Defining Generative AI/LLMs, Comparing Two Platforms

Generative AI and LLMs

Distinguishing Generative AI from other LLMs

Traditional AI models have always relied on existing information to recommend actions. An example is Netflix's movie recommendations, which relied on past interactions (Raj et al., 2024). Flagging of fraudulent activities has also been part of the traditional AI model practices. They still involved information that already existed to make predictions. Generative AI is different because it learns patterns from interactions to generate responses that did not exist. Even if there is information guiding the response, this innovation integrates creative output. Similar situations also lead to the generation of codes. Large Language Models (LLMs) are the ones that drive responses generative AI should develop. The role of LLMs is to remain neutral in evaluating sequences to develop human-like language in response to an interaction.

Key Terminologies

Prompting

A specific input guide that a user provides to an AI model regarding the outcomes they expect. There has to be clarification on everything the output should cover (Daud et al., 2026). Failure to provide precise, accurate directions results in vague output. Providing the background information and format of a response are other considerations.

Context Window

This is the short-term memory of an AI model. The maximum number of words it can remember in a given conversation. It can also represent the data that an AI model retains during a conversation (Daud et al., 2026). When the window is full, the user has to start another conversation. Failure to do so will lead to an AI generating inaccurate responses. It forgets messages it generated several prompts before.

Hallucination

This is a misrepresentation of facts that occurs in some AI model generations. An AI model can present incorrect data as the correct (Daud et al., 2026). Researching the facts of the matter confirms that it is indeed presenting wrong data as accurate information. One cause is a lack of sufficient data to answer a particular prompt. Most programmers focus on making models look accurate in their answers, even when they lack sufficient information. Asking about the information that training did not cover also causes hallucination.

Grounding

This process entails providing accurate information to an AI model to guide its responses. The practice helps counter hallucinations. AI can leverage existing data to produce accurate outputs (Daud et al., 2026). The strategy also involves training AI to develop accurate outputs. An example is the use of company data. Its access via AI helps ensure the model performs its role within relevant organizations.

Platform Comparison: ChatGPT-4o vs. Claude 3.5 Sonnet

ChatGPT-4o (OpenAI) is outstanding for brainstorming. It is also advantageous in the use of image and voice prompts for various tasks. Claude 3.5 Sonnet (Anthropic) has strengths in conducting technical analysis and reasoning. Use of Claude 3.5 Sonnet in a professional context is recommended because of its authentic tone (Mykhalko, 2025). The model's outputs are also concise. This situation leaves ChatGPT-4o useful for developing wordy outputs. There exists versatility in the responses it generates. The context window for ChatGPT-4o is lower than that of Claude 3.5 Sonnet. While ChatGPT-4o is about 300 pages, Claude 3.5 Sonnet's context window is about 500 pages (Mykhalko, 2025). Application of ChatGPT-4o in the business context has been beneficial because it supports custom app integrations. Claude 3.5 Sonnet's applications in similar contexts have included their best performance in coding accuracy. These differences indicate that each model has specific strengths and weaknesses. Improving on weaknesses can make each of them better.

Organizational Applications (3–4 Use Cases)

Implementing AI leads to the knowledge management function. It can serve as an internal librarian to facilitate easier data identification (Earley, 2023). An example exists when looking for a legal contract existing in the grounded data for the model. The second application is automated customer support. Modern businesses involve the model in meeting the functions of an agent. Businesses use the model to respond to customer concerns. Further applications include communicating CRM concerns to management. Coding functions are the third application. Businesses can develop software more quickly and with fewer errors thanks to the support it provides developers. The fourth application is in marketing. It can help a business quickly generate several marketing messages for sending to customers.

Risks and Constraints (2–3 Points)

The application of AI models can lead to misuse and data privacy violations. Company employees can feed a model confidential information about their company and use it to develop predictions about its future (Earley, 2023). The situation may have negative company implications for data exposures. There is also a higher risk of AI models misleading when they lack elaborate information. Instead of relying on facts, they develop outputs relating to the provided training. A failed training process can translate to inaccurate responses. The situation can damage the organization's reputation. Countering these risks helps ensure AI models are useful to different organizations.

 

Manager's "Getting Started" Tip Box

Implementing these applications requires starting from small activities within the organization (Earley, 2023). Use of the models in customer interactions can occur after the innovations have helped achieve small tasks, such as summarizing business meeting reports. Identification of specific information in a particular contract should also be another internal accomplishment before applying the AI models in external stakeholder interactions. Grounding should specify the data that an AI model should access. The strategy helps in countering the risk of confidential data misuse. Grounding also entails ensuring the data used to support AI outputs is accurate. Regular reviewing of developed AI functions is also necessary. It counters the risk of misrepresentation of facts due to training setbacks or other causes of hallucination.

