transformative future for enterprises
ChatGPT, the generative AI service from OpenAI, has taken the world by storm in the last three months. Today OpenAI publicly introduced ChatGPT APIs that are better quality than the GPT-3 but 90% less expensive. I’ve been following OpenAI research since 2015, and here we are, eight years later, witnessing one of the most profound tipping points in the AI revolution.
Generative AI is the new norm for how humans interact with data and computer systems in the future. While far from perfect, ChatGPT can learn on the fly and generate contextually relevant content or responses reasonably aligned with human expectations. ChatGPT is not yet another chatbot but a competent AI model with both APIs and “conversation UI” workflow. It is poised to disrupt the conventional enterprise software workflow and dashboard paradigm.
While millions of people (soon millions of developers) use ChatGPT daily, most still wonder how it will impact their work, products, and daily jobs. Many companies share this sentiment as they scramble to leverage new AI tools. I met a CEO who intends to use ChatGPT to take his company's productivity across all departments to the next level, but he is yet to figure out how to do it.
I am excited to showcase Zscaler’s vision and practice of leveraging the ChatGPT era generative AI technology. Two key motivations helped us get started last year:
- Help our customers achieve the most robust and secure digital transformation infrastructure possible
- Help enterprises, including ours, to be more efficient operationally
ZChat: Personalizing ChatGPT
ZChat is an internal tool that my team has been building, leveraging ChatGPT technology but providing more customized and vertical services based on Zscaler’s use cases. We’re excited about the prospect it holds as a “digital assistant” (some of us call it a “digital intern” to manage the expectation appropriately). Even though it is not replacing the role of a human, it can work around the clock tirelessly to collect data, parse data, distill insights, make summaries, and solve real-world problems.
Full disclosure: ZChat is not tied to Zscaler’s product roadmap yet, but the industry-first customized ChatGPT model demonstration below depicts real insights from real data. In our demo, ZChat leveraged the GPT technology but did not send sensitive information to OpenAI.
We are building ZChat on a unified platform on top of services from OpenAI, other public Generative AI model providers such as Google, and our AI team. There are a lot of long-term goals for this platform: no sensitive information leaking out of the company; resistance to “prompt injection” exploitation; sanity checking the inbound responses; metering of the service usage; making “prompt engineers” more productive and collaborative, fining-tuning foundation models, and training customized vertical models.
The seven-minute video below showcases two ZChat use cases for massive productivity improvements:
- The SaaS procurement manager is evaluating the usage of a sales enablement tool and trying to anticipate how many licenses will be needed in the future.
- The IT security admin wants to check and fine-tune a Zero Trust configuration recommended by ZPA Intelligent Policy.
As you can see from the demo, we are taking generative AI to the next level by allowing organizations to ask questions about their own data without any sensitive information leaving their environment. ZChat dramatically accelerated two business workflows. One Fortune 500 CIO told us that he could imagine how he could get some urgent questions answered on a weekend afternoon without bugging his team.
Our personalized model initiative ZChat is still in the infant stage, but it will play an increasing role at Zscaler to deliver a secure digital transformation infrastructure to our customers with reliability, availability, and serviceability.
Generic ChatGPT model practices
Besides the personalized model demos above, generic ChatGPT, soon GPT-4, and other public generative AI models can help many departments well in an enterprise. Here are just a few examples:
- Sales and marketing professionals often need help to write personalized emails at scale, as personalization is crucial for outbound communications. While many people know how to do it, the quality often suffers at scale. Fortunately, with GPT's smart tools like ”AI for Sheet - Cargo Addon,” it's now possible to generate thousands of uniquely personalized emails in seconds, saving valuable time and improving the quality of communications.
- Microsoft Excel is still the most widely used tool in the financial industry. Here are 80 useful Excel ChatGPT prompts to help automate tasks such as financial forecasting and detecting anomalies in data, which make financial teams work more efficiently and accurately. ChatGPT will make Robotic Process Automation (RPA) truly applicable to sophisticated financial tasks.
- In engineering and data science teams, programmers and scientists can leverage AI pair programming tools like GitHub’s Copilot, generative AI code analysis tools like Metabob, and data science-specific ChatGPT prompts to speed up their day-to-day tasks.
Also, many resources are helping people to learn “prompt engineering,” such as this one. Prompt Engineer is likely among the fast-growing new job categories in the coming years.
ChatGPT disrupting the software industry
ChatGPT has the potential to disrupt almost every industry, but it will hit the enterprise software industry first:
- AI-Generated Content (AIGC) for developers: "When we first launched GitHub Copilot for Individuals in June 2022, more than 27 percent of developers' code files on average were generated by GitHub Copilot," GitHub said in a Feb 14, 2023 blog post. "Today, GitHub Copilot is behind an average of 46 percent of a developers' code across all programming languages -- and in Java, that number jumps to 61 percent." Code generation, document generation, test generation, etc will be done more and more by ChatGPT, like generative AI technologies, from here on.
- A new generation, intelligent “low code, no code” platform for non-developers: The ChatGPT conversation-based platform will effectively become the new generation “low code, no code” platform for the future of the enterprise. ChatGPT will enable every non-developer, including the CEO/CIO/CFO of a company, to have “developer” capabilities, thanks to the “digital assistants” working for them around the clock.
- A CLI/API/GUI -> CUI transition for enterprise software users: The Conversation UI (CUI) powered by generative AI models will disrupt the CLI/API and GUI workflow/dashboard way we interact with computers and data for decades. More and more people will now instead desire to have a “conversation” to finish their workflow at work.
Too many people are discussing whether ChatGPT will displace jobs. I firmly believe that such a disruptive technology won't replace humans directly yet, but it may replace those who don’t keep pace with applying these new capabilities. Ultimately, ChatGPT APIs, GPT-4 (coming), and various generic and personalized generative AI services will significantly accelerate the missions of Zscaler and many companies in the coming decade.
The road ahead
In future posts, we will dive deeper into use cases and examine how to turn them into new solutions. We’ll explore opportunities to fine-tune the foundation model to get better custom use case results. We want to address bias, compliance, and quality control topics, and we are well aware of concerns about “AI hallucinations.” We want to calibrate what type of questions or prompts are better suitable for generative AI models and what’s the capability boundary or sweet spot of a given model. We will also want to compare ChatGPT APIs with alternatives. One final note worth mentioning is that historically, a critical bottleneck for business intelligence and business insights is data availability or how clean and complete the data is; the bottleneck is not always on the AI model per se. However, we’ll explore how recent generative AI advancements may help address the data pain points for enterprises. We have a lot to learn on this journey, but we are embracing the new normal set by ChatGPT.
What to read next