Newsroom

GPT (Generative Pretrained Transformer) is a language processing AI model. It can generate natural language responses to various inputs. In a press release last week, GPT-4 was shown to have performed shockingly well on a variety of tests.

Unified Bar Exam - 298/400, which was in the 90th percentile of results.

LSAT - 163/180, which was in the 88th percentile

SAT, reading and writing - 710/800, the 93rd percentile

SAT, math - 700/800, the 89th percentile

GRE, verbal section - A 169/170, the 99th percentile

AP Art History, Biology, Macro- and Microeconomics, Psychology, Statistics, US Government, and US History exams – 5/5 on each

On a livestream, GPT-4 described images provided by the audience, including an image of a snowboarder on an alien planet, and a cartoon of a squirrel holding a camera.

On the same livestream, GPT-4 was prompted to summarize the OpenAI press release into a sentence where each word begins with G. While the prior version of the AI model failed to do so, GPT-4 generated the following sentence: “GPT-4 generates groundbreaking, grandiose gains, greatly galvanizing generalized AI goals.” When prompted to exclude the term “AI,” GPT-4 stated the following: “Gigantic GPT-4 garners groundbreaking growth, greatly galvanizing global goals.”

Potential applications of the technology include the following. As you might guess, I had some help creating this list.

Programming:

App creation: Submit information about the design of a new app and connections to different API’s and GPT-4 will create an app.

Debugging assistance: By describing an error or issue encountered in code, users can receive help from GPT-4 in identifying the problem and suggesting potential solutions.

Code explanation: Users can provide GPT-4 with a code snippet, and it can explain its functionality, syntax, or specific lines of code.

Finance:

Trade Assistant: ChatGPT can be used as a trade assistant in capital markets, helping investment bankers analyze trading data, identify market trends, and monitor trade executions, ultimately aiding in informed decision-making.

Research Analyst Support: ChatGPT can assist research analysts by automating the extraction of key financial metrics, summarizing company reports, and providing insights on market sectors, which can improve the efficiency and quality of research output.

Portfolio Management: ChatGPT can support portfolio managers by simulating various investment strategies, optimizing asset allocation, and monitoring portfolio performance, ultimately helping to manage risk and enhance returns.

Legal document drafting:

Contract templates: GPT-4 can help users draft contract templates for various purposes, such as rental agreements, non-disclosure agreements, or employment contracts

Legal research: GPT-4 can provide information on legal concepts or case law summaries within its knowledge base.

Compliance checklists: GPT-4 can help users create compliance checklists for specific regulations, industries, or jurisdictions.

Business consulting:

Market research: GPT-4 can provide insights into various industries, market trends, or competitors.

SWOT analysis: GPT-4 can help users identify the strengths, weaknesses, opportunities, and threats for a given company or product.

Business plan assistance: GPT-4 can provide guidance on creating business plans, including sections on executive summaries, market analysis, and financial projections.

Production Onion

Traditional tools largely fall into one of two categories. They either perform work or act as knowledge repositories. Google is a software tool that primarily is a knowledge repository. Photoshop is primarily a tool for achieving work. AI acts as both. An effective way to think about AI implementation is through what I have dubbed the “production onion.”

Figure II: Production Onion Stages

At the center of the onion is where we are now, stage (1). Output from AI programs is frequently reviewed. The tasks are narrow. A potential prompt to the GPT model is “write me an article about the benefits of ceramides using quotes you find on the internet.” The results will be imperfect, requiring some revision before submission.

Once the GPT model yields desirable results with sufficient frequency, reviews can be less frequent. The prompt may be the same, but there is not the same requirement for correction of the output. This would be stage (2).

Stage (3) would require broadly assigned tasks. For example, “fulfill the traditional role of editor for all skincare articles that are submitted.” An editor is more than a spell checker; they help achieve the strategic direction of a publication. The role is more broadly defined than “write an article.”

Stage (4) might have the same prompt as stage (3), but the AI model has graduated to being autonomous.

