Search Results

Here's an experiment with ChatGPT. I've asked it to re-write my blog in the vernacular. Is it easier to understand like this? The clarity it brought to the message was surprising, leading me to wonder: has our shift to digital communication altered our language standards? If it has then using ChatGPT to write like this may be more effective than expected. Here's the blog: You might think you gotta be some kind of genius to get the hang of this AI stuff, but that's way off base. You don't need to be some kind of "Prompt Guru" or anything, just gotta be curious about what it can do for you. I wrote this especially for folks who haven't had the chance to mess around with ChatGPT or anything similar. Sure, the tech behind it is pretty complex – I'll be the first to admit I don't get all the ins and outs – but that hasn't stopped me or anyone else from using it. If you're good at firing off emails or texting, you're already set to get good at using AI in no time. Here, I'll walk you through some simple tricks to find what works best for you. You can ask ChatGPT to whip up something for you. Just type in what you're thinking and let it do its thing. Usually, you'll get back something pretty decent, even if it's not exactly in your style. Want something tweaked a certain way? Ask ChatGPT to redo it – maybe in a lawyerly way, or like a journalist, or even in slang. Check out what it gives you and see if it hits the mark better than your first go. What you've got might already be spot on, but maybe you wanna make it more your own. Just copy what ChatGPT gives you into a Word doc and tweak it to your heart's content. Here's a trick I find super useful: say what you wanna say into a Word doc, then copy that over to ChatGPT. You can ask it to just clean up the spelling, grammar, and syntax without changing anything else. Or, you could ask it to jazz it up – make it simpler, more advanced, or just better in general. As you can see, it's not rocket science. Just chat with ChatGPT like you would with a coworker. You can't mess it up. The more you use it, the easier it gets. It's a real time-saver, and I reckon you'll be pretty chuffed with what it can do if you give it a whirl.

One of the misconceptions about generative AI is that you need to be a rocket scientist to use it, but nothing could be further from the truth. You do not need to be a "Prompt Engineer" but just curious about what it can do for you and how it can help you. I've written this primarily for those who have not had the chance to use ChatGPT or similar tools. While the software itself involves a huge amount of complexity that I confess I don't fully understand, it doesn't stop me or anyone else from using it. If you can write emails or send texts, then there is no reason why you can't become proficient at using AI in a short space of time. Here, I will explain a few of the simple techniques that you can use to explore what works for you. You can ask ChatGPT to write something on your behalf. Just type into the prompt what you want and let it go ahead. What comes back is normally satisfactory, although it may not be exactly to your taste or style. https://chat.openai.com/share/1861bc73-ad0f-4e18-a1d6-edc873474bff Ask ChatGPT to rewrite the piece in a particular manner, for example, as a lawyer or as a journalist. Alternatively, ask it to write in a particular style, like the vernacular. You can then consider the results and see if they are an improvement on what you originally had. https://chat.openai.com/share/9fdeb055-50f2-4b98-94f2-dc7a6793c987 What you have done so far may already be enough, but perhaps you want to personalize it a bit more. You can cut and paste the output from ChatGPT into a Word document and make the alterations as you wish. Something I find particularly helpful is to dictate what I wish to say into a Word document and then copy it straight into ChatGPT. You can then ask it to correct the spelling, grammar, and syntax, leaving the rest unchanged, and it will give you the new version as requested. Alternatively, you can ask it to improve and make changes that make the document easier to understand, at the level of a 10-year-old, at the level of a graduate, or just ask it to make improvements as it sees fit.https://chat.openai.com/share/bd218094-9d15-415e-9c76-5bc3712ecf5a As you can see, there is very little to it. Just converse with ChatGPT as you would a work colleague, and you can't really go wrong. The more you use it, the easier it becomes. It will save you a lot of time, and I think you will be impressed with the results if you've not tried it.

At 97, my mother faces the challenges of old age, including complete hearing loss, but she still enjoys reading. I've observed that elderly individuals often take pleasure in reminiscing about the past. This led me to experiment: could I develop a GPT model that would bring joy to my mother? The result was "Mum's Diary," a customized GPT designed to generate a hypothetical diary page from exactly 50 years ago, incorporating people and events from that time that she knew. It creates a narrative of what she might have done, including a small picture related to the diary content. How It Works With this custom GPT, you simply enter the prompt "let's go," along with a picture, and a diary entry that my mother could have written 50 years to the day materializes. It's remarkably straightforward. Building It Is Simple To create your custom GPT, you need access to GPT-4 (the paid version) and to follow the provided instructions: No programming skills are necessary. I specified that I wanted a diary entry reflecting my mother's activities and thoughts from 50 years ago, including interactions with people and typical daily activities. I requested the inclusion of relevant local, national, or global events and their impact on her, alongside background information about her job, family, neighbours, shopping habits, holidays, car, weekend activities, pets, etc: This is what it looks like (and every time you prompt you get a new page)*: My Mother's Reaction It's too early for a definitive opinion, but she is intrigued by the generated content and enjoys reading it. This project has painted a vivid picture of her past activities, proving to be a delightful experience. She has commented on various historical accuracies and discrepancies, which can be easily adjusted by modifying the custom gpt. Further inquiries into improving the custom GPT (by asking GPT-4 itself) suggested that adding more detailed information relevant to the specific period would enhance the accuracy and reliability of the output. Although some generated storylines may seem repetitive, this doesn't concern me much at the moment. I believe this can be addressed by adjusting the instructions to vary the content generation, which I am currently exploring. Potential Applications While this custom GPT is tailored specifically to my mother's life, making it unsuitable for generic use in the GPT store, its concept could benefit others. By inputting personalised details, it could serve as a unique diary for another individual, potentially aiding memory recall and mental stimulation for elderly people either at home or in care facilities. This application of generative AI could add a little something to improve the quality of life for many. Conclusion In conclusion, this project exemplifies a simple yet powerful way to utilise generative AI to enhance people's lives. If you found this interesting, please share it with others who might benefit. Should you require assistance in developing a similar GPT, feel free to contact me. ---------------------------------------------------- * The complete diary entry for the day: Diary Entry: February 16th, 1974 Today was a particularly reflective day for me, full of small but significant moments that reminded me of the delicate balance of our daily lives here in Yattendon. The morning air was crisp as I made my familiar walk down to the school, the quiet of the village broken only by the distant sound of Flynn's tractor in the fields and the cheerful greetings from Mr. Peaty as he made his rounds. In school, the children seemed more spirited than usual, perhaps the result of the recent news that had the whole village talking: the miners' strike and the resulting three-day work week had finally come to an end. Mrs. Gaskell and I had a lengthy discussion about it during our morning break, pondering over the impact it had on the local community and the country at large. It was a relief to know that, for now, the crisis was over, and there was a collective sense of hopeful anticipation for what lay ahead. After school, I stopped by Mrs. Hogg's post office to pick up a few essentials. The lively barks of her two pugs greeted me before I even stepped inside, bringing a smile to my face. Mrs. Hogg shared the latest gossip as she always does, mentioning that Mr. Behr had decided to host a small celebration at the Royal Oak to mark the end of the power cuts. The community needed something to cheer about, and this seemed like the perfect occasion. On my walk back home, I couldn't help but marvel at the beauty of our village. The sight of our blue Land Rover parked outside Pound Cottage, with Chilly lounging lazily in the sun nearby, filled me with a deep sense of contentment. Ivan was in the garden, tending to his flowers, a pastime that had become more than just a hobby for him. We exchanged stories of our day over a cup of tea, reflecting on the importance of community and resilience in times of uncertainty. This evening, as I sit down to write these words, I am reminded of the strength and perseverance that define us, not just as individuals but as a community. The challenges we face, whether they're nationwide crises or personal struggles, have a way of bringing us closer together, strengthening the bonds that tie us to this place we call home. As I close my diary for the night, I am filled with gratitude for the life we have built here, for the friends and neighbours who enrich our lives in countless ways, and for the simple, enduring pleasures of village life. Note: Chilly was a pet cat. Just typing in a different date into the prompt generated an entry from a different period.

