Earlier this year, as I was considering the free time I’d have over the summer, I approached Salesforce Ben with an idea for a series of articles. Thinking through which technologies are relevant and garnering the attention of Salesforce and the Salesforce ecosystem, nothing seemed more so than Generative AI (GenAI).
At the same time, I’ve been on my own journey with GenAI. I was quickly underwhelmed by the ‘parlor tricks’ of write-this-pithy-thing-in-the-voice-of-this-famous-author. Fun… funny even! But with my background in enterprise software, I was skeptical as to how to turn it into something useful that could be monetized for business.
My perspective changed on taking the free ChatGPT Prompt Engineering for Developers short course from Deeplearning.ai. I began to see two categories of use cases with real potential for GenAI.
First, the possibility of a more intuitive human assistant for some content generation and productivity tasks.
Second is summarization, entity recognition, and extraction. I especially see the pulling meaningful data out of unstructured text as a rich vein to explore. Once I saw the potential of these capabilities, I went from skeptically tracking its progress to trying to read, listen, and learn something about it every day.
All this is to say, I’m not an AI expert, but I am trying to learn. And that’s what I plan to share with you over the course of this series.
Predicting What’s Next for GenAI
In making our way through this new AI hype cycle, we all want to understand what’s going to happen. This is a natural temptation that is worth resisting. To make my case, let’s examine some AI prognostication history to inform how we might view the current GenAI trend.
First, consider the following two quotes, and as a thought exercise, I want you to guess when these words were spoken:
- “Machines will be capable, within twenty years, of doing any work a man can do.”
- “Within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved.”
These two statements were made by prominent AI researchers Herbert Simon and Marvin Minsky (look up more about these guys!) in 1965 and 1967, respectively. Within ten years of these statements, disillusionment would set in for the AI research community, governments would be defunding their AI programs, and the “First AI Winter” would see AI relegated to lower league status for the next decade.
With this very thin evidence, I’m going to assert what I’ll – somewhat arrogantly – call “Chittum’s Precept”. Which is: talented computer scientists are no better at predicting the future than the rest of us.
We can test that out on a more recent prediction from “godfather of AI” Geoffrey Hinton in 2017 after some research that showed remarkable results for modern machine learning AI models to [diagnose certain types of cancer]
“It’s just completely obvious that in five years deep learning is going to do better than radiologists.”
Whilst early results looked promising, it was later found that pure AI-based radiology diagnosis was not optimal. To his credit, he later corrected his forecast to propose a vision of offloading the repetitive work to the AI and letting the human focus on problems related to reasoning and cognition.
When it comes to predictions about where the current AI hype will lead to, I prefer this “prediction” by economics journalist Kai Ryssdal on the Prof G podcast the other week:
“…we just don’t freakin’ know what’s going to happen with AI…”
An Action Plan for Today
It is too early to know where the current GenAI movement will lead us. Still, my own intuition is tugging on me. It seems to me that we’re seeing the beginnings of a long-term transformation in user interface – similar to how mobile impacted computing.
Apart from intuition, there are two other factors that make me take GenAI seriously. The first is the interest on the part of non-developers. Developers are naturally curious and adventurous about technology. But the public GenAI apps like ChatGPT, MidJourney, and Dall-e have made GenAI very accessible to non-coders. In doing so we have a vision for how conversational UIs may transform how we talk to computers in the future.
The other factor is the opinions of people I trust. This includes friends and colleagues as well as public personalities like Cara Swisher and Ezra Klein.
If you’re convinced, like me, that this is something big and worth investing your time in, here are three immediate (if obvious) steps you can take today to ride the GenAI wave.
During my journey into better understanding GenAI, I’ve realized that I’ve stepped into the middle of a 70-year conversation. The science of artificial intelligence has been hashed, rehashed, and debated for decades. As I stated previously, there are hype cycles, disillusionment, and even outright feuds over what artificial intelligence is.
Getting started can be daunting. But as Desmond Tutu observed, “there is only one way to eat an elephant – a bite at a time.”
So start with what you’re interested in. For me, my career has been defined by my ability to communicate with words. For this reason, I’ve primarily focused on text-based AI. This led me to take the prompt engineering course. From there, I’ve sought out the people and companies that are primarily focused on this part of GenAI.
Your journey may be different.
As I stated in my previous blog post, I think a good starting point for all of us in the Salesforce ecosystem is the “Getting Started with AI” trail on Trailhead, and there are more modules beyond this. But I’ve found myself deeply interested in the full scope of what’s going on in GenAI.
For instance, I’m interested in understanding the foundations of how machine learning works. For this reason, I’ve been going back over the Machine Learning Guide podcast. I’ve also put Andrew Ng’s Coursera Machine Learning Specialization course on my backlog.
Of course many will want to dive deep into Salesforce’s own AI work. In that case, I’d recommend following the Salesforce Research blog. They are sharing a lot about the work they’ve done in AI, including published work about the CodeGen project and a vision for making GenAI actionable – something very relevant to Salesforce practitioners.
Experimentation is a big part of how we learn new technologies. Once you’ve learned the theory, a hands-on experience can serve to bed in that knowledge.
Beyond this, posing a problem and attempting to solve it with technology is a sure way to get the creative process going. This leads to new ideas and, by extension, new applications and businesses. As observed by Redmonk’s Kate Holterhoff in her article AI is for Tinkerers, one of the exciting things about this new wave of AI technology is just how open it is for experimentation.
First, many existing models have an API. Integrating with these APIs presents an opportunity to see how they can fit specific use cases that might be of interest. Using foundation models with fine tuning and clever prompt engineering can produce some amazing results. Integrating ChatGPT with Salesforce was one of the first things I attempted to get to know how to work with GenAI.
Jitendra Zaa presented his GPT integration for Apex Hours which was useful to see how a generative AI might appear within Salesforce. Keir Bowden recently wrote a post showing his integration which surfaced the integration with OpenAI’s GPT model as a Salesforce CLI plugin. The tinkering has begun.
But experimentation goes beyond just integrating with one of the many APIs available. Entire models have been open sourced. Hugging Face has built a community around a large number of models, data sets, and apps.
The pace of open innovation and invention can’t be understated. And it is open to anyone to participate in. Even if you’re not a coder, you can play with creating prompts with different public models and see what you can get them to do.
Those working in the Salesforce ecosystem know the value of community, and there are loads of communities around AI. Virtually every company who has an offering, model, or API has stood up a Slack workspace or a Discord server.
Similar to starting your learning journey, it might be difficult to pick which community to join. But again, follow your interests here. If you’re already in the world of data science, you might start with Hugging Face, which has built a community through its open source activities. If you’re more focused on image generation, maybe you want to explore the MidJourney community discord. For those more interested in text-based models, Cohere’s LLMU has ways to connect with others in their discord server.
And for a more comprehensive look at all the ways people interested in AI are connecting, you might want to check out the Top AI Communities of 2023 blog post in the AI Vanguard channel on Medium.
As of now, Salesforce has released its first set of Einstein GPT features. It’s a great time to begin to immerse yourself in this exciting technology. If you’ve been bitten by the GenAI bug, or if you’re even interested, make your next step today. It’s never too soon to start your journey to learn, experiment, and meet others who share your interest.
Learn about Generative AI:
Building with Generative AI:
Meet others in Generative AI communities: