GENERATIvE AI IN MANAGEMENT TODAY AND TOMORROW Pobrane z czasopisma Annales H Oeconomia Data: 04/01/2024 12:33:30

132JERZY KORCZAK, ILONA PAWEOSZEKvaluable information regarding the meaning and structure of the words. It aims to provide the model with comprehending the context. The feedforward layers in LLM consist of several interconnected layers that apply non-linear transformations to the input embeddings of tokens or sentences. These layers let the model grasp more abstract concepts from the input text. The recurrent layers are designed to interpret information from the input text sequentially. These layers maintain a hidden state that updates each step, enabling the model to capture the relationships between words in a sentence. Another crucial component of LLM is the self-attention mechanism. It allows the model to focus on different portions of the input text selectively. It will enable the model to weigh the importance of different words (for example, depending on their sentiment). This mechanism helps the model prioritize the most relevant parts of the text, leading to more accurate answe

GPT-4 and LaMDA work on similar architectures; however, there are some key differences between the two models, notably in: Purpose: LaMDA is designed for conversation, while GPT-4 is intended for generating text. LaMDA is trained on a dialogue dataset, while GPT-4 is trained on a dataset of text and code. This means that LaMDA is better at generating natu-ral-sounding conversation, while GPT-4 is better at generating creative text formats, like poems, code, scripts, musical pieces, emails, letters, etc. Size: LaMDA has 137 billion parameters, while GPT-4 has 175 billion. This means that GPT-4 is slightly larger than LaMDA and has more learning capacity. However, it also means that GPT-4 is more computationally expensive to train and use. Availability: LaMDA is currently in limited release, while GPT-4 is available to the public. However, LaMDA is free, while GPT-4 requires a subscription. Strengths: LaMDA understands and responds to factual queries, GPT-4 gen-erates creative text f

Weaknesses: LaMDA can be robotic or repetitive, GPT-4 can be inaccurate or biased.Currently, LLMs cannot be considered comprehensive decision-making systems but only as supporting tools. Managers need data and knowledge from many sources not subject to language modeling. These sources include financial reports, competitive benchmarks, business logic, and decision-making rules. The intelligent dialogue sys-tems evolution involves integrating decision-making capabilities to enhance function-ality and provide more efficient and effective responses. For this purpose, it would be necessary to implement a managers knowledge model to generate adequate responses. The managers knowledge model can provide greater content relevance to fill gaps and reduce cognitive bias. Another expected improvement of LLMs is their integration with the organizations knowledge and data resources. This will make LLM a personalized tool with a high degree of grounding in facts necessary to make informed dec