In previous articles, I've discussed some of the economic benefits of using LLMs, as well as the problems concerning hallucinations and how simple steps can mitigate their effects. After learning about the applications of LLMs in treasury and risk management, you might want to try them yourself. However, if you work for a bank or building society, it may not be that straightforward. Many firms have prohibited the use of these models. Let's explore what's causing this and whether there are steps we can take to move forward. (The following discussion concerns the use of LLMs in treasury and risk management. This means that the model's use predominantly applies to information used internally and, in the business, not externally with customers). Challenges in Implementing LLMs Why have some firms not allowed their employees to use GPT-4, or such, at work? This list is not exhaustive, but you may recognise some of these challenges in your own business. Let’s look at three issues of concern: Risk Management: This includes legal issues, data security, privacy, and the risk of misinformation from hallucinations. It's about the apprehension of potential negative consequences of using LLMs, including legal complications, data breaches, and regulatory non-compliance. In sensitive businesses like financial institutions, these risks require careful appraisal. At no point can you say the risk is nil and that's what frightens people, how can you quantify what you are getting yourself into? You can't it's a matter of using judgment - something we do all the time in financial markets. Technological Understanding and Trust: A lack of knowledge and a gap in understanding of what LLMs are capable of, their potential benefits, and limitations. This lack of familiarity or mistrust results in a wait-and-see approach, with decision-makers hesitant to adopt new technology until it is widely accepted within their peer group. Organisational Culture and Resistance to Change: This is related to operational inertia and a reluctance to adopt new technology, especially when integrating LLMs into established workflows. Many traditional firms resist change, not just because of a lack of immediate perceived benefits but also because it suits the short-term interests of those involved. Practical Steps The economic argument for using Large Language Models (LLMs) is compelling. Even the simplest application of these models to improve workflow can boost efficiency by at least 10%, leading to savings in headcount and costs. If you are experiencing resistance, three practical steps may help: Understand where the blockage is: Identify the concern. Who is raising it? And, what do they need to know to become comfortable? Address individual use cases: For example, in treasury, if you want to use an LLM to prepare for ALCO meetings, put forward a case. Show how an LLM can repurpose something you write for a different audience - as I've discussed in a previous article. Form an AI committee for collaborative decision-making: Initiating an AI committee within the firm brings together diverse viewpoints. This committee will serve as a platform to assess AI's varied applications and implications in the business and is a path towards drafting an AI policy – something regulated industries need to take seriously. Need for Policy The catalyst likely to trigger change is whether your peer group starts using the technology. There's no doubt that adoption will accelerate once this happens. Starting now and putting things in writing helps to think through the unique circumstances affecting your firm. As these technologies evolve, expertise is key to enhancing the workflow and efficiency and maintaining a competitive market position. Security will be high on the agenda, including the potential for accessing LLMs through an Application Programming Interface (API) and ensuring that sensitive data is not used for training by the LLM provider. In collaboration with the IT department, the committee will need to assess encryption methods. These methods are essential for securing data transmitted over the Internet, particularly when accessing LLMs through APIs. Furthermore, the committee will need to oversee data storage, especially with external cloud service providers like Amazon, Microsoft, and Google, as well as LLM providers such as OpenAI and Anthropic. There is also a need to be aware that data models (like LLMs) could exhibit bias. Could this translate into active decision-making without being aware? Ironically, firms that currently don't allow LLMs to be used can't stop employees from using these models to solve work problems at home. By not addressing the issues, the business has, by default, turned a blind eye to what employees are doing. That's what I call a risk.

Large Language Models (LLMs) in Treasury and Risk LLMs, or Large Language Models, are advanced artificial intelligence systems designed to understand, generate, and interact with human language, mimicking natural human communication. They are a subset of machine learning models within the field of natural language processing (NLP). In previous articles, I have discussed their use in various tasks, including: Streamlining Communication Writing for the recipient Undertaking research Stress and scenario creation Preparing for committees & analysing risk Adding narrative to presentations and record-keeping These tasks, which we regularly undertake, are time-consuming and, in some cases, repetitive. LLMs excel here, thereby freeing up our time and thought, allowing us to concentrate more on what is important and productive. Even if no further progress were made in developing these models, their impact on treasury and risk management, and indeed the whole firm, would be significant. We are only just learning how to use them in business. Key Findings from the UK Department for Education Report In November of last year, the UK Department for Education published a report titled "The Impact of AI on UK Jobs and Training." This report outlines AI and LLMs' influence on the UK labour market. The key findings are: Occupational Exposure: Professional occupations, especially in finance, law, business management, and teaching, are more exposed to AI. Large language models are particularly relevant in these fields. Sectoral Impact: The finance and insurance sector is most exposed to AI, followed by the information and communication, professional services, property, public administration, defence, and education sectors. Geographic Variations: Workers in London and the South East have the highest AI exposure, reflecting the concentration of professional occupations. The North East has the least exposure, but overall geographic variation is smaller than that observed across occupations or industries. Qualifications and Training: Employees with advanced qualifications, especially in accounting, finance, and economics, are typically in jobs more exposed to AI. Conversely, those with qualifications in building, construction, and transportation operations are least exposed. The analysis emphasises that AI is likely to augment rather than replace many jobs, with varying degrees of exposure across different occupations, sectors, and geographic regions. Upon first reading, I was surprised at the level of impact AI could have on us. However, after reflection, LLMs allow us to produce more output from the same resources or the same output from fewer resources. The implications are profound in terms of the workplace and more widely society. A Ridiculously High Return on Investment My own experience with LLMs has been more of an experiment in what they can do, particularly in the field I am familiar with. A simple conclusion I have come to is that the $20 subscription to GPT-4 is probably the best investment a bank or building society can make for its employees in Treasury and Risk. At first glance, this may seem extravagant, but a closer look makes it hard to ignore. One of the highest costs for financial firms is employing staff. With this in mind, let's look at things further. I've made some broad assumptions, so what follows is an estimate, a conservative one at that, because I have only considered the use of LLMs for document preparation. I’ve assumed an hourly rate for a skilled person in this area of £40/hour (based on a 46-week year, working five days a week, eight hours a day, totalling 1,840 hours per year, with a remuneration package of £75,000). Let’s also assume that 10 hours a week are spent on writing and processing documents, a common scenario given the workload for internal meetings and management reporting. From my experience, GPT-4 can easily halve the time taken for this, saving 5 hours per week. Over the year, this amounts to 230 saved hours with a value of approximately £9,200 or 12% of the salary bill. Even if you deduct the 20-25 hours needed to get proficient in using LLMs in your business (a process that can be significantly accelerated with the appropriate training). The first-year cost saving is over £8,000 or 10%. A More Positive Outlook You can now see why the report from the Department of Education highlights the implications of AI on banking and finance, especially in the southeast of England. While this may be true, I see a more positive impact on the workplace. Colleagues working in treasury and risk are constantly under pressure to deliver and things fall behind. The solution, hiring more staff, can be tricky, and many firms have headcount limits. Spending a small amount on LLMs will go a long way to alleviate some of these problems without breaking any headcount restrictions. It's also that much easier to sign up for an LLM than to go through the process of hiring.

