Chat GPT Makes as Foolhardy a Bet as Apple Does a Genius One
A White Brook Capital Insight
To level set, earlier this week, Apple unveiled their implementation of artificial intelligence, branded “Apple Intelligence”. Apple Intelligence separates itself from Google’s Gemini and OpenAI’s ChatGPT by identifying and addressing user queries that contain personal information, but for capabilities it isn’t capable of performing, asking for user permission, and outsourcing the request to OpenAI’s ChatGPT, anonymously.
Thanks for reading White Brook Capital Insights! Subscribe for free to receive new posts and support my work.
Apple has smartly put itself in the position of a supra general artificial intelligence. It sees everything that occurs on the device, and it can decide if it wants to intercede. Unlike 3rd party AI, it knows its device’s capability, it can intelligence the wanted output, and it can decide if it wants to try to perform the task, and that decision can change over time.
This is genius from a branding standpoint. Anything it believes it can’t answer with an overwhelmingly likely success rate it can pawn out to ChatGPT. A successful answer by it or by ChatGPT results in consumer appreciation for Apple Intelligence, while failures are borne by the failure of ChatGPT. At the same time, unlike the outsourced AI, it can also recognize future action on the device and therefore can determine whether or not the delivered answer was successful, and use those actions to improve and extend Apple’s capability - in a way that the provider of the 3rd party AI response cannot.
Remember this is the principal problem with being an artificial intelligence provider,
One of the issues that I haven’t seen solved yet is an AI model that extracts information systematically from its waste. Google’s search rapidly improved and became a step function better than others because, for each search delivered, they observed a positive response - the user’s selection, and a negative response - the user’s non-selection. By being the default search engine, they could accurately assume that each result delivered was meaningful and that the response received was also a meaningful response to their delivery. Every other search engine that wasn’t the defacto standard could not make those assumptions, and their development suffered. AI is in a similar position to those other search engines. Given the newness of the technology and the existence of search, which often does an excellent job on its own, a user’s click to close after receiving an answer can mean that the answer was good and useful; that it was bad and quickly dismissed as inferior to search; that it was wrong but indiscernible to the user; or that it was known to be wrong and dismissed. General AI models like ChatGPT and Google’s Bard improvements thus far have been beneficiaries of indiscriminate investment rather than thoughtful revision, but these issues will become more significant for vertical models with more limited audiences and harder to deliver data.
White Brook Capital Partners Fourth Quarter 2023 Commentary
But Apple solved that problem. They, unlike OpenAI, will get value from the success and waste of the answers provided by Apple Intelligence and ChatGPT. This should allow them to rapidly improve, even if the are starting from a generation or two behind.
But just as interesting, as Apple sitting higher than the generalized artificial intelligences that were designed to be kings, is the flow of payments. There aren’t any. ChatGPT got ATT’d (The iPhone was famously exclusive to AT&T for several generations. A company which didn’t foresee the expenditures necessary to absorb new, data heavy, customers and keep Apple happy, and wasn’t appropriately compensated, in their view).
“The utility of general models will be to answer the most straightforward questions and then aggregate traffic and direct it appropriately to better-equipped vertical models to answer more advanced queries. They will monetize via advertising and kickbacks/commissions from vertical models or, for more sporadic queries, pay vertical models for answers to retain their ability to aggregate traffic. Vertical models will compete against others in the same space and must navigate the highlighted issues diligently. Their ability to earn early revenue or to attract significant early financing will prove essential to capture differentiated industry data and pay for exclusivity. At this time, there are no obvious leaders.”
White Brook Capital Partners Fourth Quarter 2023 Commentary
In exchange for OpenAI’s multi-year effort of training their AI models and spending on $40,000 a piece for a fleet of Nvidia processing units they buy or lease to use, ChatGPT will earn $0 on the billions of requests they will have to service once iOS 18.1 is in the hands of users. Instead, reportedly, both sides can participate in any future revenue share - as we predicted in the 4Q commentary.
For an AI provider, you really need to make money on either advertising or use of the product. I didn’t think that was a difficult concept, but OpenAI leadership apparently needed to hear it, so it’s worth saying. Because Apple Intelligence removes identifying information that make advertising plausible for an outside vendor (but importantly, I believe, Apple will be able to inform an ad profile for their own network), payment for usage is the only way to monetize, well. Apple, in probably IOS 19, will open an AI Store and monetize by taking 100% margin of 30% of the revenue of each AI a user seeks to use. But for ChatGPT, how will they make money?
What’s interesting about that is also how it appears the deal came to be. Apple reportedly negotiated with ChatGPT, then negotiated with Google to use Gemini, and then came back to ChatGPT. After the event they explained that in the future Gemini and other AIs will also be offered for Apple hardware users. It seems clear that Google wasn’t going to pay for distribution because they know better than anyone, you need to know whether a delivered answer was successful, and they weren’t getting that information. But OpenAI, the most famous, well regarded, AI provider in the world, jumped at the chance to not earn money providing the most valuable technology the world may have ever seen. “Distribution” is a fatal reason to make a very expensive product free to run on the backend. With ChatGPT reportedly only at a $3.4bn revenue run rate despite the considerable name recognition and astronomic valuation, perhaps Sam felt this was the move needed so that OpenAI could finally really flip the monetization switch. I think it was a bad idea, AI processing isn’t free to OpenAI and this move put a gun to the Company’s head.
At White Brook we traffic exclusively in this year's laggards, small and midcap stocks. Generally these companies have little part to play in the development of AI, and none of the companies mentioned would fit in our universe. AI is a major part of the future, but it's still one where we don’t know how you, as the owner of the Company developing it, can make money at, given current economics.
Instead, we continue to own Box, Inc (BOX). We like that it is a foundational technology for a business, and while we disagree with CEO Aaron Levie’s optimistic appraisal of AI’s likely role in future society, he’s smart, they acted early, and they have pursued AI in almost exactly the same way as Apple. Namely a way, where they can be a ROI positive beneficiary of AI’s continued development and success. They benefit from the growing competition between AIs, and, by supporting all of them, they grow their attractiveness to new customers and deepen their relationship with existing customers. In these early days of AI, rather than making their customers make a large infrastructure bet that relies on a single company who operates at the whim of a powerful CEO, that can make the wrong decision, their customers can pick and choose AI’s for the process and manipulation necessary and change them as much or little as necessary - much as I assume consumers will be able to do in Apple’s IOS 19.
Thanks for your attention, please reach out if you have any questions or comments.
Thanks!
Basil
All investments involve risk, including loss of principal. This document provides information not intended to meet objectives or suitability requirements of any specific individual. This information is provided for educational or discussion purposes only and should not be considered investment advice or a solicitation to buy or sell securities. The information contained herein has been drawn from sources which we believe to be reliable; however, its accuracy or completeness is not guaranteed. This report is not to be construed as an offer, solicitation or recommendation to buy or sell any of the securities herein named. We may or may not continue to hold any of the securities mentioned. White Brook Capital LLC and/or their respective officers, directors, partners or employees may from time to time acquire, hold or sell securities named in this report. It should not be assumed that any of the securities transactions or holdings discussed were or will prove to be profitable, or that the investment decisions we make in the future will be profitable or will equal the investment performance of the securities discussed herein