The State of AI in 2024

Introduction

We stand at the dawn of a transformative era. Much like the internet in the 1990s, artificial intelligence (AI) in 2024 is brimming with potential yet surrounded by uncertainty. While AI's progress has been rapid, with powerful models like GPT, Claude, and LLaMA captivating the public’s imagination, the full extent of its impact remains largely unexplored. Businesses are racing to adopt these technologies, but the line between realistic, immediate applications and overhyped promises is often blurred.

This whitepaper explores AI’s current state—where it’s being applied productively, how it compares to previous technological revolutions, and why it is both overestimated and underestimated. By examining genuine use cases where AI is delivering measurable value today, we provide a roadmap for businesses to harness its potential without getting lost in the fog of hype.

AI has already demonstrated its power to enhance productivity, automate complex tasks, and revolutionise industries. Yet, its limitations are equally apparent, with challenges in scalability, non-deterministic outcomes, and concerns over data privacy. This whitepaper will help distinguish between genuine innovations and overpromises, offering practical guidance for leveraging AI to drive meaningful results in your organisation—without the need for significant research or deep technical expertise.

As AI continues to evolve, we must remain mindful of its potential and pitfalls. The goal is not to be swayed by the hype, but to navigate this landscape with clarity, practicality, and a vision for long-term success.
‍

Executive Summary

In 2024, AI is at a pivotal moment, much like the internet was in the early 1990s. The explosion of interest in artificial intelligence, driven by rapid advancements in large language models (LLMs) and other machine learning systems, is drawing comparisons to the early days of the web. However, just as businesses and society took years to fully utilise the internet's potential, the same can be said for AI today. This section explores AI's current state, drawing parallels to the internet’s early development and lessons learned from past technological revolutions.

Introduction to AI’s Current Moment (2024)

Much like the internet in the late 1990s, AI has reached mainstream awareness and is seen as a transformative technology. The public is increasingly aware of AI’s potential, but like the internet decades ago, there is widespread uncertainty about its most impactful uses. This growing interest mirrors the anticipation that surrounded the web in its early stages, where the technology was clearly transformative, yet its full impact was not yet understood.
‍
The major catalyst for AI’s current status was the release of ChatGPT by OpenAI in late 2022 which marked a watershed moment for public understanding of AI capabilities. ChatGPT’s conversational abilities, passing the so-called “Turing Test” in many casual contexts, sparked excitement and alarm about AI’s future. For the first time, many people encountered AI systems that could engage in seemingly intelligent dialogue, which shifted the perception of AI from a niche research interest to a mainstream topic.

Foundational models such as GPT (OpenAI), Claude (Anthropic), and LLaMA (Meta) have laid the groundwork for today’s AI applications. 

These models are trained on vast datasets and are versatile, serving as the backbone for a wide range of AI solutions, from chatbots to creative tools. Much like early internet infrastructure laid the foundation for websites, search engines, and e-commerce, foundational models are enabling the proliferation of AI-powered applications across industries.
‍

Parallels Between AI (2024) and the Internet (1990s)

As businesses in the 1990s were unsure of how to harness the internet’s potential, companies today are grappling with how best to integrate AI into their operations. Both eras are marked by a mixture of excitement and uncertainty—knowing that the technology will change the landscape but not yet fully grasping how. This lack of clarity creates both opportunities and challenges, as businesses experiment with AI use cases.

Similar to the internet’s rapid growth in the 90s, AI is expanding at an unprecedented rate, but its most transformative applications remain undefined. The early internet saw a flood of innovative websites and services, but it wasn’t until much later that the most impactful use cases—like social media, e-commerce, and streaming—emerged. Likewise, AI’s current applications, while impressive, may not represent the most revolutionary ways the technology will ultimately reshape industries.
‍
Companies like OpenAI, Google, and Meta, which were early in developing and deploying foundational models, enjoy a significant first-mover advantage, much like early internet pioneers such as Netscape or AOL. These AI leaders have set the standard for what foundational models can achieve, while other companies rush to catch up or differentiate themselves with specialised applications or improvements.
‍
Both AI today and the internet in the 90s have experienced a hype cycle marked by inflated expectations and speculative investments. The early internet bubble saw a rush of capital into dotcom companies, many of which collapsed during the dotcom bust. Similarly, there is currently a "fog of war" around AI’s true potential, with overpromises and fast-talking startups promising revolutionary AI solutions that may not live up to the hype.

