Reaching 1 million users quicker than any other application, ChatGPT's rapid ascent has raised some fundamental questions about the nature of AI, and what it means for the broader economy.
Stephanie Roth, Senior Markets Economist
Bruno de la Sota, US Equity Strategist
This year’s JPMorgan Chase&Co. Shareholder Letter from Chairman & CEO Jamie Dimon highlights the importance of Artificial Intelligence (AI), Data and the Journey to The Cloud. AI is an extraordinary groundbreaking technology which may not only reshape the finance industry, but accelerate the next era of innovation and efficiency. In our view ChatGPT is just the beginning of what is possible.
Since its launch in November 2022, ChatGPT has taken the world and media by storm. Reaching 1 million users quicker than any other application, its rapid ascent has raised some fundamental questions about the nature of AI, and what it means for the broader economy.
In this piece, we analyze the various factors driving this phenomenon
What is ChatGPT?
We asked ChatGPT who – and exactly what – it is. This is what we got.
“ChatGPT is a large language model trained by OpenAI that is capable of generating human-like responses to natural language queries. It works by utilizing a deep neural network architecture called a transformer, which has been trained on a massive amount of textual data to learn patterns and relationships between words, phrases, and sentences.
In essence, ChatGPT works by processing a large amount of textual data to learn patterns and relationships between words and phrases, which it then uses to generate contextually relevant and linguistically coherent responses to queries. The result is a system that can understand and respond to natural language queries in a way that mimics human conversation.”
Not perfect, but quite impressive.
Why all that buzz?
OpenAI launched GPT-3.5 in November of 2022. The public prototype quickly gained traction, reaching 1M users in only 5 days. By January 2023, it surpassed the 100 million user mark, making it the fastest growing platform ever.
The large language model (LLM) successfully distinguishes itself from the traditional chatbot by making itself look and feel like a genuine human interaction with a great level of expertise across a wide range of topics. LLMs have the potential to redefine how we approach daily tasks increasing efficiency and improve utility and engagement. Perhaps unsurprisingly, Big Tech quickly announced multi-year commitments to Generative AI:
Some of the headlines:
- “Microsoft Invests $10 Billion in ChatGPT Maker OpenAI” (Bloomberg, 01.23.2023)
- “Google Releases ChatGPT Rival AI ‘Bard’ to Early Testers” (Bloomberg, 02.06.2023)
- “Baidu Surges as Hope Mounts over Chinese Answer to ChatGPT” (Bloomberg 02.06.2023)
- “Salesforce follows Microsoft in launching A.I. tools for salespeople with help from OpenAI” (CNBC 03.07.2023)
- “Salesforce Launches $250 Million Fund for Generative AI Startups” (Bloomberg 03.07.2023)
- “Citadel Negotiating Enterprise-Wide ChatGPT License” (Bloomberg 03.07.2023)
ChatGPT is the fastest growing application
Investor Interest in generative AI soared in 2022
Generative AI investments will likely focus on 3 layers: 1) Applications that integrate AI models into user-facing products, 2) Models that power AI products running on proprietary models (or on open source) and 3) Infrastructure vendors that run the computing power required to service the models.
Use cases for ChatGPT and LLMs
Owing to their adaptability, LLMs have become very powerful. Only recently, ChatGPT passed: 1) a Wharton MBA exam, 2) a U.S. medical licensing exam, 3) several law school exams and 4) a Google coding interview for a Level 3 engineer with a $183,000 salary. Evidently, the prospects are wide-ranging.
Some other use cases could include:
- Conversational Agents: Think of an e-commerce chatbot acknowledging the availability of inventory, then recommending a product that could be delivered the following day. Customers receive an expedited service, and operations are improved by accounting for stock levels, planning, supply chain disruptions and lead times. Put simply, the AI can undertake such tasks quicker and more efficient.
- Virtual Assistants: Recently, Salesforce confirmed that they would integrate ChatGPT into EinsteinGPT, their CRM service. The virtual assistant will automate more time-intensive sales tasks like composing emails, scheduling meetings and preparing notes for future interactions. By using generative AI to summarize past customer exchanges, sales teams will be able improve their chances of closing. It could even automate the process of prospecting altogether.
- Language processing: Industries relying on extensive contract edits could also benefit. LLMs have the capacity to draft a contract suggesting suitable language and even review existing agreements. The technology could also be leveraged during legal research to provide summaries of cases, laws and even translations.
- Medicine and Diagnosis: Virtual assistants for telemedicine, clinical decision support, evidence-based recommendations that improve patient outcomes. These are just some of the ways AI could facilitate medical processes.
Should I be using ChatGPT instead of Google?
In short, not really. Broadly speaking, ChatGPT and LLMs are not designed to become better versions of existing search engines. Alphabet could certainly embed functionalities within the classic search engine tool to offer conversational response similar to ChatGPT. However, improvements in efficiency and productivity by integrating LLM into existing services & products is where we think the real growth opportunity is. As key players start to position, we would expect investments in the following segments for commercial AI applications:
1. Productivity and Efficiency Applications
Microsoft CEO Satya Nadella has been very vocal about the integration of ChatGPT into Bing’s search business, following Microsoft’s investment into OpenAI. The upside potential comes from growth in productivity and an Office365 offering. GitHub Copilot (powered by OpenAI Codex and acquired by Microsoft in October 2018) already shares autocomplete-style suggestions from an AI pair programmer as you code.
