First, I fed Voleon’s job listing for the Staff Editor position into ChatGPT, which I find is the most capable LLM for understanding language, style, and domain-specific nuances. I then asked it to find examples of imperfect writing in articles on machine learning in investing. It came up with four, explained the merits of each, and suggested one in particular. I had to steer it back on course because it made a suboptimal choice. But when I settled on a different article, it was able to select the most confusing section of the paper and explain its reasoning for doing so.

Below is the original passage and all of the steps that went into creating an improved, AI-assisted and human-authored version, along with explanations of the editorial and writing processes.

Ronen Israel, Bryan Kelly, and Tobias Moskowitz, "Can Machines 'Learn' Finance?," Journal of Investment Management 18, no. 2 (2020): 23–36. Excerpt pp. 24-25.

So why have we moved to using new terminology to describe old ideas? Above and beyond the marketing angle—machine learning is a sexy name that carries the connotation of bleeding edge Silicon Valley technology—there are at least three substantive reasons for this shift. First, the historical practical usage of statistics was frequently confined to “small” models—those with a handful of input predictor variables (or “features” in ML terminology) and simple, often linear, association rules between those inputs and the output (i.e., dependent variable) of interest. The term “machine learning” has come to serve as a shorthand to signal an explicit interest in “large” models, those with many input variables and/or those allowing for complex nonlinear associations between the inputs and output.

This idea is captured by part (i) of the GXK definition above. In order to learn through experience, the machine needs a representation of what it is trying to learn, which requires a research choice. Machine learning brings an open-mindedness for statistical representations that are richly parameterized and often nonlinear. Such models are of course not new to statistics, so it would be misleading to describe this as a contrast with “traditional” statistics. But it is fair to say that machine learning specializes in this sophisticated end of the model spectrum. Small models are rigid and oversimplified, but have the virtue that they can be used with small data sets. They are also “robust” in the sense that their behavior can be relatively insensitive to reasonable changes in the data. Large and sophisticated models are much more flexible, but can also suffer from poor out-of-sample performance when they overfit noise in the system. Researchers turn to models like these when they believe the benefits from more accurately describing the complexities of real-world phenomena outweigh the costs of potential overfit. Part (i) of this definition also points out that the primary objective of machine learning is to generate accurate predictions. As emphasized by Breiman et al. (2001), its focus on maximizing prediction accuracy in the face of an unknown data model is the central differentiating feature of machine learning from the traditional statistical objective of estimating a known data-generating model and conducting hypothesis tests.

Second, machine learning seeks to choose a preferred model from a “diverse collection” of candidate models. Again, this idea has a long history in statistics under the heading of “model selection” and therefore is not a new contribution of machine learning. But the process of searching through many models to find the best performer is characteristic of essentially all machine learning methods—it is closely connected with what machine learners call model “tuning.” Of course, looking at multiple models and selecting the top performers in-sample mechanically leads to overfit and poor out-of-sample performance. Because of this, the model search process is always accompanied by so-called “regularization” techniques and methods for identifying models that are likely to perform best out-of-sample. Regularization is a blanket term for constraining the size of a model. An optimal model is a “Goldilocks” model. It is large enough so that it can reliably identify the true and potentially complex predictive relationships in the data, but not so flexible that it overfits and suffers out-of-sample. Regularization methods encourage smaller models, and make sure that a richer model only gets selected if it is likely to give a genuine boost to out-of-sample prediction accuracy. A cornerstone method in the model selection process is cross-validation, in which the researcher simulates out-of-sample tests in historical data and picks models that would have performed best in these “as-if” out-of-sample scenarios. Element (ii) of our machine learning definition describes refinements in implementation that emphasize reliable out-of-sample performance in order to explicitly guard against overfit.

Note that this version is condensed, reorganized, and uses simpler language. Those moves in themselves saved hours of labor. However, it is quite long and robotic, and registered as 96% AI text when checked.

Why call it machine learning?

The term “machine learning” covers concepts that significantly overlap with traditional statistics, so why is it considered a separate field? Part of the reason is marketing, as the phrase suggests that it involves advanced technology. More importantly, the distinction reflects meaningful changes in how researchers develop predictive models.

The first difference concerns scale. Traditional statistical models were often designed around a small number of predictors and relatively simple relationships between those inputs and the outcome being predicted. Machine learning expands this approach by allowing researchers to analyze much larger sets of variables and more complex, nonlinear relationships.

Machine learning does not replace statistics. Many techniques originate from statistical research. Instead, machine learning extends statistics toward greater flexibility, allowing researchers to identify complex patterns without specifying relationships in advance.