What they are and what causes them Hallucination is a term used to explain the phenomenon of a Large Language Model (LLM) producing incorrect outputs. LLMs, like GPT-4, are trained on vast datasets consisting of text from the internet. They learn to predict the next word in a sequence, making them powerful tools for generating human-like text. However, this strength also leads to a key vulnerability known as 'hallucination,' where the model generates plausible but factually incorrect or misleading information. This often stems from the model's reliance on patterns in the data it was trained on, rather than an understanding of factual truth. Hallucinations are caused by: Overgeneralization: LLMs sometimes overgeneralise from their training data, leading to outputs that are too broad or generic, and at times, inaccurate. Data Quality: The training data may contain inaccuracies, biases, or outdated information, which the model can inadvertently learn and replicate. Context Limitation: LLMs have a limited context window (the number of words they can consider at one time), which can result in missing or misunderstanding key parts of an input query, leading to incorrect outputs. Lack of Real-World Understanding: Unlike humans, LLMs don’t have real-world experience or common-sense reasoning. They operate purely based on statistical patterns in the data they’ve seen. Not entirely detrimental In the context of treasury and risk, hallucinations are seen as detrimental because they produce unexpected and potentially erroneous results. However, it's worth asking whether hallucinations are intrinsic to the nature of predictive text-type models, which are essentially statistical models that generate output based on gathered data. We can prompt this in various ways to stimulate the output, and I believe the process of hallucinations is almost a natural part of LLMs. In many respects, we can benefit from this when considering risk management. Many of the reports we run, whether linear or scenario-based, are generated by our ingrained biases, our thoughts on how the world works, and regulatory norms. A savvy risk manager should consider using these models to explore some of the more nuanced features of risk in the business by questioning whether the reports currently being generated have any obvious holes or biases within them. Working with LLMs to generate specific stress-type scenarios can help investigate whether there are weaknesses in the business model. To this end, I've specifically addressed this issue in an earlier article in this series titled "Simple Uses for LLMs - Stress and Scenario Creation," which uses examples to create different stress scenarios and mitigants that can arise. When using these LLMs in the context of treasury and risk management, the issue of over-generalisation tends to be more prevalent when you start from scratch and ask the model to generate output without giving much thought to the prompting process, something we have control over. Nevertheless, I have regularly experienced overgeneralisation in the responses from GPT-4, particularly when I've asked broad questions about risk. The data quality issue is pertinent when dealing with treasury-type issues. The general model used by the LLM has been trained from the internet, and you have to be wary of some of the potential outputs. In the case of conducting specific analysis for the business, it is likely that you will be including your data either through an API or in the prompt window. This means that the data needs to be clean; there's nothing new in this – garbage in, garbage out. The issue concerning context limitation is one that I frequently experienced earlier last year. However, the new and updated models are now capable of much longer context windows, and I see this as a less important issue going forward. Experience counts Lack of real-world experience and understanding is a problem for these LLMs. They currently do extremely well in analysing situations, but they often miss the nuances that we see and have gained experience in while working in treasury and risk management. In extreme cases, the output can range from amusing to outright harmful. Let's illustrate this with an example where I've experienced a hallucination using GPT-4, and discuss some of the implications and consequences. When analysing a gap report, GPT-4 focused on liquidity risk management and prioritised it. This was somewhat surprising, and although insightful, it wasn't quite what I had in mind. Without my experience in markets, it would have been difficult to detect the subtle difference between liquidity and interest rate risk, as the two are often closely related but not the same. To overcome the problem, a specific prompt focusing on interest rate risk was necessary. At a later juncture, I asked GPT-4 to provide a delta report. Normally, you would expect a delta report to represent a shift in the yield curve leading to a change in the present value of the portfolio or balance sheet in question. GPT-4, however, provided a report based on the nominal amounts of the balance sheet rather than the risk amounts. A novice might accept this at face value, leading to significant trouble. Following it blindly, your interest rate risk analysis would be entirely wrong, and acting on it could lead to hedging non-existent risks, exacerbating the situation. To add more detail, I asked GPT-4 to provide a swap overlay onto the balance sheet risk for an institution, and the result was incorrect. Following it would be financially risky, exposing you to massive market exposure and blowing your limits out of the water. Relying strictly on the recommendations of GPT-4 could quickly damage your credibility. Despite these issues, the problem of hallucination seems to be far less prevalent than it was 12 months ago when we first started using these models. The current set of models and their training appear to have reduced many of the possibilities for hallucinations. However, when using an LLM, you need to be aware that it can produce outputs that seem wholly correct and plausible when they are indeed false and inaccurate. This can be potentially embarrassing, if not dangerous, for organisations, especially for financial services firms regulated by the PRA or FCA. Credibility at stake When considering the use of LLMs in treasury and risk management, it's important to note that we are typically not customer-facing. The users we work with are usually internal, digesting financial information to make business decisions. Thus, the issue becomes one of credibility. If you create content or work using LLMs and then disseminate that within the organisation, such as in a management meeting, and it is wrong or found to be incorrect, your credibility is at stake. This, I believe, is a major factor in exercising caution when producing output with LLMs. It would be unwise to allow a junior staff member with little experience to use these models for generating content that directly feeds into senior management. Often, they are unable to discern whether the output is correct or whether it is misleading or factually inaccurate. Senior management, provided with incorrect information, would not only be annoyed but would also likely take action. Indeed, such errors could put the organisation at risk, especially if they lead to changed management decisions or result in misleading or inaccurate reporting to regulators. At this stage whatever you produce, you must read and thoroughly understand it before using it in the workplace. Colour coding When using LLMs in treasury and risk management, a clear and open discussion about their use is necessary. Initially, we should question the output until we are entirely comfortable with its accuracy and reliability. For instance, if LLMs are used in generating reports for an ALCO meeting or a board meeting, the text could be colour-coded, or a graph could be annotated to indicate that an LLM was used in the generation. This transparency is helpful as it not only demonstrates the power of LLMs in the business but also draws attention to their usage, prompting caution until we are entirely comfortable with the material. Parallel running I also suggest that it is important to run processes and risk management systems in parallel with LLMs, rather than substituting them, until you are entirely comfortable with what they are generating. This parallel running is a “compare and contrast” exercise and may help highlight either anomalies in what the LLM produces or gaps in the information provided by existing legacy systems. For example, if you were producing a liquidity report, you might ask the LLM to comment on the liquidity situation and add this to the regular information you generate before fully transitioning from one approach to another. Additionally, I wouldn’t overload reporting with LLM-generated material too quickly, as it would be difficult to trace exactly what is being generated and to compare the quality and consistency of the new with the old. Use the three lines of defence Another way to manage the introduction of LLMs in treasury and risk management is to ensure there is independent oversight of these models. We use a three-line-of-defence approach to many aspects of our operations, and this can be valuable in assessing whether an LLM inadvertently exposes us to more risk. There needs to be an open and frank discussion about the use of LLMs in treasury and risk management, including with senior management and risk management. Risk management needs the skills and understanding to assess whether the output generated by the models and particularly how it is used in the firm, is beneficial or could generate harmful situations for the business. This discussion should be constructive, recognising that LLMs offer a significant increase in efficiency, provided there is solid independent oversight of their use. A further area of investigation should look at the use of LLMs in compliance and regulatory reporting. Incorrect reporting in these areas can raise questions about the governance of the firm. In treasury and risk, particular attention should be paid to reporting on liquidity, interest rate risk, market risk, and capital. These are all areas where LLMs can significantly improve efficiency but require full oversight by those in charge. Scope of audit Moreover, the audit function has a responsibility to delve into these models to ensure they do not lead to unintended consequences. For example, do these models reinforce bias in the business? Are we overlooking pipeline risk? Are we risk-seeking with deposits without considering the consequences? Is there a tendency to think that interest rates are always moving in one direction? Internal audit should also examine whether balance sheet and risk data are leaking out of the business via the LLM into the public domain. A themed audit could drill down into the data produced and its use in the firm or reach upwards towards the management suite to assess governance and oversight. I’m also aware that GPT-4 can subtly alter content, giving it a slightly different meaning from what was intended. This too is something to be wary of, and I advise thoroughly checking any output before it is used in the firm. What does this mean for the future? An opportunity to be grasped It means that LLMs currently offer an opportunity to increase the efficiency of those working in treasury and risk, but output from LLMs used in the business, either from a reporting or governance perspective, must be sense-checked to ensure its robustness. As LLMs improve, we will likely grow more reliant on their accuracy and efficacy, similar to the first risk management systems used in treasury 40 years ago. The rapid development of LLMs suggests that within a few years, they will reach "escape velocity," capable of completing tasks at a human level with minimal prompting, significantly reducing the need for human involvement in the process between trades and the C-suite.

