Foreword

While other animals at times utilize sticks and stones as tools, humans have harnessed a sequence of transformative technologies: irrigation, levers, boats, guns, locomotives, electricity, computers – and recently, scalable applications of artificial intelligence (AI). It could turn our to be the most transformative yet.

Some argue that attributing intelligence, and with it the title of “AI” to Large Language Models (LLMs) such as ChatGPT is inappropriate. As per the real-world use of the word, LLMs have been bundled under the term Artificial Intelligence or at least AI-related in scientific literature, professional networks, regulatory policy, social media, in common discourse, by the creators of LLMs themselves, and by technology companies offering “AI” services. This report too, considers LLMs as a progression in the field of AI.

This project aims to showcase state of the art AI to trade finance professionals, with the explicit goal for readers: an understanding of the landscape of opportunities in AI facilitating International Trade, as well as for general use. To achieve this, I present a simplified foundational understanding of the mechanisms behind modern AI systems in earlier chapters, as well as their current use cases, limitation, and compliance considerations later on.

Each chapter is designed to be complete in content when read on its own. You may start reading whichever section you would like first. Should you stumble upon an unfamiliar word of phrase, consider consulting the Wordbook below.

For comments or inquiries you may reach me at gregor.mihelac (at) outlook (dot) com

Thanks ❤️

Special thanks to Silja Calac and Duarte Pedreira for their incredible mentorship and energy.

Thanks to: André Casterman; Caitlin MacNamara; Charlotte Russo; Pia Reyes; Sam Fowler-Holmes
for engaging in meaningful conversation that helped me see this topic from a different perspective.

And thank you, for reading!

Wordbook

AI - Artificial Intelligence; systems designed to mimic human intelligence and perform tasks autonomously.
AI chatbot - Conversational software using AI to interact with users, often powered by large language models (LLMs)
Anthropic - AI research company, creators of Claude AI.
Attention - Mechanism in neural networks that allows the model to focus on relevant parts of the input.
Autonomous vehicles - Vehicles capable of sensing their environment and operating without human involvement.
Base64 - Encoding scheme that represents binary data in ASCII string format.
Black box - A system where the internal workings are not visible or understood, only inputs and outputs.
Blockchain - Decentralized, distributed ledger technology used for secure and transparent record-keeping.
Bots - Automated software programs designed to perform specific tasks or simulate human behaviour online.
ChatGPT - Popular AI chatbot developed by OpenAI, capable of generating human-like responses.
Claude - AI assistant created by Anthropic, designed for various tasks and conversations.
Convolutional neural nets - Type of deep learning algorithm commonly used in image recognition tasks.
Copilot – AI assistant developed by Microsoft.
Copyright infringement - Unauthorized use or reproduction of copyrighted material.
GDPR - General Data Protection Regulation; EU law on data protection and privacy.
GitHub - Web-based platform for version control and collaboration, popular among developers.
Hallucinations - Inaccurate or fabricated information generated by AI models.
Inference - Process of an AI model generating outputs based on learned patterns and input data.
Internet of Things (IoT) - Network of interconnected devices that collect and exchange data.
Jailbreak - Exploiting vulnerabilities in AI systems to bypass built-in restrictions or safeguards.
JSON - JavaScript Object Notation; lightweight data interchange format.
Kernel - Core component of an operating system that manages system resources.
LLM - Large Language Model; AI system trained on vast amounts of text data to understand and generate human-like language.
Llama - Open-source large language model developed by Meta.
Machine Learning - Branch of AI that enables systems to automatically learn and improve from experience without being explicitly programmed. It uses algorithms and statistical models to analyze patterns in data, make decisions, and predict outcomes based on infromation found in the traning data.
Meta - Technology company formerly known as Facebook, focused on social media and virtual reality.
Midjourney - AI-powered image generation tool.
Multimodal - AI systems capable of processing and generating multiple types of data (text, images, audio, etc.).
Neural net - Computing system inspired by biological neural networks, used in machine learning.
OpenAI - AI research laboratory consisting of the for-profit corporation OpenAI LP and its parent company.
Output - Information or results produced by an AI system in response to input or queries.
Phishing - Fraudulent attempt to obtain sensitive information by disguising as a trustworthy entity.
Prompt - Input or instruction given to an AI model to generate a specific output or response.
Prompt injection - Manipulating AI systems by crafting input prompts that exploit vulnerabilities.
Python - High-level, general-purpose programming language widely used in data science.
Sentiment analysis - Process of determining the emotional tone or attitude expressed in a piece of text.
Smart contracts - Self-executing contracts with the terms directly written into code that is non-changeable.
Social engineering - Manipulating people into divulging confidential information or performing certain actions.
Stream - Continuous flow of data or information, often used in real-time AI processing.
Token - Unit of text used by the Transformer to process and generate outputs.
Tokenized - Process of breaking down text into individual units (tokens) for AI processing.
Training stage / AI training - Process of teaching AI models to recognize patterns and make decisions using large datasets.
Transformer - Neural network architecture used in many state-of-the-art AI models, particularly in natural language processing.
Wav2Vec - Self-supervised learning approach for speech recognition developed by Facebook AI.

Mentioned sources:

Goldman Sachs (2024) “AI is poised to drive 160% increase in data center power demand”. Article.

Greshake et al. (2023) “Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection”. ArXiv 2302.12173v2

Ji, H., et al. (2022), “Predicting within and across language phoneme recognition performance of self-supervised learning speech pre-trained models”. ArXiv: 2206.12489.

McKinsey (2015) “Four fundamentals of workplace automation”. Article, McKinsey Quarterly.

Paterson, et al. (2022) “The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink”. ArXiv:2204.05149

Radford, A., et al. (2018) “Language Models are Unsupervised Multitask Learners”.

Vaswani, A., et al. (2017) “Attention Is All You Need “. ArXiv: 1706.03762.

Wei, et al. (2023) “Jailbroken: How Does LLM Safety Training Fail?”. ArXiv: 2307.02483