That flexibility comes with a tradeoff. Smaller models require stronger assumptions, but they are often more stable when data are limited. Larger models can capture more complicated relationships, but they also risk identifying patterns that exist only in historical data rather than in the underlying system. This problem, known as overfitting, occurs when a model learns noise instead of meaningful signals.

Researchers seek a balance: models must be flexible enough to identify genuine predictive relationships, yet constrained enough to perform well on new data. Machine learning emphasizes this balance by focusing on both developing and rigorously testing many possible models.

A second defining feature of machine learning is systematic model selection. Traditional statistics has long included methods for comparing alternative models, but machine learning places greater emphasis on searching across large collections of possible models and identifying those that perform best. Because searching among many models can itself create misleading results, machine-learning methods rely heavily on techniques such as regularization and cross-validation.

Regularization limits unnecessary complexity, preventing models from becoming more elaborate than the data supports. Cross-validation tests whether a model that performs well on historical data will remain effective on new data. Together, these methods help researchers distinguish genuine predictive power from patterns that appear meaningful only by chance.

While ChatGPT’s final written output is nowhere near as good as a skilled human’s, the analysis and organizational synthesis it performs is quite useful as an intermediate part of the process—in fact, truly remarkable. Here is its explanation:

ChatGPT: This edit is guided by a simple editorial principle: readers should grasp the author's central argument before they are asked to absorb its technical details. The original passage contains a clear and persuasive thesis: machine learning is best understood as an extension of statistical modeling rather than a departure from it. However, that argument is difficult to find because of all the definitions, qualifications, and technical explanations.

 My revisions preserve the authors' substantive claims while reorganizing the discussion around the reader's cognitive experience. The revised version introduces the central argument first, then develops it through a logical progression from differences in model complexity, to the tradeoffs those differences create, to the techniques machine learning uses to manage those tradeoffs.

Throughout, I replace abstract phrases with multiple nouns with more direct verbs. I also seek to reduce unnecessary jargon, eliminate repetition, and make implicit relationships explicit. The goal is not to simplify the underlying ideas, but to make their logical structure immediately visible so that readers can devote their attention to understanding the argument rather than decoding the prose.

Grammarly’s “Humanize” function tries to mask LLM authorship. Though it is not very good, it is still a useful part of the process of wringing the robotic voice out of LLM output.

Even with my tweaks of common AI words and phrases, it still flagged its own “humanized” output as likely AI-generated. I could have used any number of other tools, such as GetsolvedAI, but using Grammarly twice made the point that its humanizer can’t even outsmart itself.

What follows is written by ChatGPT & “Humanized” by Grammarly, with my own tuning.

Why call it machine learning?

Why call it machine learning? The term "machine learning" covers ideas that overlap quite a bit with traditional statistics, so why is it treated as a separate field? Part of the reason is marketing, since the phrase suggests advanced technology. More importantly, the distinction reflects real changes in how researchers build predictive systems. The first difference is about scale. Traditional statistical models were usually designed around a small number of predictors and relatively simple relationships between those inputs and the outcome being predicted. Machine learning broadens this by letting researchers analyze much larger sets of variables and more complex, nonlinear relationships.

Machine learning does not replace statistics. Many of its methods come from statistical research. Instead, it extends statistics toward greater flexibility, permitting researchers to find complex patterns without defining relationships in advance. That flexibility comes with a tradeoff. Smaller models need stronger assumptions, but they are often more stable when data are limited. Larger models can capture more complicated relationships, but they also risk finding patterns that exist only in historical data rather than in the underlying system. This issue, called overfitting, happens when a model acquires noise instead of meaningful signals. Researchers seek a balance: models must be flexible enough to find genuine predictive relationships, although constrained enough to work well on new data. Machine learning stresses this compromise by concentrating on both developing and carefully testing many possible models.

A second key feature of machine learning is systematic model selection. Traditional statistics has long included ways to compare alternative models, but machine learning puts more weight on searching across large collections of possible models and identifying those that perform best. Because searching among many models can itself produce misleading results, machine learning methods make heavy use of techniques such as regularization and cross-validation. Regularization limits unnecessary complexity, keeping models from becoming more elaborate than the data supports. Cross-validation checks whether a model that performs well on historical data will still be effective on new data. Together, these methods help researchers distinguish genuine predictive power from patterns that seem meaningful only by chance.

All the preliminary AI-assisted steps so far are forms of research and prewriting. Now, I craft my own version, with wit, rhetoric, logic, and style that are distinctly human and beyond the capabilities of LLMs, at least for now.