On March 16th 2023, Microsoft introduced Copilot to a small number of enterprises and individuals, testing the integration of ChatGPT GPT4 into Office365. Copilot in Word gives users a first draft to edit and iterate, eliminating the worry of writer’s block. Copilot in PowerPoint helps create presentations with a simple prompt. The Excel iteration analyzes trends, creating professional-looking data visualizations in seconds. With Teams, those late to meetings will be able to get a summary transcript of what they missed. Copilot will even be able to assist the organization of Outlook inboxes.
It’s only a matter of time until companies decide to monetize improvements in utility. However, the real advancement will only be as good as the underlying data in which the algorithm is trained on.
2. Cloud Infrastructure
Running queries on large datasets can increase workloads on the public cloud infrastructure. The scaling needed to satisfy the demand for computing power will only put upward pressure on infrastructure investment. It is here that having a cloud partner, like Microsoft Azure, Amazon Web Services or Google Cloud Platform, is key to develop a solid LLM.
3. Computing power and Chips
Estimating the capacity required to absorb increasing workloads ranges widely. Latest forecasts (source: Dell’Oro) expect unit shipments for AI datacenters to double from ~7% today to ~15% by 2027, and revenue to grow from ~20% today to ~45% by 2027. That expansion appears to be driven by a 3x-10x increase in AI server costs.
What are the costs and challenges for AI?
Running & training LLMs isn’t cheap. Ultimately, cost per search will come down as algorithms get more efficient. LLMs perform poorly when models can’t access enough data for training, resulting in “underfitting.” The flipside is also true: overtraining may lead to tunnel vision, leading to a lack of knowledge in other fields. Regulation is also likely to pick up. With each product cycle, regulators have been quicker to respond to potentially contentious issues, especially in relation to content moderation proliferation of disinformation and biases.
What does AI mean for the economy and labor market?
We figured we’d put ChatGPT to work. We asked it key questions about the impact of AI, then analyzed its answers.
Is AI good for the economy?
ChatGPT: “Artificial intelligence (AI) can have a positive impact on economic growth in a number of ways: 1) Increased productivity: AI can automate routine tasks and perform them faster and more accurately than humans, which can increase productivity and output; 2) Cost savings: By automating tasks and processes, AI can reduce labor costs; 3) New business opportunities: AI can help businesses identify new opportunities for growth and innovation; 4) Improved decision-making: AI can help businesses make more informed and data-driven decisions.
However, it is important to note that some industries may see greater benefits than others, and the impact of AI on employment and income inequality also needs to be considered.”
Our take: Just like disruptive innovations of the past, AI is likely to boost the productivity of the US economy. Jobs may be displaced during the rebalancing process, but growing pains should give way to new jobs and greater worker output.
Long run growth is a function of productivity and employment
Estimates are wide-ranging, but many suggest company productivity improves by 2-4%1 when AI is adopted. Around 25% of companies are expected to embrace AI in a meaningful way (based on estimates from Goldman, more on this later) and that could amount to an annual productivity boost to GDP of 0.5%.
How will AI impact the labor market?
ChatGPT: “Artificial intelligence (AI) has the potential to automate many jobs that are currently performed by humans, which can lead to job displacement in certain industries. However, it's important to note that AI can also create new job opportunities and change the nature of work.
Some jobs may see more automation than others, and some industries may see a greater demand for workers with AI-related skills. Jobs that require creativity, social intelligence, or physical dexterity, are less likely to be automated by AI.”
Our take: AI will impact the labor market, much like the advent of prior technological innovations. To maximize profits, US corporations have often found cheaper labor abroad and/or automated processes.
As we look forward, “the race between automation and augmentation” is on2. Goldman Sachs estimates that 25% of current work tasks could be automated by AI in the U.S. and Europe3. They find that traditionally high skill, nonroutine jobs are highly susceptible to automation from AI (legal & financial operations). Meanwhile, manually intensive jobs or roles that require interpersonal skills are less susceptible (construction, personal care).
Jobs at risk of automation are higher paying
What comes next?
“AI will be critical to our company’s future success [..] The importance of implementing new technologies cannot be overstated.”
Chiarman & CEO Jamie Dimon Annual Shareholder Letter
Expect more companies to announce multi-year commitments with OpenAI, or share developments on their own language processing models. The opportunities are immense, and it remains unclear who will lead the race of innovation. What is clear is that the battle is not only between the Technology companies, nor is it about the winner taking it all. In an era of efficiency and productivity, those that respond the quickest will lead. The incumbents have a head start, but new players will undoubtedly emerge.
1 Damioli, Van Roy, and Vertesy (2021), Alderucci et al. (2020), Czarnitzki, Fernandez, and Rammer (2022)
2 Autor et al. (2022)
3 Based on the review of existing literature on the probable use cases of generative AI, GS classified 13 work activities (out of 39 in the O*NET database) as exposed to AI automation, and in the base case assumes that AI is capable of completing tasks up to a difficulty of 4 on the 7-point O*NET “level” scale. GS then takes an importance and complexity-weighted average of essential work tasks for each occupation and estimates the share of each occupation’s total workload that AI has the potential to replace. GS further assumes that occupations for which a significant share of workers’ time is spent outdoors or performing physical labor cannot be automated by AI.