Accelerating AI Adoption

The pace of AI advancements is extraordinary. Foundational models have evolved rapidly, with significant progress made in just a few years. Specialised applications such as GitHub Copilot for code completion, vector search for content retrieval, and AI-enhanced creative tools are examples of how quickly AI is being integrated into everyday workflows.
‍
Just as the internet gradually became embedded in nearly all aspects of life—from email to e-commerce—AI is being integrated into a growing number of tools and platforms. From Microsoft Office to Photoshop, AI-enhanced functionality is becoming ubiquitous, automating tasks, improving efficiency, and enhancing creativity.
‍
The rise of open-source AI models is democratising the field, much like open-source technologies such as Linux and Apache drove innovation during the early internet era. Open-source AI allows smaller companies and individuals to experiment with AI technologies without the massive computational resources required to train large proprietary models, fostering innovation and wider adoption.
‍

The Internet’s Lessons for AI

The internet evolved from a curiosity—used primarily by academics and hobbyists—into an essential tool for businesses and consumers. AI may follow a similar trajectory. While it is currently viewed as impressive but experimental, AI will likely transform into a vital utility for industries ranging from healthcare to finance, as its capabilities are refined and use cases solidified.
‍
The early internet boom was marked by unsustainable business models and overinvestment in speculative startups, leading to the dotcom crash. These early missteps provide cautionary lessons for today’s AI landscape. Companies must be careful not to overpromise and should focus on sustainable, realistic applications of AI that provide measurable value.
‍
Just as competition between internet giants like Microsoft, Google, and Amazon drove innovation, the AI race is spurring rapid advancements. Companies like OpenAI, Google, Meta, X and various open-source communities are pushing the boundaries of AI, creating an ecosystem where breakthroughs happen faster and more frequently thanks to competitive pressure.
‍

AI's Potential to Transform Industries

AI has the potential to redefine entire industries, much like the internet did with retail, communication, and media. As foundational models become more sophisticated and specialised applications are developed, AI will likely become indispensable in areas like customer service, software development, and healthcare.
Just as the internet took years to reach its full potential, AI is still in its early stages. The most transformative AI use cases may not have been imagined yet. Businesses must remain flexible and open-minded, ready to pivot and adapt as the technology evolves.
‍
Businesses today have an opportunity to stay ahead of the curve by experimenting with AI technologies. While the full business potential of AI may not yet be realised, those who invest in exploring AI now stand to benefit when the technology matures, and its use cases become clearer.
‍

Key Takeaways

AI is at a tipping point. Like the internet in the 1990s, AI has reached a critical mass of public awareness, driven by significant developments like ChatGPT, but its full potential and practical business applications are still emerging.
‍
Rapid growth, unclear use cases. AI is advancing at an unprecedented rate, much like the early internet, but its most transformative use cases are not yet fully realised. Businesses are still experimenting to find the best ways to leverage the technology.
‍
First-mover advantage. Companies like OpenAI, with early foundational models, have gained significant market advantages, similar to the early internet pioneers. However, competition and open-source innovation are also shaping the AI landscape.
‍
AI embedded in everything. From coding tools to smartphones, AI is starting to integrate into mainstream products, much like how the internet eventually became part of everyday business and personal tools.
Lessons from the internet era. The dotcom boom and bust teach us that while there is immense potential, overestimating short-term gains or rushing in without a sustainable business model can lead to missteps. Companies should balance excitement with careful, long-term planning.
‍
AI’s future is promising but unclear. Just as the internet took years to fully transform industries, AI's most significant impacts may still lie ahead. Companies that experiment with and invest in AI now will be better positioned to capture long-term value.
‍

AI Hype Cycle—Separating Fact from Fiction

Overestimation of AI Capabilities

AI, particularly models like ChatGPT, has been portrayed as more advanced than it actually is. Many people, including influential podcasters and commentators, believe AI is on the verge of general intelligence or will soon replace humans across a wide range of industries.
‍
Early successes, such as language generation and image creation, have led to overconfidence in AI's potential to solve problems outside its current technical limits, such as true creativity, deep understanding, or context-aware reasoning.

Eye-catching demonstrations of AI capabilities, like generating artwork or passing exams, can mislead people into thinking that these models are applicable in real-world situations without limitation.

People often overestimate the scope of AI, thinking it can seamlessly transfer knowledge across different domains. However, many AI models are highly specialised and may only excel in narrow use cases.

There’s a tendency to anthropomorphise AI, assuming it has deeper reasoning capabilities akin to human thought. In reality, current generation AI lacks true understanding and only provides probabilistic outputs based on patterns in data.

