Recently, the fear of AI taking writing and other creative jobs has emerged bigger and badder than ever before. The worry that a smooth talking studio head will wheel-and-deal with a virtual Sam Altman, and emerge with a business model that requires negligible labor to produce movies.
To understand the impact of something like a large language model on movie-making, one should understand the underlying structures of machine learning, along with their corresponding challenges.
The first category of machine learning is regression. Regression is basically allowing the computer to find the various values of Y = Mx+ B, with multiple variables. Regression can be helpful to find continuous values. Regression can help predict what Johnny’s SAT score will be if he studies for 20 hours, and his previous score was 1150. This regression is done through a supervised process where the input variables are entered, and the correct output is included.
The second category of machine learning is classification. Classification aims to map input variables to categorical outputs, rather than continuous outputs. Classification would solve a problem like “Will Beth divorce her husband in Good Girls?” Similar to regression, this is learned through a supervised process where the input variables are entered, and the correct output is included.
A third category of machine learning is association, or clustering. Association tries to find and segment similar groups. Clustering would essentially be used to segment different types of people. A lot of people buy Stanley Cups, but may be react differently to its advertising. Utah mothers and construction workers should probably be on different advertising campaigns. Clustering would help sort people into the different advertising campaigns. Unlike the other two functions, clustering is oftentimes done without having explicit correct outputs identified.
The final category of machine learning is recommendation. Recommendation tries to give users recommendations based on their preferences. Netflix uses recommendation to predict what their subscribers would like based on how they’ve shared their preferences so far. Recommendation finds things based on their preferences that it thinks the user will enjoy.
My theory is that GPT’s writing function is largely regression. It likely has a writing score based on grammar and perhaps content. Given this theory, I predict that when I ask GPT to improve this essay through multiple iterations, the grammar changes will be insignificant, or GPT will change the underlying content (maybe part of good writing to GPT is writing about specific topic areas) to things more highbrow.
I will run this essay through Chat GPT 3.5 7 times, and put the first output and the seventh output below and label them. Finally, I’ll evaluate what I think the differences seem to be.
(1st iteration)
In recent times, the apprehension surrounding the potential displacement of writing and other creative jobs by artificial intelligence (AI) has reached unprecedented levels. There is a growing concern that shrewd studio executives might engage in negotiations with virtual personalities, such as a digital Sam Altman, to devise business models that require minimal human labor for movie production.
To comprehend the profound impact of large language models on the art of movie-making, one must delve into the foundational principles of machine learning and the challenges inherent in these structures.
The first facet of machine learning is regression, a process wherein computers discern various values of Y = Mx + B across multiple variables. Regression proves instrumental in predicting continuous values, as exemplified by its ability to forecast Johnny’s SAT score after 20 hours of study, given his previous score of 1150. This supervised process involves inputting variables and corresponding correct outputs.
The second category, classification, seeks to map input variables to categorical outputs rather than continuous ones. For instance, addressing a query like “Will Beth divorce her husband in Good Girls?” involves a supervised learning process, akin to regression, where input variables and correct outputs are inputted and learned.
The third category, association or clustering, endeavors to identify and segment similar groups. Clustering becomes valuable when categorizing diverse consumer demographics, recognizing that Utah mothers and construction workers may respond differently to advertising. Unlike regression and classification, clustering often operates without explicit identification of correct outputs.
The final category, recommendation, focuses on providing users with personalized suggestions based on their preferences. Netflix, for instance, utilizes recommendation algorithms to predict subscribers’ preferences and suggest content accordingly.
My conjecture is that the writing function of GPT predominantly aligns with regression. It likely assigns a writing score based on grammar and, to some extent, content. Given this hypothesis, I anticipate that multiple iterations of refining this essay with GPT may yield minimal changes in grammar, or GPT might alter the underlying content, possibly gravitating towards more sophisticated topics, as it perceives excellence in writing to be synonymous with addressing specific, higher-level subject matter.
(7th iteration)
In contemporary discourse, the growing apprehension surrounding the potential displacement of writing and creative roles by artificial intelligence (AI) has garnered considerable attention. A prevailing concern is that discerning studio executives may engage in negotiations with virtual personalities, such as a digital representation of notable figures like Sam Altman, to devise business models that necessitate minimal human involvement in movie production.
To comprehensively grasp the profound impact of large language models on the art of filmmaking, it is imperative to delve into the foundational principles of machine learning and the inherent challenges embedded within these structures.
At the core of machine learning lies regression, a process wherein computers discern various values of Y = Mx + B across multiple variables. Regression plays a pivotal role in predicting continuous values, illustrated by its ability to forecast Johnny’s SAT score after 20 hours of study, factoring in his prior score of 1150. This supervised process entails inputting variables and their corresponding accurate outputs.
The second facet, classification, seeks to map input variables to categorical outputs rather than continuous ones. Addressing a query such as “Will Beth divorce her husband in Good Girls?” involves a supervised learning process akin to regression, where input variables and accurate outputs are provided and learned.
