I’m having a hard time stepping away from ChatGPT. Even with its flaws, it’s a very powerful consumer AI product. I admire how you can repeat prompts to get responses worded differently yet equally accurate. As I think about the impact of ChatGPT on the world, several points of potential disruption emerge: greater spending on chips and battery backups, increased value of data center REITs, less spending on consulting, more productivity per knowledge worker, more cheating in schools and more teaching in classrooms. All potential disruptions orbit a central question that investors are asking: What does it mean for Google?
In the end, I believe that Google will be fine because they have the vision, money, and users to respond and compete with the rise of ChatGPT.
The ChatGPT Basics
The company behind it is, of course, OpenAI. Founded in late 2015 with a $1B investment from Sam Altman, Elon Musk, Greg Brockman, Reid Hoffman, Jessica Livingston, Peter Thiel, AWS, Infosys, and YC Research. The purpose of OpenAI was to become an AI research lab offering products to the public domain to reinforce learning. The company’s two most visible projects include a more text-focused ChatGPT, launched in Nov. 2022, and DALL-E, launched Jan. 2021, which renders images from text inputs.
ChatGPT is the most common use of GPT-3 (Generative Pre-Trained Transformer 3), which is a language processing AI model developed by OpenAI. It uses GPT-3’s natural language processing to power chatbots which, in practice, enables you to have a human-like conversation with a machine.
How many users are currently using ChatGPT is an unknown, though the service had 1m users within five days of its launch. I don’t have a guess on what the current number is beyond the fact that its growth has been exponential.
ChatGPT does have its shortcomings. First, it’s not trained in real-time. You can’t ask it questions on the weather, store hours, sports scores, or any events before 2021. Second, it doesn’t cite sources. Adding transparency to where information comes from will be critical to the platform’s long-term success.
Near-term use cases of ChatGPT:
- Suggest replies to customer reviews or social media comments.
- Detailed fitness plans based on your unique goals.
- Craft Q&A prompts, pen a novel or help prepare for an interview.
Long-term use cases of ChatGPT:
Since 2019, Microsoft has invested $3B into OpenAI and is rumored to be ready to invest another $10B at a $29B valuation. This week, the company announced Azure OpenAI, which will enable businesses to integrate the capabilities of ChatGPT and DALL-E into their own cloud services. This is a similar approach to those of Google Cloud and AWS. For consumers, I imagine this could mean asking Outlook to write an email, Powerpoint to build a presentation, or Excel to build an earnings model.
It’s possible that Microsoft will also try to incorporate Bing results into ChatGPT to solve the chatbot’s real-time information problem. While sensical, there’s a debate as to whether the combination is feasible due to the enormous computing that would be required for merging the two.
ChatGPT vs. Google
Google Search’s strong suit is in answering queries that fall into four categories: internet navigation, maps, shopping and general queries like daily weather, copywriting or summarizing records. The use of Google is more weighted toward shopping and general queries. While ChatGPT overlaps with Google Search on some prompts, ChatGPT’s responses are drastically different than Google’s in that they mimic human dialogue and fail to cite sources.
ChatGPT’s strength is in its personalized assistance with email responses, content creation, brainstorming, and summary of long documents into bite-sized analysis. It can also correct grammar, categorize items or extract specific information from documents. To really get the most out of ChatGPT, the system requires you to enter a clear prompt (a question, statement, or a few sentences). The more detailed the query, the more detailed the response will be.
Potential long-term risk to Google
At this point in GPT-3’s development, it seems like training data is what will differentiate one large language model (LLM) from another. So, if you have unique sources of data like Tweets or customer reviews, you might receive unique responses from the AI model. One possible future is that these LLMs could be built into the backend of many of the tech services we use. This is the outcome that could hurt Google in the long term. Fewer eyeballs means less ad revenue, but I would characterize this as an unlikely outcome.
Google will respond to ChatGPT
- The vision. In 2017, Google shifted how it communicates its focus to investors from “organizing the world’s information” to an “AI-first” company. That year, they mentioned “AI” and “machine learning” 23x on each of their quarterly conference calls. While the use of these words has declined over the years, it’s clear that Google still sees itself as an AI-first company. To date, AI is most clearly present in making Google Search results more relevant, the Waymo self-driving project and in TensorFlow, Google’s open-source software library.
- The money. Currently, ChatGPT is not trained in real-time and it can only call upon historical information. Some of the most valuable information for a search engine like Google is about the present: Is this store open right now? Who is leading the Red Sox game? How long will it take me to get to my meeting? Where can I buy a new pair of boots? Google can answer a lot of general queries and get paid to serve ads to users. In the September 2022 quarter, Google reported an operating income of nearly $24B in its Google Services segment, of which 65% is attributed to Search performance. The rough math is that Google can count on >$60B in annual operating income from Search. More than enough money to fund investments that will yield a ChatGPT competitor, if it doesn’t act on a rumored launch of DeepMind’s chatbot called Sparrow.
- The users. Google is part of the fabric of the internet based on number of users. I believe that the company has four products with >1B monthly users including Search, Chrome, Gmail, and YouTube. Additionally, I believe they have two products with >500m monthly users including Google Drive and Google Photos. Maintaining users is a powerful competitive advantage for Google, and it means that GPT-3 needs to be 10X better than whatever Google will announce.
- ChatGPT may be just a feature. When people go from not understanding what OpenAI or NLP is to being blown away by the results from ChatGPT, they see exponential progress and anything seems possible. If we can go from nothing to something as useful as ChatGPT then, in 10 years, it could be able to answer anything. Taking a step back: Time will tell if ChatGPT is more than a shiny black box.