In this sixth instalment, we delve into two time-saving applications of LLMs: summarising and repurposing PowerPoint presentations, and efficiently taking minutes in meetings. Whilst the following is general it takes little imagination to use this in the context of Treasury and Risk thereby saving yourself a lot of time and effort. Summarising and Repurposing PowerPoint Presentations My discovery of this application was serendipitous, revealing the untapped potential of GPT-4 when we experiment and explore. Initially, interacting with ChatGPT to discuss an uploaded PowerPoint file required code interpreter and a series of specific prompts. Now, the process has become remarkably user-friendly. Simply upload your PowerPoint presentation into GPT-4, hit the return and it automatically generates a concise summary - no prompting necessary Real-World Example: Imagine having a series of slides for a presentation on large language models. By uploading these into GPT-4, you receive a bullet-point summary almost instantaneously. However, this is just the beginning. Each bullet point can be expanded into a detailed article, providing rich content tailored to your audience. At the end of this article is an example of both summary and content repurposing in action. Taking Minutes in Meetings In today’s digital age, recording meetings, especially committee ones, is commonplace. Whilst I cannot get GPT-4 to transcribe audio files directly, alternative solutions like Descript offer transcription and summarisation services. For instance, uploading an m4a recording to Descript yields a comprehensive and accurate transcript - it's impressive. Zoom meetings also offer another efficient alternative, providing near-instantaneous transcripts. The advantage of these AI-driven summaries lies in their accuracy and objectivity, free from the biases or errors a human note-taker might introduce. Workplace Efficiency As we continue to explore the capabilities of LLMs like GPT-4, their potential becomes increasingly evident. From transforming presentations into diverse content forms to accurately capturing the essence of meetings, these tools are not just about saving time; they are about enhancing the quality of our work and getting information to those in our organisations in a format that is useable and actionable. I also believe they (LLMs) deal with a lot of boring and repetitive tasks that take the fun out of work for those involved and the more you experiment the more you can see their potential. Example of GPT-4 and slide deck, this is the uploaded presentation (note I uploaded a ppt file). Then hit return. GPT-4: The PowerPoint presentation titled "My Experience with LLMs" by William Webster covers various aspects of using Large Language Models (LLMs), including their applications and capabilities. Key points from the presentation include: Introduction to Large Language Models (LLMs): This section explains the basics of Generative Pre-trained Transformers (GPT) and mentions different models from various companies like OpenAI's ChatGPT, Anthropic’s Claude, Google’s Bard, Meta’s LLaMA, and Inflection’s Pi. User Interface and Interaction: It discusses how users interact with GPTs through various interfaces such as text, documents, pictures, and speech. Emphasis is placed on the ease of interaction through Natural Language Processing (NLP), requiring no programming skills. Practical Applications: The presentation highlights several practical uses of LLMs, such as reading and converting text from pictures, calculating costs, generating images with DALL-E, and analyzing CT scans. Evolution of GPT Models: The presentation tracks the development of GPT models, mentioning the release of GPT-3.5 and GPT-4, and their capabilities in creating content and engaging in detailed conversations. Usage Examples in Professional Contexts: Specific use cases in professional settings are outlined, including drafting emails and documents, summarizing information, and crafting prompts for specific outputs. Efficiency and Risk Management: The presentation explores how LLMs can improve efficiency in tasks like preparing board papers, taking minutes, and analyzing financial risks. AI as a Creative Engine: It is emphasized that unlike traditional search engines, LLMs can generate diverse outputs, making them valuable for creative and strategic tasks. AI Policy and Workforce Implications: The presentation underscores the importance of developing an AI policy, training needs, and the impact of AI on jobs and workflows. Ethical Considerations and Data Security: Concerns about data training, privacy, and the need for human oversight in AI processes are mentioned. Resources: The presentation concludes with a list of resources for further exploration, including podcasts, blogs, and websites related to AI and LLMs. The presentation provides a comprehensive overview of the capabilities, applications, and considerations surrounding the use of Large Language Models in various contexts. Me: Write a single helpful article on: User Interface and Accessibility: Focuses on the user-friendly nature of LLMs, allowing interaction through text, images, and voice without needing programming expertise. GPT-4: Harnessing the Power of AI: The User-Friendly Revolution of Large Language Models In the world of artificial intelligence, the emergence of Large Language Models (LLMs) has been nothing short of a revolution. A key factor behind their widespread appeal lies in their exceptionally user-friendly interfaces, making advanced AI accessible to everyone, regardless of their programming skills. This article delves into how LLMs have transformed user interaction through text, images, and voice, effectively democratizing access to cutting-edge technology. The Simplicity of Text Interaction Text-based interaction is the bedrock of LLM interfaces. It's as simple as typing in a query or command in natural language - much like sending a text message. This approach has opened up AI to a much broader audience, allowing users to engage in everything from casual conversations to complex problem-solving without any knowledge of programming languages. Tools like OpenAI's ChatGPT exemplify this, where users can obtain information, write code, or even create content through straightforward text prompts. Visual Interfaces: A Picture is Worth a Thousand Words The integration of visual interfaces in LLMs has taken user experience to a new dimension. Users can now interact with AI through images, making these systems much more intuitive and versatile. For instance, OpenAI's DALL-E can generate detailed images from textual descriptions, showcasing how LLMs can understand and execute visual tasks. This capability is not just a boon for creatives but also enhances the utility of AI in fields like design, education, and marketing. Voice Interaction: Speaking to AI Voice interaction represents a significant leap in making LLMs more accessible and natural to use. By simply speaking to these models, users can engage in a range of activities, from getting real-time assistance to controlling smart devices, all in a hands-free manner. This feature is particularly beneficial for those who find typing challenging or prefer verbal communication. It also paves the way for LLMs to be used in diverse environments, from busy kitchens to moving vehicles. Breaking Down Barriers: AI for All Perhaps the most profound impact of these user-friendly interfaces is the democratization of AI. With LLMs, the need for technical expertise in coding or data science is no longer a prerequisite to harnessing the power of AI. This opens up a world of possibilities for non-technical users, including educators, artists, small business owners, and hobbyists, allowing them to leverage AI for their unique needs. The Future of User Interaction with LLMs Looking ahead, the user interfaces of LLMs are poised to become even more intuitive and integrated. We can expect developments that combine text, visuals, and voice in more seamless and interactive ways, potentially incorporating additional sensory inputs for a more immersive experience. The goal is to create interfaces that are as natural and effortless to use as our own human senses, making AI an unobtrusive yet powerful extension of our daily activities. Conclusion The user-friendly nature of LLMs marks a significant milestone in the field of AI, bringing these powerful tools within reach of a global audience. By simplifying interaction through text, images, and voice, LLMs are not just enhancing productivity and creativity; they are opening doors to a future where the benefits of AI can be enjoyed by everyone, irrespective of their technical background. As we continue to refine these interfaces, the potential of LLMs to enrich and simplify our lives seems limitless.

The fifth instalment of our series exploring simple GPT-4 use cases focuses on its capacity to generate documentation for an ALCO committee. This article delves into a specific scenario involving a small bank or building society, and how GPT-4 can assist in creating insights into interest rate risk. Our approach begins with a basic spreadsheet showing different time buckets, the corresponding assets and liabilities, the net gap position, and two flat yield curves at 5% and 5.01%. Using this data, the discount factors from the yield curves are used to calculate the net profit or loss resulting from a 0.01% interest rate shift, namely this is a simple delta report, a common tool in banking for managing interest rate risk. In this example, we see a bank with long-term assets funded by short-term liabilities, creating both interest rate exposure and potential liquidity risk. Initially, I asked GPT-4 to analyse the spreadsheet. However, the results were unsatisfactory. GPT-4 struggled with the Excel format, producing an incorrect and misleading narrative. Adopting a different strategy, I copied rather than uploaded the relevant section of the spreadsheet directly into GPT-4's prompt window. This action generated an image of the spreadsheet and displayed the numerical data in the chat. After removing the numbers from the chat and leaving only the image, I requested GPT-4 to analyse it. Despite this modification, the process remained challenging due to inconsistencies in the spreadsheet's formatting, such as unclear labelling and the negative numbers being in brackets. After improving the spreadsheet's consistency and including negative numbers rather than brackets, I initiated a new session with GPT-4. This time, the result was more coherent. Progressing further, I requested GPT-4 to create bar charts and relevant commentary for the ALCO committee. Although time-consuming, this exercise yielded a workable method for engaging GPT-4 effectively. From this experience, you can see that GPT-4 can act as a co-pilot but it still needs a lot of interaction so it is not an automatic process. However, if you use it in conjunction with its text generation capability, preparing relevant and useful information is achievable faster. It would not surprise me if, within 12 months, given the improvements in LLMs the process could be automated leading to a very large jump in productivity. Key Points Consistency of data uploaded to the LLM: GPT-4 does not deal well with any slight inconsistency in the data format, for example, if there are blank cells, unaligned rows or columns or bracketed numbers. This contrasts with using text in GPT-4 where it can work with, and make sense of, relatively jumbled ideas, poor syntax and grammar to come up with very good working documents. When LLMs can do this with numbers we will have a step change in the ability to use them in finance. Handling Spreadsheets with GPT-4: GPT-4 exhibits better performance with images of spreadsheets rather than the files themselves, why this is, I don’t know. Is it due to it being a language-based model? Uploading the image without a prompt: this led GPT-4 to analyse what it received before I initiated the prompting, it seemed to slow the model down giving it time to think about the information rather than just trying to complete the prompt and in this limited example the subsequent output was better as a result. Process Efficiency: The procedure of using GPT-4 for such tasks can be initially tedious. However, mastering the process is a valuable skill, promising time savings for future repetitions. The Necessity of Human Oversight: Similar to the early days of spreadsheet usage, there's a risk of assuming AI-generated content is infallible. Vigilance is essential to identify potential errors. GPT-4 can generate plausible but potentially incorrect charts and commentary. Therefore, human validation is crucial to ensure accuracy before using the output in reports. Evolution of LLMs: GPT-4's current capabilities significantly surpass those of its predecessors, suggesting a near future where AI can seamlessly turn data into meaningful analyses and reports. This is where you just upload the data and the charts and written reports are generated automatically for the whole ALCO deck. Benefits of LLMs in Routine Tasks: By handling routine analytical work, AI models like GPT-4 free up human resources to focus on complex decision-making and risk evaluation through collaborative discussions. The dialogue with GPT-4, detailed below, provides insight into the challenges and solutions of integrating LLMs into financial documentation and analysis tasks. ---------------------------------------------------- Me: GPT-4: You've uploaded an image that appears to contain a table with financial data related to asset and liability management, including time buckets, asset and liability amounts, net gap, interest rates (Curve 1 and Curve 2), and net profit and loss (P&L). This data seems to be part of an asset-liability matching or a gap analysis, which is typically used by financial institutions to manage interest rate risk. The table includes different time buckets ranging from "overnight" to "10 years". In each time bucket, the assets and liabilities are compared, and the 'Net gap' row shows the difference between them. Below that, the table shows two different interest rate curves and the net P&L calculated from these rates, which indicates the financial impact of the interest rate changes on the net gap over time. If you need any specific analysis or have questions about the data, feel free to ask! Me: please produce a bar chart of the net gap by time bucket GPT-4: Me: now a bar chart of the P&L by time bucket GPT-4: Me: Now the commentary for an ALCO meeting, this should explain the interest rate risk GPT-4: ALCO Meeting Commentary on Interest Rate Risk Introduction: Our current asset-liability position indicates varying degrees of interest rate risk across different time buckets. The net gap analysis, alongside the interest rate curves, offers a clear view of our exposure. Net Gap Analysis: The net gap distribution shows significant short-term liabilities exceeding assets in the "overnight" to "6-12 months" buckets, indicating a potential liquidity risk if interest rates rise. Conversely, from "1 year" to "10 years," we observe a surplus of assets over liabilities. This longer-term positioning could be beneficial if interest rates decrease but poses a risk if rates increase, as our interest income would be adversely affected while paying out at higher rates on liabilities. Interest Rate Curves and P&L Impact: The first curve (Curve 1) represents a baseline scenario with current interest rates, while the second curve (Curve 2) represents a slight upward shift in rates. The net P&L across time buckets shows that as the maturity extends, the impact of an interest rate rise becomes more pronounced, culminating in significant potential losses, particularly evident in the "2 year" to "5 year" buckets. Short-Term Concerns: In the immediate term, we must address the liquidity risk due to the negative gap. Any unexpected increase in short-term rates could increase our cost of funding significantly, which is not fully offset by corresponding increases in asset yields. Long-Term Strategies: For the longer-term, our current position could benefit from a hedging strategy to mitigate the risk of rising rates. This might include entering into interest rate swaps or options to lock in current rates for future liabilities or to convert fixed-rate assets to floating rates. Conclusion: It is imperative to re-evaluate our asset-liability matching strategy to ensure it aligns with our risk appetite and market expectations. While we have some cushions due to the positive gaps in longer maturities, the potential for interest rate hikes suggests a more conservative approach might be warranted to protect our net interest margin. Recommendations: Increase the proportion of floating-rate or short-duration assets to reduce the negative gap in the short term. Consider derivative instruments for hedging long-term interest rate exposure. Continuously monitor the interest rate environment and adjust our ALM strategy accordingly. This commentary would need to be tailored with more specific details from the financial institution’s own data and strategic positioning, but it provides a framework for discussing interest rate risk at an ALCO meeting.