The third category, association or clustering, endeavors to identify and segment similar groups. Clustering becomes invaluable when categorizing diverse consumer demographics, acknowledging that Utah mothers and construction workers may respond differently to advertising. Unlike regression and classification, clustering often operates without explicit identification of correct outputs.
The final category, recommendation, focuses on providing users with personalized suggestions based on their preferences. For instance, Netflix employs recommendation algorithms to predict subscribers’ preferences and suggest content accordingly.
My conjecture posits that the writing function of GPT primarily aligns with regression. It likely assigns a writing score based on grammar and, to some extent, content. Given this hypothesis, I anticipate that multiple iterations of refining this essay with GPT may yield minimal changes in grammar, or GPT might alter the underlying content, possibly gravitating towards more sophisticated topics, as it perceives excellence in writing to be synonymous with addressing specific, higher-level subject matter.
I then wrote a query to compare the two essays and here’s the results of the comparison:
The two essays share a similar structure and cover the same topics, exploring the impact of large language models on the creative industry and delving into the foundational principles of machine learning. However, there are notable differences in the language, style, and emphasis of certain points. Here’s a summary of the key distinctions:
- Language and Style:
- Essay 1 uses more technical and formal language, employing terms like “facets of machine learning,” “regression,” “classification,” “association or clustering,” and “recommendation.”
- Essay 2 maintains a formal tone but opts for slightly more accessible language, using phrases like “at the core of machine learning” and “the growing apprehension surrounding.”
- Phrasing and Sentence Structure:
- Essay 1 tends to have longer and more complex sentences, utilizing a more intricate sentence structure.
- Essay 2 uses slightly shorter sentences and a more straightforward structure, making it potentially more reader-friendly.
- Detail and Elaboration:
- Both essays cover the same categories of machine learning (regression, classification, clustering, and recommendation), but the level of detail and elaboration varies.
- Essay 1 provides more detailed explanations of each category and includes examples to illustrate the concepts.
- Essay 2 offers concise explanations, presenting the information in a more condensed form.
- Emphasis and Focus:
- While both essays touch upon the potential displacement of creative roles by AI, Essay 1 places more emphasis on the impact of AI on movie production and the negotiation with virtual personalities for business models.
- Essay 2 maintains a balance between discussing the displacement of creative roles and delving into the foundational principles of machine learning.
- Repetition:
- The two essays are almost identical in content, with some sentences and phrases being repeated verbatim.
In terms of which essay is better, it depends on the audience and purpose. Essay 1 provides a more in-depth exploration of machine learning concepts, making it suitable for a technically inclined audience. On the other hand, Essay 2 may be considered more accessible and reader-friendly, potentially making it preferable for a broader audience. The choice between the two may also depend on the specific requirements of the assignment or the preferences of the reader.
Conclusion (human generated):
Judging from the differences it summarized and a quick review, the main differences I took from this essay is that through additional iterations, the writing gradually became simpler. This suggests that my theory about underlying content was incorrect, and that Chat-GPT over time eventually started to value simplicity through additional iterations.
I asked Chat GPT why it revised an essay this way. Here’s what it wrote:
The revised essay incorporates several improvements in terms of clarity, conciseness, and sophistication. Here are some key differences and the reasons behind the changes.
- Clarity and Conciseness:
- I revised sentences for clarity and simplicity, making the information more accessible to a broader audience.
- The language was streamlined to eliminate unnecessary complexity, making the text more straightforward.
- Word Choice and Expression:
- Synonyms and varied expressions were used to enhance vocabulary and maintain engagement.
- The revised essay employs more precise and impactful language to convey ideas effectively.
- Sentence Structure:
- Sentences were restructured for improved flow and coherence, facilitating a smoother reading experience.
- Long sentences were broken down into shorter, more digestible segments for better comprehension.
- Professional Tone:
- The tone was refined to maintain professionalism and align with formal writing standards.
- The language was adjusted to enhance the essay’s overall tone and convey a sense of authority.
- Consistency and Parallelism:
- Consistency was maintained in terms of terminology and formatting throughout the essay.
- Parallelism in sentence construction was improved to create a more polished and cohesive piece.
- Precision in Technical Descriptions:
- Technical descriptions, particularly regarding machine learning categories, were refined for accuracy and precision.
- The revised essay maintains a balance between technical detail and accessibility for a wider audience.
- Enhanced Engagement:
- The language was adjusted to enhance engagement and maintain the reader’s interest.
- Changes were made to create a more compelling narrative and keep the reader focused on the content.
- Elimination of Redundancy:
- Redundancies and unnecessary repetitions were removed to ensure a more efficient and impactful communication of ideas.
The goal of these changes was to elevate the overall quality of the essay by improving clarity, precision, and engagement while maintaining the integrity of the original content.
(Me again): In short, it started to simplify when it thought the writing quality was otherwise there
Isadore Johnson's Website 