The Taiwan Banker

The AI craze has gained great traction over the past five years, reaching a new peak after it became a mainstream topic in the media and on social media. In order to increase financial practitioners’ familiarity with ChatGPT, TABF has invited Lai Zhao-rong to speak on its potential uses in finance. Mr. Lai told us that there is a future for ChatGPT and generative AI in data analytics and prediction. He believes that it is necessary to understand AI before one can understand ChatGPT, and that AI is defined as, “a tool or program with which users can interact, with apparently human responses.” Reasonable, but incorrect answers AI is defined by this Turing test. Lai stated, “The responses and feedback of AI allow the user to think they are interacting with a real person, not a robot.” One factor driving the popularity of ChatGPT is that its responses are becoming practically human. The biggest advantage of generative AI is its ability to generate data. Lei points out that ChatGPT’s responses include words and data not input by the user. “One could even say that ChatGPT is like a chatbot or a word game.” ChatGPT is a Natural Language Processing (NLP) technology which uses machine learning and deep learning. Natural language refers to human language, rather than programming languages. NLP can allow computers to understand, analyze, and generate natural language, and is already used in customer service chatbots and in smart home technology. What is a chatbot? ChatGPT focuses on users’ prompts to produce new information. For further prompts, ChatGPT answers according to the second question, and so on in the manner of a human conversation. “The AI can only give reasonable answers, not correctanswers.” “Some people are not satisfied with the responses given by ChatGPT’s, or see them as ambiguous, evasive, or even silly,” inconsistent with what users think is the correct answer. “Some users also give it strange questions to which there is no ideal answer.” Furthermore, ChatGPT is not a search engine, may be unable to come up with a good to questions related to people, events, time, or places. Lai observed that most Taiwanese people use the free version of ChatGPT 3.5; only a few users and companies are willing to pay for the more advanced ChatGPT4.0. “In terms of the quality of responses, 4.0 greatly exceeds 3.5.” The newer version is able to browse web pages and cite information from articles. The newer version is not only far superior, but also has more adaptive functionality. It is now accessible as a plugin: it can watch movies, search local computer files,and even make PDFs. Due to the breadth of analysis, ChatGPT 4.0 tends to take longer to respond to prompts. Don’t forget to delete sensitive prompts “Before using ChatGPT, users should remember at all costs to only enter personal information, corporate secrets, or sensitive information when it is absolutely necessary.” Lai warns, if users must ask about sensitive data, it is important to quickly delete the conversation afterward in order to prevent data leaks. “Furthermore, even if the information is deleted, ChatGPT may retain the information for up to a month.” How is data turned into AI? Lai recommends using Google’s standard operating procedures, which include seven steps: data collection, data preparation, model selection, model training, model evaluation, parameter optimization, and prediction. “AI’s most awaited function is to self-execute predictions. The seven steps all lack one thing: statistics.” “The difference between statistics and AI is that the former relies on human analysis and forecasts, while the latter uses formulas to calculate trends and make predictions.” Lai says when following Google’s seven steps, most people neglect the importance of data collection and the quality of data collected, which bleeds into the final product. “Users should remember to properly prepare the data. It is important to delete deficient information, turn words into variables, and rearrange the data into the pattern of a model.” After collecting and preparing the data, users should choose a suitable model based on their objectives: if a user uses AI Cloud data to analyze stocks, for instance, they could also rely on ChatGPT’s model. ChatGPT is Open AI’s most developed large-scale language analysis model. The source of the name is Generative Pre-trained Transformer: a pre-trained model that can save users’ time. “To create an cloud-based AI stock analysis robot, users should gather related Application Programming Interfaces (APIs) and draw data from cloud databases.” Lai noted that within every banking index and market, stock information, news reports, and all other necessary data is available. If the API is insufficient, “users can deploy their own web crawlers (automatic indexers) to search for data on the internet.” The purpose of an API or crawler is to automatically pull data from the web, but they still may be unable to compile all related data, depending on the openness of each data source. In Taiwan, for example, Taipei is significantly more open about its data than other counties, making it difficult to compare. Special applications of ChatGPT After gathering sufficient data, users should analyze it. Lai explains that users can take advantage of ChatGPT, BERT, CKIP, or similar programs using programs to evaluate data such as Chinese word breaks and sentiment analysis. Sentiment analysis means analyzing a paragraph or article for positive, negative, or neutral undertones. “Users who want to post text or an article should ask ChatGPT to analyze the tone. ChatGPT tends to be fairly accurate.” To create an cloud AI stock analysis robot, one should analyze at least a year’s worth of sentiment data as the basis for a cloud computing, web framework, image, instant notification, situation monitoring, or keyword search function. In addition to open-market data and news reports regarding listed companies, comments by social media users can include relevant financial information. Lai noted that different countries used different use finance-specific vocabulary. “Retail investors can immediately draw attention to market phenomena and market sentiment, which ChatGPT can process.” “Although ChatGPT is a trained language analysis system, users can still continue to train it, making it better suited to handle future prompts, such as banking analysis and predictions. Some companies will also use it as a safety control.” ChatGPT 3.5 is free, but not open source. “Users can pay for ChatGPT4.0 and use it for more advanced purposes, or exchange it for other software.” Nevertheless, Lai says, most people still use ChatGPT for fun or to sample its functionality, so demand for the paid version is not overwhelming. He suggests that ChatGPT 3.5 users not only ask very clear questions, but also introduce clear information: the more thorough and complete, the better. One should not focus on if parts are repeated. “If two users both ask ChatGPT ‘What is intelligent money management?’ the first may get an incomplete answer, but the second may be offered theories about money management, one answer being considerably more useful than the other.” Changing lifestyles and workplace dynamics Besides just responding to questions, ChatGPT can also be used to summarize articles. From Lai’s personal experience, it can not only summarize Chinese and English articles, but also translate the summaries from English to Chinese. “When using ChatGPT, one should not act with haste. If you set out the steps to the answer, it will give a longer and better answer.” Lai emphasizes that clear instructions, strictly explaining the key points and sources of the data in the prompt, and asking for a specific output, will improve the results. “However it is still important to remember that clear instructions will only help improve the reasonableness of the answer, not its accuracy.” “Interest in this emerging technology has skyrocketed, creating a ‘feast for the eyes.’ It will abruptly change lifestyles and the workplace environment alike, and finance is no exception.” Lai predicts that some industries may avoid using ChatGPT because experienced professionals are not accustomed to its proclivities; yet the AI surge will continue growing deeper and broader, carrying mankind into a new era.