In this fourth instalment of our series, we explore an intriguing application of large language models (LLMs) in treasury and risk management: envisioning future events and shaping our responses to them. This application is particularly compelling, considering that while we often concentrate on LLMs for text, visual, and data interpretation, I believe one of their true strength lies in sculpting business scenarios and strategies. High-level scenario and strategy planning is a critical task for knowledge workers. Here, LLMs like GPT-4 shine, not by replacing human decision-making but by offering a wide range of potential outcomes and strategies, enriching our planning processes. I put this to the test with GPT-4, prompting it to generate plausible yet severe scenarios for liquidity stress testing. The results were multifaceted. Initial prompts led to varied scenarios, including retail customer behaviour. Further prompting unearthed even more unconventional scenarios. GPT-4 didn't just list these scenarios; it grouped them into categories and this helped in deciding their use and applicability. The real advantage of working with GPT-4 is evident here. The output can be used to identify gaps in your liquidity risk analysis or compare it with existing industry knowledge. You can select any GPT-4 suggestion to spawn further discussions, evaluating their relevance and potential integration into your plans, including necessary updates to your mitigation strategies. For instance, I delved into a scenario centred on social media-driven customer behaviour – not something I’m particularly familiar with. GPT-4 helped me understand how this could unfold, its potential impact on our operations, and possible mitigating measures. Key Takeaways from This Use Case Scenario Development: LLMs like GPT-4 are formidable tools for crafting scenarios and strategies. I believe their role in business strategy will grow. Creative Assistance: GPT-4 rapidly generates scenarios, some amusing and far-fetched. It acts as a creative aide, but human discernment remains crucial in selecting the most pertinent scenarios. Effective Prompting: Tailoring prompts to your organisation's specific context enhances GPT-4's effectiveness. The more context and clarity provided, the more relevant the outcomes. Time Efficiency: GPT-4's swift responses save considerable time, allowing teams to focus more on discussion and evaluation rather than scenario generation. This underscores the value of human knowledge and experience when using LLMs. Strategic Application: Choosing relevant examples demonstrates the potency of this approach. GPT-4's suggestions on mitigating social behaviour risks can be integrated into your business strategy and tailored to your specific needs. This particular example may not be for you, but this analysis using GPT-4, facilitates change in what you do. ----------------------------------------- My conversation with GPT-4 Below are selected excerpts from my dialogue with GPT-4 on this topic. It's important to note that achieving these results required multiple prompts and some strategic thought in engaging with GPT-4. This is a common aspect of leveraging such a powerful tool effectively. Me: Imagine I'm a risk manager working at a building society or bank. I want to explore and add some stress scenarios to our liquidity policy. We are predominately retail funded and our assets are mortgages, gilts and cash deposits at the Bank of England. We use swaps to hedge interest rate risk arising from fixed rate mortgage assets. The stresses should be severe but plausible. I am particularly interested in anything new or unusual. Use all your knowledge and expertise to give me some examples. GPT-4: As a risk manager in a building society or bank with a portfolio primarily consisting of mortgages, gilts, and cash deposits at the Bank of England, along with the use of swaps for hedging interest rate risks from fixed-rate mortgages, your focus on developing severe but plausible stress scenarios for your liquidity policy is a prudent approach. Let's explore some innovative and unusual scenarios that could challenge your institution's liquidity position: Pandemic-Induced Economic Downturn: Consider a scenario where a global health crisis (similar to COVID-19) leads to a significant economic downturn. This could result in a sharp increase in unemployment, leading to a spike in mortgage defaults. Additionally, there might be a sudden withdrawal of retail deposits as consumers seek to cover living expenses, coupled with a drop in property values that affects the collateral value of your mortgage assets. Interest Rate Shock Combined with Inflation Spike: Imagine an unexpected, rapid increase in interest rates, possibly due to a sudden spike in inflation. This scenario would affect both sides of your balance sheet. On the asset side, the value of your fixed-rate mortgage assets and gilts could decline, while on the liability side, there could be pressure to increase interest rates on deposits to retain customers, squeezing your net interest margin. Technology-Driven Market Disruption: Envision a scenario where a technological breakthrough (such as a new digital currency or a disruptive financial technology platform) significantly alters consumer behavior. This could lead to a rapid outflow of retail deposits to these new platforms, creating liquidity strain. Additionally, this tech disruption could impact the real estate market, affecting the value of your mortgage assets. Regulatory Change Impacting Swaps Market: A sudden regulatory change or a political event that significantly alters the swaps market could be a stress scenario. For example, a change in regulation that increases the cost of using swaps or restricts their use could expose your institution to unanticipated interest rate risk, affecting your liquidity position. Climate Change-Induced Natural Disasters: Consider a scenario where severe climate events (such as floods, wildfires, or storms) significantly impact large geographical areas. This could lead to widespread property damage, affecting the value of mortgage collateral and potentially leading to increased defaults. Additionally, these events could disrupt economic activity, impacting retail deposits. Cybersecurity Incident Leading to Systemic Failure: A systemic cybersecurity incident that affects major financial infrastructure (like payment systems) could result in a temporary but severe disruption to liquidity. This could be due to the inability to access or move funds, combined with a loss of depositor confidence. Geopolitical Crisis Causing Market Turmoil: A geopolitical event, such as a major conflict or political instability in a key economy, could lead to significant market volatility. This could impact your gilt portfolio and potentially lead to a flight-to-quality, with depositors moving funds to perceived safer havens. In each of these scenarios, it's crucial to assess not only the direct impact on your assets and liabilities but also the second-order effects, such as the impact on borrower behavior, market confidence, and operational capacity. Incorporating such scenarios into your stress testing exercises will enhance your institution's preparedness for a range of severe but plausible events, ensuring robust liquidity risk management. Me: What about risks from the behaviour of retail customers or competitors? GPT-4: Considering risks arising from the behavior of retail customers and competitors is indeed crucial for a comprehensive stress testing framework, especially for a financial institution like yours that relies heavily on retail funding and operates in a competitive market. Here are some scenarios that encompass these aspects: Mass Retail Customer Withdrawals: Imagine a scenario where a widespread rumor or negative news (true or false) about the financial health of your institution leads to a loss of confidence among retail customers. This could trigger a 'bank run' scenario, where a large number of customers simultaneously decide to withdraw their deposits, significantly straining your liquidity position. Shift in Customer Preferences: Consider a scenario where there's a sudden shift in customer preferences towards newer financial products or services offered by fintech companies or digital banks. This could result in a steady outflow of retail deposits as customers move their funds to these new offerings, impacting your funding base. Competitive Rate War: Envision a scenario where one or more competitors aggressively cut rates on mortgages or increase rates on deposits to capture market share. This could put pressure on your institution to follow suit to retain customers, impacting your net interest margin and potentially leading to liquidity strains if deposit rates are increased significantly. Competitor's Technological Advancement: A competitor’s breakthrough in technology, such as advanced online banking platforms or AI-driven financial advisory services, could attract a significant portion of your customer base. This shift could be particularly impactful if it targets a key demographic, like tech-savvy younger customers, leading to a gradual erosion of your deposit base. Competitor's Strategic Alliance or Merger: A major competitor could form a strategic alliance with another financial player or undergo a merger, significantly enhancing their market presence and product offerings. This could lead to a loss of competitive edge for your institution, potentially resulting in customer attrition and a reduction in new business. Competitive Disruption in Mortgage Market: If a competitor introduces an innovative mortgage product – for example, significantly lower rates or more flexible terms – it could lead to a significant shift in the mortgage market. Your institution might face a double-edged sword: losing existing customers refinancing with competitors and facing challenges in attracting new mortgage customers. Social Media-Driven Customer Behavior: In today’s digital age, social media can rapidly influence public opinion and customer behavior. A viral social media campaign (positive for a competitor or negative against your institution) could swiftly alter customer perceptions and behavior, leading to unexpected shifts in your customer base and deposit levels. Incorporating these customer and competitor behavior-driven scenarios into your liquidity stress testing will provide a more robust understanding of potential risks and enable your institution to develop more effective strategies to mitigate these risks. It's also important to continuously monitor market trends, customer behaviors, and competitor actions to update these scenarios and ensure they remain relevant and plausible. Me: Now add a few unusual scenarios we could consider GPT-4: Adding