Underestimation of AI's Potential

While the hype around AI leads to overpromising in some areas, genuinely impactful applications—such as code assistance, fraud detection, or supply chain optimisation—are sometimes under-appreciated because they don’t make headlines.

Many businesses underestimate the productivity gains AI can offer today, such as automating mundane tasks, personalising customer experiences, or improving data management, due to a lack of understanding of where to apply AI effectively.
‍
Implementing AI is often seen as requiring heavy investments in research and infrastructure. This has led to many companies being slow to adopt AI, even when existing technologies are mature enough to offer tangible benefits without massive R&D.
‍
Some industries, such as healthcare or legal services, are more conservative and slower to adapt to AI, underestimating its potential to drive improvements in diagnostics, personalised medicine, or legal research.
‍

Distinguishing Genuine Innovation from Overpromise

AI companies and startups promise cutting-edge innovations that are years from practical application, creating a flood of overpromising. The pressure to attract investment leads to exaggerated claims about AI's capabilities and future potential.
‍
Investors and companies may not always be able to distinguish between short-term marketable innovations (like improved chatbots) and long-term breakthroughs that require more substantial research (e.g., fully autonomous driving).
‍
Unlike other technologies, AI lacks industry-wide performance standards or benchmarks to objectively evaluate its capabilities, making it difficult for non-experts to discern between functional, scalable solutions and speculative demos.
‍
Many AI models, especially LLMs, are non-deterministic, meaning their outputs can vary for the same input. This inconsistency can create an impression of instability or inadequacy in real-world applications, further confusing users about AI’s reliability.
‍
AI models, particularly deep learning systems, are often "black boxes," where it is difficult to explain how they arrived at a particular outcome. This complexity leads to misunderstandings about the limitations and potential risks, making it hard to evaluate genuine breakthroughs.
‍
The fast pace of AI development makes it difficult to keep up with new innovations. What is considered cutting-edge today might be surpassed within months, blurring the line between legitimate, sustainable innovations and fleeting trends.
‍
While AI demos can be impressive, translating them into reliable, scalable products for real-world use is a major challenge. For example, an AI model may perform well in controlled environments but struggle with the nuances and unpredictability of actual operational conditions.
‍
Implementing AI in real-world environments often reveals issues related to data quality, bias, and privacy concerns, which are often overlooked in early-stage demonstrations but become critical when attempting to scale.
‍
AI’s current state is both overestimated and underestimated because of the complexity, hype, and the gap between demo-level promise and operational reality. While it is essential to recognise the hype and overpromises, AI already has substantial, under-utilised value in many industries. The key challenge is discerning genuine innovations from overpromises, which requires both a cautious, informed approach and a focus on proven use cases that provide tangible, immediate benefits.

How Big Tech is Riding the AI Wave

The rapid advancement of AI, particularly in the development of large language models (LLMs) and other machine learning technologies, has triggered a wave of investment and excitement—often referred to as the "AI Gold Rush." Companies across industries are pouring resources into AI to stay competitive and capture a share of the economic benefits. However, as demand for more powerful AI models grows, so does the need for enormous computing power and specialised hardware, shaping the landscape of AI innovation.
‍

The AI Boom: Why Companies are Rushing In

Companies like OpenAI, Google, and Meta, which were early adopters of foundational AI models, gained a massive first-mover advantage by developing tools that now underpin a wide range of applications. As AI continues to gain traction, companies that fail to invest risk falling behind, similar to what occurred during the early days of the internet.
‍
The rush to capitalise on AI capabilities has led to significant innovation in sectors such as customer service (via chatbots), software development (e.g., GitHub Copilot), and creative industries (e.g., AI-generated images and videos). These applications are already proving profitable and demonstrate the immense financial potential AI holds.
‍

Computing Power as the Bottleneck

The development of large language models (LLMs) and other AI systems requires massive amounts of computational resources. Training cutting-edge models like GPT-4 or Google Gemini involves processing vast datasets through complex neural networks, which demands exponentially increasing computing power.
‍
To meet these demands, companies rely on specialised hardware, particularly Graphics Processing Units (GPUs) and Tensor Processing Units (TPUs). These are critical for accelerating machine learning workloads. NVIDIA the leading supplier of GPUs, has emerged as one of the most valuable companies in the world, driven by the AI boom.
‍
The need for hardware has transformed AI into a capital-intensive endeavour, with large companies like Amazon Web Services (AWS), Google Cloud, and Microsoft Azure investing heavily in data centres optimised for AI workloads. Cloud providers offer AI-as-a-service platforms (e.g., AWS Bedrock, Azure AI), allowing businesses to access computing power without owning the hardware directly.
‍

The Hardware Supply Chain: Constraints and Challenges

The demand for GPUs has led to global supply chain constraints, with companies scrambling to secure enough hardware to train their models. In some cases, the scarcity of GPUs has slowed down AI development, particularly for startups and smaller enterprises.
‍
The computational resources necessary to train and run AI models are expensive. As companies scale their AI operations, the costs of cloud computing, hardware procurement, and energy consumption for data centres become significant burdens.
‍
Training LLMs is also energy-intensive, raising concerns about the environmental impact of AI development. As companies compete to build bigger and more powerful models, energy efficiency is becoming a key challenge.