Machine Learning and Investing

A Summary of an excerpt of: Ronen Israel, Bryan Kelly, and Tobias Moskowitz, "Can Machines 'Learn' Finance?," Journal of Investment Management 18, no. 2 (2020): 23–36.

by Jacob S. Dorman, Ph.D.

Many machine-learning investment techniques rely on statistical methods; they do not replace them. To the contrary, they use statistics to find complex patterns in mountains of financial data. In other words, they don’t just find the needle in the haystack; they analyze the position of every piece of hay. Using machine learning to expand statistical techniques enables analysis of much larger and more complex datasets in a greater variety of ways. It enables statistics to operate with greater flexibility, allowing researchers to let the machine choose statistical methods and patterns rather than determining those choices in advance.

Frankly, another reason we speak of “machine learning” or “AI” investing is marketing: those phrases are sexier and have greater commercial cachet than “computer-assisted advanced statistics.”

But applying stats and artificial intelligence to investing is not without peril; searching for models can itself produce misleading results, so researchers use model-tuning techniques known as “regularization” and “cross-validation”—kind of like using various methods to “check the math” of LLM-generated computer code. Regularization limits unnecessary complexity, keeping models from becoming more elaborate than the data will support. Cross-validation checks whether a model that performs well on past data will still be effective on new data. Together, these methods help researchers distinguish useful patterns from those that might seem significant but are actually just the result of chance. The goal is to find the “Goldilocks” model: just right.

After all, not all machine-learning investment models are created equal. Smaller models require more assumptions, but are often preferable to larger models when there isn’t much data. On the other hand, large models can find more complex patterns in vast datasets, but also risk “overfitting,” which happens when a model cannot distinguish between “noise” and meaningful patterns. As a result, researchers must balance the size, sophistication, and flexibility of their models to find genuinely predictive patterns. In other words, researchers need to analyze a bale of hay but adapt so everything doesn’t go haywire when they try to make sense of a barnful of data. As a result, they carefully develop and test many possible models. Humans need to be in the loop—the hay is not going to bale itself!

Throughout this process, I used generative AI as a research assistant and thought partner—not as an author. I delegated the tasks AI performs well, such as locating and summarizing sources, while verifying its work at every step. When it came to writing and editing, however, the final decisions were entirely human.

The finished piece differs substantially from the intermediate AI drafts. I reorganized the argument, removed repetition, and reduced the original 619-word passage to 366 words without sacrificing its substance. I also introduced literary devices—metaphor, alliteration, aphorism, subtle humor, and internal rhyme—to make the ideas both more memorable and more engaging while preserving technical accuracy.

The central hay metaphor illustrates the difference between human and machine writing. It grew from my own memory of stacking fifty-pound hay bales on my best friend's farm in upstate New York three decades ago. LLMs can imitate metaphor, but they cannot draw on lived experience. Likewise, I restored several vivid words ("sexy," "tuning," and "Goldilocks") that the models had discarded because I recognized their rhetorical value.

My goal was for readers to finish with both a smile and a genuine, if introductory, understanding of a complex subject, rather than the fatigue that often accompanies dense technical prose.

The editing process itself combined human judgment with AI assistance. After roughly twenty rounds of revision, I used Microsoft Word and Grammarly for grammar and spelling checks. These tools are valuable, but they routinely make stylistic suggestions I reject. They are aids, not substitutes for editorial judgment. Writing and editing this exercise took about five hours—work that might otherwise have occupied me for a week.

Writing is not simply the production of grammatical sentences. It depends on judgment, structure, audience awareness, and decades of reading, writing, and editing. Those are skills developed over a lifetime. Unlike many technical fields, editing benefits from long experience, which is one reason I believe my background is an asset for a Staff Editor role.

As a final experiment, I submitted my finished version to Grammarly's AI detector, which classified none of the text as AI-generated. Success! That’s one small step for AI, and one giant leap for humankind.

More broadly, I believe this workflow could be systematized at Voleon and tracked with project management software. Employees could use LLMs to brainstorm ideas, locate supporting evidence, summarize research, and produce early drafts. Editors would then reshape those drafts into polished communications while ensuring technical accuracy, consistency, and a distinctive institutional voice.

I would develop a house style guide in collaboration with leadership that establishes not only grammatical conventions but also principles of tone, clarity, and style. My preference is for intellectually rigorous prose that remains engaging and accessible, but the editor's role is to express the organization's voice rather than impose a personal one.

Finally, I see the Staff Editor as a collaborator throughout the publication process. Depending on an author's needs, I could help develop ideas, draft documents, perform substantive editing, line editing, copyediting, and proofreading, then coordinate revisions and approvals before publication. The objective is not to replace human writers with AI, but to combine AI's efficiency with the judgment, creativity, and accountability that only people provide.