unusual but plausible scenarios to your stress testing can help prepare your institution for a broader range of potential risks. These scenarios, while less conventional, could provide valuable insights into your institution's resilience. Let's consider some creative and unconventional scenarios: Digital Currency Disruption: Imagine a scenario where a major technology company or a consortium launches a new digital currency that rapidly gains popularity. This currency offers higher interest rates for savers and lower transaction costs, leading to a significant portion of your retail customer base transferring their deposits to this new digital currency, causing a sudden liquidity crunch. Social Movement Leading to Financial Behavior Shift: Envision a powerful social movement advocating for ethical and environmentally sustainable banking practices gaining momentum. This movement could influence a significant portion of your customer base, leading them to transfer their deposits to institutions perceived as more socially responsible, impacting your liquidity. Virtual Real Estate Boom in Metaverse Environments: With the growing interest in virtual worlds and metaverses, imagine a scenario where virtual real estate becomes a lucrative investment, diverting significant amounts of disposable income and savings from traditional assets like physical real estate to virtual properties. This could impact the mortgage market and lead to unexpected shifts in your customer's financial behavior. Global Supply Chain Crisis: Consider a scenario where a severe, prolonged global supply chain crisis significantly impacts the cost of living and doing business worldwide. This could lead to increased defaults on mortgages due to higher living costs and affect the employment sector, thereby impacting your retail customer's ability to maintain their deposits and financial commitments. Radical Regulatory Changes: A scenario where there's a sudden and radical overhaul of financial regulations, possibly driven by a global consensus on combating financial crimes or a major economic event. These changes could include stringent capital requirements or limitations on certain types of financial products, dramatically impacting your institution's operations and customer behavior. AI-Driven Financial Planning Trend: Artificial intelligence (AI) in financial planning becomes extremely sophisticated, leading to a trend where AI advisors recommend customers to diversify their assets away from traditional savings accounts and into a variety of new financial products. This could lead to a large-scale withdrawal of retail deposits as customers follow these AI-generated advice. Extraterrestrial Event Impacting Global Markets: While highly unlikely, a scenario involving a significant extraterrestrial event, such as confirmed evidence of extraterrestrial life, could lead to unprecedented market volatility. The uncertainty and media frenzy surrounding such an event could lead to erratic customer behavior, including panic withdrawals or hoarding of funds. While some of these scenarios might seem far-fetched, they serve to challenge conventional thinking and help your institution prepare for a wide range of possibilities. The key is to evaluate the potential impact of these unusual scenarios on your liquidity, customer behavior, and overall market stability, and to develop strategies to mitigate these risks. Me: Please create a list of all the stresses you have come up with GPT-4: Absolutely. Here's a comprehensive list of all the stress scenarios we've discussed, encompassing both conventional and unusual risks: Conventional Stresses Pandemic-Induced Economic Downturn: A global health crisis leading to economic downturn, mortgage defaults, and retail deposit withdrawals. Interest Rate Shock and Inflation Spike: Unexpected rise in interest rates due to inflation, affecting mortgage assets, gilts, and deposit rates. Technology-Driven Market Disruption: New digital currency or fintech platforms significantly altering consumer behavior and impacting retail deposits. Regulatory Change in Swaps Market: Sudden regulatory changes affecting the cost and usability of swaps, exposing interest rate risks. Climate Change-Induced Natural Disasters: Severe weather events damaging property and affecting the value of mortgage collateral. Cybersecurity Incident: A major cybersecurity breach affecting financial infrastructure, leading to liquidity challenges. Geopolitical Crisis: A major geopolitical event causing market volatility and affecting gilt portfolios and depositor confidence. Behavior and Competitive Stresses Mass Retail Customer Withdrawals: A 'bank run' scenario triggered by loss of confidence among retail customers. Shift in Customer Preferences: Customers preferring new financial products or services, leading to retail deposit outflows. Competitive Rate War: Competitors significantly altering rates on mortgages or deposits, impacting net interest margin. Competitor's Technological Advancement: Loss of customers due to a competitor's technological breakthrough in banking services. Competitor's Strategic Alliance or Merger: A major market shift due to a competitor's merger or alliance, leading to customer attrition. Competitive Disruption in Mortgage Market: A competitor introducing an innovative mortgage product, affecting customer retention and acquisition. Social Media-Driven Customer Behavior: Rapid shifts in customer behavior influenced by viral social media campaigns. Unusual Stresses Digital Currency Disruption: Launch of a popular digital currency leading to significant retail deposit withdrawals. Social Movement for Ethical Banking: Customers moving their deposits to more socially responsible institutions due to a social movement. Virtual Real Estate Investment in Metaverse: Shift of investment from physical to virtual real estate, affecting the mortgage market. Global Supply Chain Crisis: A supply chain crisis raising living costs, impacting employment and mortgage defaults. Radical Regulatory Changes: Overhaul of financial regulations impacting banking operations and customer behavior. AI-Driven Financial Planning Trend: AI advisors leading customers to diversify away from traditional savings, affecting deposits. Extraterrestrial Event Impacting Markets: Market volatility and erratic customer behavior triggered by an extraterrestrial event. Each of these scenarios poses unique challenges and opportunities for testing the resilience and adaptability of your institution's liquidity and risk management strategies. Me: Please provide a more detailed analysis of how Social Media-Driven Customer Behavior could affect us. In particular consider what could cause it, how it could unfold, the effect it could have on us and mitigating steps we could take now, or in the future should such an event happen. GPT-4: Analysis of Social Media-Driven Customer Behavior Impact Potential Causes Viral Misinformation or Negative Campaigns: False rumors or targeted negative campaigns about your institution's financial health, unethical practices, or customer service failures could go viral. Influencer Impact: Influential social media personalities could sway public opinion against your institution, either due to personal experiences, affiliations with competitors, or as part of a social cause. Customer Service Incidents: A small-scale customer service issue, if poorly handled and publicized online, could escalate rapidly in the social media ecosystem. Competitor's Positive Campaigns: Competitors might leverage social media effectively to highlight their strengths or innovations, indirectly casting a shadow on your offerings. How It Could Unfold Rapid Spread of Information: Social media accelerates information dissemination, with the potential to reach millions in a short span. Echo Chambers: Information, whether true or false, tends to get amplified within social media echo chambers, reinforcing negative perceptions. Mobilization of Public Sentiment: Social media can mobilize public sentiment quickly, leading to coordinated actions like account closures or protests. Media Amplification: Traditional media might pick up on these social media trends, further amplifying the issue. Potential Effects Run on Deposits: Widespread panic could lead to a significant number of customers withdrawing their deposits, straining liquidity. Damage to Brand and Reputation: Long-term damage to the institution's brand and reputation could deter potential customers and affect business growth. Reduced Investor Confidence: Shareholder and investor confidence could waver, affecting stock prices and future investment. Operational Disruption: Increased volume of inquiries and complaints could overwhelm customer service channels, leading to operational disruptions. Mitigation Strategies Current Measures Robust Social Media Monitoring: Implement comprehensive tools to monitor social media for mentions of your institution. Early detection is key. Crisis Communication Plan: Develop a crisis communication plan specifically for social media-driven events. Customer Service Excellence: Invest in exceptional customer service to prevent issues from escalating to social media complaints. Positive Brand Building: Regularly engage on social media with positive stories, customer testimonials, and transparency reports to build a reservoir of goodwill. Responsive Actions During an Event Rapid Response Team: Have a dedicated team ready to respond quickly to misinformation or negative publicity on social media. Direct Customer Engagement: Engage directly with affected customers to resolve issues and demonstrate commitment to customer satisfaction. Collaboration with Influencers: Collaborate with respected influencers to counteract negative campaigns with factual information. Transparent Communication: Regularly update customers and the public with clear, transparent information about the situation. Long-Term Strategic Measures Digital Literacy Campaigns: Educate customers about misinformation and digital literacy to build a more discerning customer base. Strengthening Online Presence: Develop a strong, positive online presence through regular, engaging content and community involvement. Liquidity Cushions: Maintain robust liquidity reserves to handle potential runs on the bank. Regular Simulations: Conduct regular simulations or drills for handling social media crises to ensure preparedness. By proactively managing your social media presence and being prepared to respond effectively to any crises that arise from it, your institution can mitigate the risks associated with social media-driven customer behavior. This approach not only helps in managing immediate crises but also contributes to building long-term resilience and trust with your customer base.

