ChatGPT and the Chocolate Factory

It has been four days since I penned the first part of my reflections on ChatGPT, an incredibly long period in which much has transpired beyond the frantic press releases from brands in every industry talking about their new ChatGPT integrations. Before we dive deeper into the “why” everyone might be doing that, I’d like to highlight some interesting ones (use cases).

First, remember all the clunky legacy customer service chatbot windows you’ve found yourself frantically trying to close the pop-ups for when accessing net banking, travel, e-commerce, or mobile services? Those getting replaced by something like SupportGPT by Forethought might make those erstwhile dumb bots a bit more human-centric and valuable. Forethought, said to have been developing its own model for the last few years, now seems to be dumping its own in favor of an outright ChatGPT integration instead. Safer bet.

Second. ChatGPT integrations are coming to the web3 messaging platform of choice - Discord. Discord will use ChatGPT to make its own bot - Clyde, a bit more intelligent- and improve its content moderation services for its 150 million monthly active users (MAUs). Discord is also trying to position its platform as the home of AI developers by dedicating its ecosystem fund and resources to AI incubator programs designed to help developers build AI apps for Discord.

And lastly, EinsteinGPT from Salesforce will make the cute icon within the CRM more intelligent and intuitive and improve field marketing and sales by orders of magnitude.

Chocolate factory (of use cases)?

Back to the title, then a giveaway. “Charlie and Chocolate Factory” is the acclaimed and much-loved children’s novel by Roald Dahl that you might have read. Or you might perhaps be more familiar with the second theatrical rendition from 2005, which starred Johnny Depp as Willy Wonka. Either way, you might ask what that has to do with ChatGPT.

Here is what. After being an operator for two decades, my recent stint as an investor helped me better appreciate a simple technique that allows you to triangulate the trends underway without getting overwhelmed by the sheer pace of change.

• Simplify (break things down to first principles).
• Develop a thesis (what does it all mean to you?).
• Refine and revalidate the thesis (seek feedback, validation, dovetail new developments)
• Repeat.

So my mental model of ChatGPT being shoved into everything works better with an analogy and a mental model designed to obfuscate and abstract complexity. And I chose the Chocolate Factory.

TL;DR (too long, didn’t read) of the novel’s plot for those unfamiliar. Willy Wonka, the eccentric but benevolent genius chocolatier, has been operating his chocolate factory in stealth with the help of bots called “Oompha Loompas” for fear of competitors stealing trade secrets. Willy suddenly announces that five raffles hidden in chocolates will land five lucky winners a guided factory tour of this factory which has been out of bounds until now. Charlie is one of the five lucky winners, alongside four other children. The four “ill-intentioned” children eventually lose out to Charlie during the tour. Willy reveals that he has been looking for a rightful heir to run the factory, and Charlie is that chosen one. (underdog winner takes all). Curtains.

The apps building on or embedding ChatGPT in their products as a feature as the five children experiencing the prowess of the factory with one (or a few) emerging as real winning use cases.

It is the timing, also

Stripe - my favorite fintech brand had its internal valuation cut by half in recent months, despite being on track to clock $1 trillion in payment volumes this year! Private valuations and funding in the start-up ecosystem have plummeted to new lows. Across fintech, insuretech, edtech, health tech, food delivery and fast fashion, D2C (direct-to-consumer) companies are beginning to realize the limitations of a digital on-ramp with digital advertising-dependent customer-acquisition-costs (CAC). While most of their CFOs are scrambling to control the “burn rate” (cash spend) and provide nice looking “adjusted EBITDA” (obtuse adjustments adjusted continuously) numbers for their next press release on “our path to profitability,” these companies are looking to beef-up their real competitive moats. Along comes the saviors - ChatGPT and other large language models based on generative artificial intelligence that can augment their core products.

Most might be in the “let’s do it, announce it, and figure this out along the way” mode, but the real winners of the chocolate factory tour are yet to reveal themselves.

As I noted in the previous post, there are problems aplenty with LLMs, one that it isn’t cheap. Retrofitting ChatGPT to your existing app layer and business model without a more precise roadmap and utility might not be sustainable for most.

Owning value accrual vs. renting it

It remains to be seen whether the value accrual from ChatGPT will be at the app layer or further away from it. While OpenAI is keen to offset the cost of running ChatGPT, the path to monetization will look very different at the app layer depending on whether the generative AI is being offered as part of the standard or the freemium model. It also remains to be seen whether more value will accrue to the likes of Notion, Snap, and Discord or to the foundational model developers like OpenAI or Stability AI.

The one clear area of robust value accrual is “no or low code” developer tooling which allows for the abstraction of code complexity in the writing of software or smart contracts in the web3 world—application of the now soon-to-be-introduced GPT4, which will replace GPT3.5. Again trying not to be overwhelmed with how many trillions of parameters more does the model include beyond the current one, and focussing on the factory, aka use cases.

It helps to have a more coherent model about attribution to sources than the current one. I am bullish on new enterprise use cases in customer service and developer tooling. Obviously, no one is going to run massive software written by ChatGPT alone, but this is about cutting time-to-market for, say, new games by orders of magnitude with the code being co-written by publishers and ChatGPT (reminds me of the Beatles song, “With a little help from my friends”)

So while big tech firms push their teams to embed generative AI into more of their products and start-ups embed generative AI into theirs without having figured out why they might be doing so, it is vital to have a simple mental model to make sense of this Cambrian explosion. Mine is an attempt. Good Charlie(s) will come along to save the Chocolate Factory and reveal himself as the rightful heir aka use case of ChatGPT. Until then, more chocolates and more press releases!

Disclaimer: Not one word here has been written by any LLMs.