The Role of Semiconductor Companies

NVIDIA has become the cornerstone of AI hardware, with its GPUs powering the majority of AI training processes. As demand for more powerful AI grows, NVIDIA’s position in the hardware supply chain has strengthened, with some calling it the “arms dealer” of the AI gold rush.

Companies like Intel, AMD, and Google are also innovating in AI-specific chips to meet the computing power demands. For instance, Google developed TPUs, which are optimised specifically for AI workloads. These companies are working on reducing power consumption and increasing the speed of AI training processes.
‍

The Impact on Startups and Smaller Players

The immense hardware requirements for developing frontier AI models create significant barriers to entry for startups. While open-source models and cloud services like AWS or Google Cloud reduce some of these barriers, the cost of training large models can still be prohibitive.
‍
Some startups are turning to collaborations and partnerships to access the hardware they need, pooling resources to share data centre costs or using open-source models that don’t require the same level of computational power as proprietary LLMs.
‍

Long-Term Trends: Scaling Beyond Hardware Constraints

As the AI industry matures, there is growing interest in developing more compute-efficient AI models that require fewer resources to train and run. Techniques such as model pruning, knowledge distillation, and fine-tuning smaller models for specific tasks are gaining traction to reduce hardware dependencies.
‍
Another trend is moving AI processing to edge devices, allowing models to run locally on consumer hardware such as smartphones, laptops, and IoT devices. This reduces the reliance on centralised cloud computing and data centres, potentially alleviating some hardware and energy constraints.
‍

The Future of AI and Hardware Innovation

The AI gold rush is being fuelled not just by breakthroughs in algorithms, but by massive advancements in hardware and cloud infrastructure. Companies that invest early in both AI and the hardware ecosystem have a significant advantage, but the cost and complexity of AI hardware pose challenges, particularly for smaller players.
‍
As demand for computing power continues to rise, companies must navigate supply chain constraints, manage energy costs, and explore new ways to build more efficient AI models. In the long term, innovations in AI hardware will be just as critical as breakthroughs in AI algorithms, defining which companies succeed in the race to scale AI.
‍

Practical Applications of AI in Business Today

The following AI and LLM applications are widely usable with today’s technology, providing immediate value to businesses in various industries without requiring new research. These applications focus on efficiency, automation, and improving decision-making in existing processes.
‍

Agriculture

• Precision Agriculture. AI helps farmers optimise crop yields by analysing data on weather, soil conditions, and plant health, recommending the best times to plant, water, and harvest.
• Pest and Disease Control. AI models identify early signs of crop diseases or pest infestations, allowing for timely intervention.
• Automated Farming Equipment. AI-powered machines are used for planting, harvesting, and managing crops, improving efficiency and reducing manual labour.
  ‍

Customer Service and Support

• Chatbots and Virtual Assistants. AI-powered chatbots can handle routine customer service queries, reducing costs and freeing up human agents for more complex tasks.
• Call Centre Automation. Large-scale call centres use AI to automate customer interactions, enhancing efficiency and saving costs in industries prioritising volume over personalisation.
  ‍

Content and Knowledge Management

• Vector Search for knowledge management. AI-driven search algorithms improve content retrieval by indexing and finding relevant information faster than traditional keyword-based methods.
• Document Summarisation. AI tools summarise lengthy documents and reports, making information easier to digest and act upon across industries like law, healthcare, and finance.
  ‍

Creative Tools and Media

• Text Generation. Large language models (LLMs) can be used to generate written content such as articles, reports, and marketing copy, especially in cases where high-quality human-level writing isn’t required.
• Image and Video Editing. AI has been integrated into creative tools (e.g., Photoshop) for automating tasks such as image generation, retouching, and enhancement, saving time for designers.
  ‍

Project Management and Workflow Automation

• Workflow Optimisation. AI-powered platforms help businesses optimise workflows by managing tasks, scheduling, and summarising project progress.
• Task Automation in Tools. AI assists in automating repetitive tasks in project management software (e.g., ClickUp, Notion), improving operational efficiency and reducing manual workload.
  ‍