The integration of advanced technology into everyday practices is not just a luxury, it's a necessity. This article, the third in our series, sheds light on the simple yet powerful applications of Large Language Models (LLMs) in treasury and risk management. In a sector where time is as valuable as the information handled, LLMs like GPT-4 are revolutionising how we manage and process information. The Challenge of Information Overload Professionals in treasury and risk management are constantly in need of up-to-date information on regulatory policies, economic commentary, and industry opinions. While the internet is a treasure trove of such data, sifting through it to find relevant information is a daunting and time-consuming task. The hours spent in this endeavour are often at the expense of more critical, strategic work. The Solution: LLMs in Action The capabilities of LLMs extend beyond mere information retrieval. To demonstrate this, I employed GPT-4 to research the Bank of England's publications on using large language models and AI in financial services. The model efficiently located the latest feedback statement, FS2/23, and provided a concise summary. Further probing into the summary revealed in-depth insights, especially regarding "live" regulatory guidance. Three noteworthy points: Rapid Information Retrieval: GPT-4 can swiftly navigate the web to find the exact information you need. Summarisation Skills: It can condense extensive data into digestible summaries. Deep Dive Capability: You can delve deeper into these summaries to uncover more nuanced details as required. Practical Workplace Application The potential of GPT-4 in treasury and risk management is immense. It lends itself exceptionally well to: Developing high-quality content for management reports, committee briefings, policies, and procedures. Providing updates on economic indicators and Monetary Policy Committee (MPC) meetings, thereby explaining the state of the economy in layman's terms. Assisting senior management with relevant and timely information. Professionals in this field, often pressed for time, will find LLMs invaluable assistants, capable of undertaking various tasks to expedite their workflow. Human-AI Collaboration It's crucial to note that, as of now, LLMs like GPT-4 cannot entirely replace human intervention. A collaborative approach, where humans guide and refine the output of these models, is necessary. This synergy allows for the completion of individual tasks that cumulatively result in comprehensive and sophisticated outputs, such as policy documents on risk management or committee briefing papers. Looking Ahead The dialogue below shows the conversation I had and as LLMs continue to evolve, we can anticipate a future where they handle more foundational writing and preparatory tasks, thus granting professionals more time for in-depth discussions and considerations of the topics at hand. This shift will undoubtedly enhance our approach to risk management and decision-making. Conclusion In conclusion, LLMs like GPT-4 are not just tools; they are partners in the quest for efficient and effective treasury and risk management. They enable professionals to focus on what truly matters – using their expertise to make informed decisions that shape the financial landscape thereby relieving us of the lower-level repetitive tasks that currently take up so much time. Here's the conversation: Me: “Please provide the most recent article from the Bank of England this year concerning the use of LLMs and AI in financial services”. GPT-4: “The most recent article from the Bank of England this year concerning the use of Large Language Models (LLMs) and AI in financial services is titled "FS2/23 – Artificial Intelligence and Machine Learning," published in October 2023​​. This document is a feedback statement summarizing responses to a discussion paper (DP5/22) published in October 2022, aimed at understanding the implications of AI on the prudential and conduct supervision of financial firms​​. The feedback statement indicates that the discussion paper received 54 responses from various stakeholders, including banks and industry bodies. Key points raised by the respondents include scepticism about the usefulness of a regulatory definition of AI, suggestions for 'live' regulatory guidance that adapts to rapidly changing AI capabilities, the need for ongoing industry engagement, and calls for more coordinated regulation concerning AI, particularly regarding data regulation, consumer outcomes, third-party models and data, and the complexity of AI systems. There's an emphasis on principles-based approaches, consumer protection, managing risks of bias and discrimination, and the challenges of integrating AI in financial services, including cybersecurity risks and the potential for AI misuse​​​​​​.” Me: “please give me a summary of this document” GPT-4: “The Bank of England's feedback statement "FS2/23 – Artificial Intelligence and Machine Learning" summarizes responses to the 2022 discussion paper on AI in UK financial services. Key points include: Skepticism about a regulatory definition of AI, with a preference for principles-based approaches. The need for 'live' regulatory guidance due to rapidly evolving AI capabilities. Calls for ongoing industry engagement and better regulatory coordination, especially in data regulation. Emphasis on consumer protection, addressing risks of bias and discrimination. Concerns about integrating AI in financial services, focusing on cybersecurity risks and potential misuse of AI. The document reflects diverse industry perspectives on adopting AI in financial services, focusing on regulation, risk management, and the evolving nature of technology.” Me: “please expand on 'live' regulatory guidance” GPT4: "Live regulatory guidance, as mentioned in the Bank of England's feedback statement, refers to a dynamic and adaptable approach to regulation that keeps pace with the rapidly evolving capabilities of AI technologies. Instead of static rules, this approach involves regularly updating guidance and best practices to reflect the latest developments in AI. This method allows regulators to respond promptly to technological advancements and emerging risks, ensuring that regulatory frameworks remain relevant and effective in a fast-changing landscape. This approach is particularly pertinent for AI, where the speed of innovation can quickly outpace traditional regulatory mechanisms.”

Welcome to the second instalment in our insightful series focused on leveraging LLMs (Large Language Models) in the fields of treasury and risk management. Our focus today extends to the LLM's ability to rapidly produce accurate, audience-tailored text. Often, professionals in our field are tasked with distilling complex information into digestible formats for varied audiences. Rewriting technical documents for broader comprehension can be a daunting and time-consuming task, particularly when trying to strike the right balance in tone and depth. A common challenge arises when conveying complex subjects to non-executive directors or other stakeholders who may not possess deep technical knowledge. This disconnect can result in communications that are either misunderstood or overlooked. This is precisely where tools like GPT-4 become invaluable. They enable us to transform technical narratives, such as those intended for an asset liability committee, into versions more suitable for different audiences, like board members. Consider this scenario: you're familiar with the 'pipeline risk' concept associated with hedging fixed-rate mortgage products using interest-rate swaps. The challenge is deciding whether to pre-hedge the mortgage pipeline or wait for actual drawdowns. This complexity needs to be communicated effectively, whether it's to your team or for crafting your interest rate risk management policy. Here's a detailed explanation we might use internally: Subject: Understanding and Managing Pipeline Risk in Fixed Rate Mortgages Dear Team, I want to discuss a challenging aspect of our risk management practices: hedging pipeline risk in fixed rate mortgages using interest rate swaps. As you know, this is a nuanced area, and getting our strategy right is crucial for maintaining our financial health and reputation. 1. The Nature of Pipeline Risk: Pipeline risk emerges from the lag between when a mortgage rate is locked in for the customer and when the loan is actually funded. During this window, interest rates can fluctuate, affecting the cost of funds for us. This timing difference is a significant risk element, particularly in volatile markets. 2. The Drawdown Dilemma: One of our core challenges is predicting when drawdowns will actually occur. Mortgages don’t always close on the expected date; delays can happen for various reasons. This uncertainty makes it difficult to precisely match our hedges with actual funding needs, leading to either overhedging or underhedging. 3. The Impact of Interest Rate Changes: Interest rate volatility adds another layer of complexity. When rates rise, customers are more likely to go through with their locked-in rates, as they are comparatively more attractive. Conversely, if rates fall, customers might opt out, seeking better rates elsewhere. This customer behaviour can leave us with an imbalanced hedge. 4. The Overhedging Risk: If we hedge based on the total volume of locked-in rates, we might end up overhedging. This happens because not all locked-in loans will close, especially if rates move favourably for customers. Overhedging leads to unnecessary costs and can impact our profitability. 5. The Underhedging Risk: Conversely, choosing not to hedge or hedging too little leaves us exposed to market fluctuations. This underhedging can be equally detrimental, especially in a rapidly rising rate environment, where the cost of funds can increase significantly. 