 

References

Daud, A., Khan, M., Ghanem, S., Alesawi, S., Yonbawi, S., Alsini, R., ... & Gharawi, A. A. (2026). Generative AI: A Comprehensive Overview of Large Language Models for Prompt Engineering and Applications. Research Square. https://doi.org/10.21203/rs.3.rs-8809658/v1

Earley, S. (2023). What executives need to know about knowledge management, large language models, and generative AI. Applied Marketing Analytics, 9(3), 215-229. https://doi.org/10.69554/YQBV7690

Mykhalko, Y. O. (2025). Language-dependent performance of large language models in medical diagnostics: A comparative study of ChatGPT-4o and Claude 3.5 Sonnet. Polski merkuriusz lekarski: organ Polskiego Towarzystwa Lekarskiego, 53(6), 821-825. https://doi.org/10.36740/merkur202506117

Raj, S., Sharma, A., Saha, S., Singh, B., & Pedanekar, N. (2024). Transformative movie discovery: Large language models for recommendation and genre prediction. IEEE Access, 12, 186626-186638. https://doi.org/10.1109/ACCESS.2024.3482461

 

 

 

 

 

Prompt Iteration and Evaluation for a Workplace Communication Task

Artificial intelligence (AI) is being used by professionals in their daily workplace activities to help generate reports, summarize information and draft emails, among other tasks. When these tools are utilized effectively, they have the potential to greatly increase a worker's productivity. Nevertheless, the quality of the output produced by AI models can depend on the user providing clear and detailed descriptions of what he or she wants produced. As Al Naqbi, Bahroun, and Ahmed (2024) stated, the use of generative AI will increase workplace productivity if the user provides the AI model with a description of what is desired including all necessary context; this will allow the AI model to produce a product that is relevant to the user and of use. This assignment will illustrate how one can create and revise prompts to improve the quality of the output of an AI model for a workplace communication assignment.

Initial Prompt

Draft a memo/email to employees regarding the implementation of a remote work policy.

Output A

The AI model created a brief notice informing employees that a new remote work policy would be put into place shortly, and that further information would follow at a later time. The message did not include key details (e.g., start date, expectations, etc.) that are typically included in an employee communication related to a new policy. In addition, the message had a casual tone and was not suitable for business communication.

 

 

Revised Prompt

Write an email from an HR manager that includes a clear description of a new remote work policy that begins on April 1. Explain the remote work policy in the body of the email. Include an invitation for employees to contact HR with any questions they may have regarding the new policy.

Output B

The revised prompt resulted in the production of a more complete email that explained the new policy, included the date it would begin, and had a tone that was professional and acceptable for business communication. Nevertheless, the organization of the message was less than ideal and did not have distinct sections to break up the content.

Revised Prompt

Write an official e-mail from HR to all employees informing them of the new remote work policy going into effect on April 1. The e-mail should have an introduction; clearly state what the new policy entails; identify at least two expected behaviors from employees while they are working remotely (i.e., check-in with your team, meet productivity standards); and close the e-mail with encouragement to contact HR if the employee has any questions regarding the policy.

Output C

The revised prompt generated a structured and professional e-mail. The e-mail had an introduction explaining why the policy was implemented; it had a clear description of how remote work will operate; it had at least two expectations for employees (i.e., maintain open lines of communication, meet productivity levels); and it had a supportive closing urging employees to contact HR with any questions regarding the new policy.

Explanations of Improved Prompt Elements

• Revising the prompt provided additional context and identified the e-mail author as an HR manager, and specified when the new policy goes into effect, therefore generating a more applicable response.
• Incorporating information related to the intended audience (employees) made the e-mail more formal and professional.
• Adding clear structural guidelines generated a more organized and readable output.
• Incorporating examples of expected behaviors (i.e., communicating with colleagues, achieving productivity levels) created a more functional response to the actual needs of workplace communications.

The revised prompts illustrate the improvements in utility and clarity that result from revising the initial prompts to be more applicable for generative AI to assist in enhancing workplace productivity. According to Al Naqbi et al. (2024), the ability to effectively utilize generative AI is contingent upon thoughtfully designing prompts to allow the technology to provide greater assistance to enhance workplace productivity.

 

 

Workflow Infographic: Prompt Evolution

The changes in the prompt within the multiple iterations are shown in Figure 1 to enhance the quality of the AI-generated workplace communication. The first prompt was not contextual, and it gave an ambiguous answer. A start date and role context were added in the first revision, which made it more understandable. The last revision brought in order and behavioral norms leading to an orderly and professional work.

 

 

 

References

Al Naqbi, H., Bahroun, Z., & Ahmed, V. (2024). Enhancing work productivity through generative artificial intelligence: A comprehensive literature review. Sustainability, 16(3), 1166. https://doi.org/10.3390/su16031166