With this framework in mind, the question “will AI replace humans?” is the wrong question to ask. A better question is “will humans lose their jobs due to AI?” There are some cases, like a telephone operator where there is a 1:1 replacement of a person with a tool. However, it is more common that tools make human labor more efficient and thus increase output. The result is that less labor is needed, which can occur at each stage.

Additional thoughts

Censorship and Output Moderation – ChatGPT and other AI models have censorship and moderation rules to avoid providing offensive output. The primary goal is to avoid negative press and a repeat of Microsoft’s Tay.ai. However, a concern is that the models are presenting responses as fact when they reflect opinion. Regardless, output moderation will continue. Every society has limits on speech.

ESG response – How will ESG proponents respond if productivity increases sufficiently as to reduce the necessity for labor? ESG proponents broadly have not been opposed to outsourced labor to developing countries. The same will likely be the case of outsourcing work to AI.

Reviewers vs producers – A common insult lobbed at reviewers of art, novels, food, movies, etc. is that they failed at creating, so they settled for reviewing. AI has reduced the skill requirement to produce many types of creative products. So long as you know what people like in art, writing, programming, etc., you might not need to have the skillset to jump the gap from idea to reality. GPT can even help you understand what people like.

Data Sources – The advent of GPT has led to a shift in the balance of power between data sources and data analysis providers. As GPT continues to develop and enhance its capabilities, the significance of data sources becomes increasingly vital. GPT's ability to process and analyze vast amounts of information efficiently and accurately reduces the need for specialized data analysis providers. In this new paradigm, the quality, accuracy, and comprehensiveness of data sources take center stage, as they become the primary determinant of the value derived from GPT's analytical prowess. As a result, organizations and stakeholders will need to place a greater emphasis on procuring reliable and diverse data sources to fully leverage the potential benefits offered by GPT's cutting-edge technology.

Bicycle for the brain – Compared to other animals, humans are about average in movement efficiency. That is, it takes a human about the same amount of energy to transport one pound of mass per mile as a cow. Give a human a bicycle and a flat surface, and we are far more efficient than any other animal. Steve Jobs described a computer as a bicycle for the mind. It allows one to achieve more output per unit of mental energy. AI seems like an inflection point. It seems like it could be more like a car than a bicycle. The level of abstraction on which people can work is much greater than before.

Tech Company Distinction – The distinction between a tech company and non-tech company may no longer make sense. This was the case before tools like GPT, but the technology may make this fact truer.

Concentrations of Power – It is unclear what GPT will mean for concentrations of power among providers and users of the technology. As is usual with new technology, groups that move quickly will win and incumbents are threatened. Google’s GPT struggled in live demos on February 8th and the company’s stock fell by 7% as a result.

Ukraine War

Ron DeSantis received flak from Republicans and Democrats alike for his lacking support of the Ukraine War, calling it a “territorial dispute.” He may hold the position not to win support today, but because support for the war will decline over the next 12 months. Interest in the Ukraine war has nowhere to go but down unless Russia makes a direct attack on the West.

The business of empire requires financial means and motivation. The British Empire fell because it lacked both. The Soviet Empire fell because it lacked the financial means. Chinese influence is growing because it has both means and motivation. The American Empire, which projects power through ~600 overseas bases in 85 countries and territories, will retrench if it lacks either.

Is it any coincidence that the US troop drawdown in Iraq began as the financial crisis was heating up in December 2007 and lasted until December 2011? The troop drawdown was likely to occur regardless, however support for the war was negatively affected by domestic financial woes. If the US enters another recession, support for the Ukraine war will wane too. The effects of a recession would be insufficient to eliminate the West’s financial means to fund the war but would undermine its motivation.

Further, the West’s motivation for the war is less concentrated and immediate than that of Russia. The war is occurring in former Soviet territory and is directed largely by one man, Putin. He would be inclined to wait out the West, though his pockets are not as deep as ours. Thus, support from China is vital.

China has the pockets and manufacturing base to provide support to Russia, and in turn is being granted access to Russian natural resources at low costs. We are effectively trading the power of one adversary for another. It is likely to be another significant cost of the war.