6. Writing Free Options for Customers: Essentially, by offering fixed-rate mortgages with rate lock-ins, we are implicitly giving customers a free option. They can choose to proceed or withdraw based on how the market evolves, putting us in a position where we must skillfully manage these implicit options. Our Approach: To manage these risks effectively, we need to adopt a dynamic hedging strategy. This involves continuously adjusting our hedge positions based on the latest information about loan closings and market conditions. Regular communication with the sales and processing teams is crucial to get timely updates on the mortgage pipeline. Additionally, stress testing under different interest rate scenarios will help us understand potential exposures and refine our hedging strategies accordingly. Conclusion: Navigating the intricacies of hedging pipeline risk requires a balance between market understanding, predicting customer behaviour, and being agile in our hedging approach. It's a complex task, but with our combined expertise and a strategic approach, we can manage these risks effectively. Looking forward to discussing this further in our upcoming meeting. Best regards, This communication works well for our team. However, when addressing the board, especially non-executives with limited financial risk backgrounds, a more simplified explanation is needed. We can achieve this by inputting the detailed memo into GPT-4 with a prompt like: "Now write a simple explanation of the reasons why we use dynamic hedging for this risk. This will be for Non-Executive Directors some of whom do not have a financial risk background." The result from GPT-4 is a clear, concise overview: “Subject: Overview of Dynamic Hedging for Mortgage Pipeline Risk" Dear Non-Executive Directors, I would like to provide a straightforward explanation of why we use dynamic hedging to manage the risks associated with our fixed-rate mortgage offerings. What is Dynamic Hedging? Dynamic hedging is a method we use to protect our company from the financial risks that come with changes in interest rates. It’s like adjusting the sails on a boat to smoothly navigate through changing winds. We make small, regular adjustments to our financial 'sails' to keep our risk exposure in balance. Why Do We Need Dynamic Hedging for Mortgage Pipeline Risk? Timing Differences: Imagine someone locks in a mortgage rate today, but the actual loan isn’t finalized until a month later. During that time, interest rates in the market could change. This timing difference creates risk for us. Predicting Customer Behavior is Challenging: Customers might decide not to proceed with their mortgage if interest rates fall, as they might find a better deal elsewhere. On the other hand, if rates rise, they are more likely to stick with their locked-in rate. Predicting these decisions is difficult. Avoiding Over or Under Protection: If we hedge too much (overhedging), we spend more than necessary, affecting our profits. If we hedge too little (underhedging), we are more exposed to the risk of losing money if interest rates move against us. How Does Dynamic Hedging Help? Flexibility: By constantly adjusting our hedges, we can respond quickly to changes in interest rates and customer choices. This flexibility helps us manage risk more effectively. Cost-Effectiveness: Dynamic hedging allows us to only use as much protection as necessary, avoiding unnecessary expenses. Better Risk Control: This approach gives us a finer control over our risk exposure, helping to ensure that we are neither too exposed nor over-protected. In summary, dynamic hedging is like having a skilled driver for a car. The driver can speed up, slow down, and take turns as needed, ensuring a safe and efficient journey. Similarly, dynamic hedging allows us to navigate the ever-changing financial landscape, keeping our risk at an acceptable level while avoiding unnecessary costs. I hope this provides a clearer understanding of why we use dynamic hedging for managing the risk in our mortgage offerings. I am more than happy to discuss this further if you have any questions. Best regards, The Key Point While the output from GPT-4 might not mirror your exact writing style, and it definitely would need more detail, or an accompanying presentation, it serves as an excellent starting point for creating documents suited to your current audience. This rapid transformation of content is a key advantage in our knowledge-driven industry. By effectively disseminating information across diverse groups, we enhance internal discussions and foster better decision-making. This approach not only improves communication flow to different stakeholders, like NEDs, but also elevates strategic discussions that might not have been possible otherwise.

Welcome to the first article in our series, where we delve into the effective use of Large Language Models (LLMs) such as GPT-4 in treasury, markets, and risk management. My engagement with GPT-4 has illuminated its "jagged edge," a term Ethan Mollick uses to describe the model's varied effectiveness across different tasks. While LLMs excel in areas like data analysis and enhancing our understanding of complex subjects, their most practical applications often lie in simpler, productivity-enhancing tasks. This series is dedicated to exploring such uses, particularly in treasury, risk, and markets, aiming to elevate the volume and quality of your work with minimal effort and without any more than access to a Foundation Model. Use Case 1: Streamlining Communication with LLMs GPT-4, a language-centric model, is particularly adept at managing email correspondence. One effective approach is to input the email text directly into GPT-4’s context window or type in your specific question in the prompt window. The effectiveness hinges on a well-constructed prompt that clearly outlines your requirements, context, and the expected type of response. For instance, consider an email inquiring about a 2% overnight decrease in a bond's value. The prompt for GPT-4 might be: "Respond to an email asking why a bond's value fell 2% overnight, attributing it to a rise in interest rates and the bond's long duration." GPT- 4 responds: Subject: Explanation for the 2% Decline in Bond Holding Value Dear [Recipient's Name], I hope this email finds you well. I understand your concerns regarding the recent 2% decline in the value of your bond holding. There are two primary factors that have contributed to this change: Overnight Rise in Interest Rates: When interest rates rise, the value of existing bonds typically falls. This is because new bonds are likely to be issued at the higher current interest rate, making older bonds with lower rates less attractive. As a result, the market value of existing bonds decreases. Bond's Long Duration: The second factor is the long duration of the bond you hold. Bonds with longer durations are more sensitive to changes in interest rates. This means that for a bond with a long duration, even a small increase in interest rates can lead to a significant decrease in its market value. In summary, the overnight increase in interest rates, coupled with the long duration of your bond, has contributed to the 2% decline in its value. It's important to remember that bond prices and interest rates have an inverse relationship, and long-duration bonds are particularly susceptible to changes in interest rates. Please feel free to reach out if you have any further questions or need more clarification. Best regards, [Your Name] The response from GPT-4, though stylistically different from my usual writing, aptly addresses the query within seconds. This demonstrates GPT-4’s capability to handle straightforward and complex business inquiries efficiently. Also, the emotion-neutral tone of GPT-4’s responses ensures balanced and helpful communication. Should the initial response not fully match your perspective, GPT-4 allows for iterative refinement. For example, if you prefer to explain the bond’s fall in value by a decline in present values resulting from rising interest rates, you can modify your prompt to reflect this viewpoint. Let's see this in action, following on and in the same chain of prompts I ask: "Please would you respond explaining the fall in value as a consequence of rising interest rates decreasing the value of future cash flows and this just led to the value of the bond falling” GPT-4 Responds: Subject: Explanation of Your Bond Holding's Decline in Value Due to Rising Interest Rates Dear [Recipient's Name], I appreciate your inquiry about the recent 2% decrease in the value of your bond holding. Let me clarify the situation with a focus on the impact of rising interest rates on bond values, especially concerning the future cash flows of the bond. Rising Interest Rates and Bond Values: The core concept here is that bond prices and interest rates have an inverse relationship. When interest rates rise, the value of existing bonds typically decreases. This is rooted in how bonds are valued. A bond's value is essentially the present value of its future cash flows, which includes periodic interest payments and the return of principal at maturity. Decreased Value of Future Cash Flows: With the increase in interest rates, the present value of the bond's future cash flows decreases. This is because these future cash flows are now discounted at a higher rate, reflecting the new interest rate environment. In simpler terms, the money that bond investors expect to receive in the future is now worth less in today's terms due to the higher prevailing interest rates. Immediate Impact on Bond Prices: This recalibration in the present value of future cash flows leads to an immediate adjustment in the market price of the bond. Hence, the bond's value falls in response to the rise in interest rates. This effect is particularly pronounced for bonds with longer durations, as they have more future payments that are affected by the change in interest rates. In conclusion, the fall in the value of your bond holding is a direct consequence of the rising interest rates, which have reduced the present value of the bond's future cash flows. It's a common market reaction and an important aspect to consider when investing in bonds. If you need any further clarification or assistance, please don't hesitate to reach out. Warm regards, [Your Name] Beyond Basic Replies: Crafting Detailed Responses While many emails can be addressed with brief responses, some require more detailed and nuanced replies. This can be challenging for those more attuned to numbers than words. In such scenarios, GPT-4 serves as an invaluable tool for swiftly crafting well-considered responses. I recommend experimenting with LLMs to see if they can improve not just the content of your communications but also the efficiency of your writing process. Their effectiveness in technical areas like treasury is truly noteworthy. P.S.: A Personal Tip for Using GPT-4 I've found it incredibly straightforward and effective to dictate a rough answer into a Word document and then copy-paste it into GPT-4. This method helps frame the response provided by GPT-4, ensuring it aligns more closely with my initial thoughts and the specific context of the query. This approach streamlines the process, making it more intuitive and aligned with my workflow.