An Executive’s Cheat Sheet to Generative AI

Everyone is talking about Generative AI (Gen AI) and the hype is deafening. Almost every media outlet, technology company and influencer is trying to stake a claim in this burgeoning market. So, in an effort to cut through the noise and provide some clear, straightforward advice, here is a cheat sheet to Gen AI for Executives:

  1. When most people say Gen AI, their mental model is of Open AI’s ChatGPT – in many cases, people are thinking about what it would look like to plugin a model like (Chat)GPT into their systems. Gen AI technically includes another generative type of model called a GAN (Generative Adversarial Network), but all the current excitement is about GPT models.
  2. Gen AI is not new technology – GPT-3.5/4 is a step change within AI (but we are on increment 4 of the underlying model for ChatGPT). The current hype cycle is due to two things 1) the latest model release from OpenAI is able to respond with human-like intelligence and 2) a refined chat interface built on this model allows it to be queried as if it were a person, making the underlying technology breakthrough extremely accessible.
  3. The value case should be the focus – not the technology case. Many companies are scrambling to implement Gen AI solutions, and in a rush to tick the Gen AI box, are considering use cases where the fit isn’t optimal. While the underlying technology is ground breaking and can sometimes feel like magic, like any other tool, it only generates value by being applied to the right use case.
  4. Common Gen AI use cases will be commoditized this year – use cases such as creating a chat bot over a knowledge repository, asking questions of corporate documents or data repositories, summarizing reports, etc. will all be features available in popular enterprise products (see early examples: Office Co-pilot, Confluence AI-tools, Salesforce: ChatGPT for Slack / Einstein GPT in CRM, Five9 Contact Center OpenAI integration and many more).
  5. Be wow’d by the underlying model, not the user interface – it’s incredibly easy to build an impressive Proof of Concept (PoC – a prototype to test the feasibility of an idea) on top of an impressive model. For many simple use cases, the models are performing 99% of the heavy lifting. When considering a build team, understand how they’re differentiated beyond the packaging of an existing model.
  6. There is a (very) low barrier to entry – a competent engineering team (without data science experience) can build PoC level Gen AI solutions that use public model APIs in days. Everyone has access to the same GPT model-as-a-service APIs and open source repositories to quickly build state of the art Gen AI solutions.
  7. Differentiation doesn’t (always) mean building your own model – many use cases can be solved by using a model-as-service either directly, or by combining it with adjacent proprietary data. Fine-tuning and custom models have their place but in most cases you shouldn’t start there.
  8. You don’t need specialized skills – most companies already have the skills in-house to build a Gen AI PoC using their own data. There are some new skills and technologies needed (such as understanding vector databases, the langchain framework, prompt engineering, etc.) but for a competent developer these are very easy to pick up (think days).
  9. Your technology teams are already building Gen AI applications – the enthusiasm felt by business-focused executives for Gen AI likely pales in comparison to the excitement experienced by technical professionals. There is a huge groundswell of excited engineers who have built something using new Gen AI APIs, with the more adventurous training their own models. See 6 and 8.
  10. You can probably trust your data in a model-as-a-service product – many executives are concerned about sending their data to hosted Gen AI services. This is similar to the concern that was experienced by IT executives a decade ago when moving platforms from “on premises” servers to the cloud. If a business currently trusts Amazon, Google or Microsoft with their data and systems, in most cases this trust can extend to the model services these providers (will eventually) provide.
  11. You won’t be left behind – finally, it’s prudent to carefully evaluate the right use cases for Gen AI in your business. The space is moving incredibly quickly, so technology that is state-of-the-art today, may be obsolete in months. In many cases it can be advantageous to wait before investing too heavily in certain use cases (see 4). In the meantime, businesses should experiment internally to develop experience and ensure technology teams are keeping abreast with the latest developments.