Cybersecurity and Risk Management

• Incident Response Automation. AI systems automatically respond to low-level cyber threats, reducing response time and human intervention.
• Threat Detection. AI systems monitor and detect cybersecurity threats in real-time, automating the prevention of data breaches and responding to low-level incidents.
• Fraud Detection. In finance, AI models detect suspicious activity and fraud patterns, improving risk management without extensive new R&D.
  ‍

Education and Personalised Learning

• Automated Grading. AI automates grading for assignments and tests, freeing up teachers’ time for more critical teaching tasks.
• Personalised Learning. AI can tailor educational content to individual student needs, adapting in real-time based on their performance and learning style.
• Tutoring and Assistance. AI-powered virtual tutors help students understand concepts and complete tasks by offering instant feedback and assistance.
  ‍

Finance and Fraud Detection

• Risk Management. AI models are being used to predict and manage financial risks, helping companies and investors make informed decisions.
• Fraud Detection. Financial institutions use AI to detect and prevent fraudulent transactions by identifying patterns and anomalies in real-time.
• Algorithmic Trading. AI-driven algorithms analyse vast amounts of market data to execute trades based on pre-programmed criteria, aiming to maximise profits with minimal human intervention.
  ‍

Healthcare

• Personalised Treatment. AI models analyse patient data to suggest personalised treatment plans, optimising care without replacing existing protocols.
• Diagnostics Assistance. AI helps in analysing medical images and data to assist in diagnosing diseases, speeding up the diagnostic process without requiring major new breakthroughs.
  ‍

Manufacturing and Maintenance

• Supply Chain Optimisation. AI helps optimise supply chain logistics, improving efficiency and reducing waste.
• Predictive Maintenance. AI systems predict when machinery is likely to fail, helping companies maintain equipment proactively and avoid costly downtime.
• Quality Control. Automated visual inspection systems powered by AI detect defects and ensure product quality in real-time.
  ‍

Retail and E-Commerce

• Customer Insights. Retailers use AI to analyse customer data, predicting trends and helping them stock products more effectively.
• Recommendation Engines. AI powers product recommendations, improving customer experiences and increasing sales conversion rates.
• Dynamic Pricing. AI adjusts pricing based on demand and market conditions, maximising revenue for retailers without manual intervention.
  ‍

Smart Cities and Urban Planning

• Public Safety. AI-driven surveillance and monitoring systems assist law enforcement by identifying potential threats or suspicious activities in real-time.
• Traffic Management. AI is used to monitor and optimise traffic flow in cities, reducing congestion and improving urban mobility.
• Energy Efficiency. AI systems help manage energy consumption in smart grids, reducing waste and improving sustainability in urban environments.
  ‍

Software Development and Engineering

• Problem Breakdown. AI’s ability to help developers deconstruct complex problems into actionable steps.
• Technical Documentation. Automatically generating and enhancing documentation for software, APIs, and systems.
• Getting Started Guides. AI-driven frameworks for initiating projects and overcoming the "blank page" problem.
• Code Assistance. Tools like GitHub Copilot provide real-time code completion, refactoring, and debugging support, improving developer productivity.
• API Understanding. AI helps developers quickly understand and interact with APIs, reducing time spent on documentation.
• Documentation and Problem-Solving. LLMs can generate technical documentation, break down complex problems, and assist with prototyping.
• Generating Throwaway Code. How LLMs help developers create non-production code quickly, useful for prototyping and exploration.
  ‍

Where AI Falls Short Today

Despite the remarkable advancements in AI, particularly with large language models (LLMs), there are significant limitations that restrict AI’s current utility in certain contexts. These limitations affect its output quality, reliability, and scalability, posing challenges for businesses looking to integrate AI into real-world applications.
‍

Trading-off Quality for Quantity in AI-Generated Content

AI models can generate vast amounts of text, code, or images quickly and at scale, which is particularly useful for tasks like drafting content, summarising information, or generating ideas. However, the quality of this output often falls short when compared to human experts. AI lacks deep understanding and context, which can lead to generic, shallow, or even factually incorrect content.
‍
While AI can produce creative works like artwork or poetry, it typically operates within the confines of its training data. It mimics patterns rather than generating truly original ideas. For tasks requiring deep expertise, such as writing technical documentation or developing complex legal arguments, AI-generated content is often not precise or nuanced enough.
‍
In many applications, especially where human-level quality is essential (e.g., medical diagnoses, legal documents), AI-generated content can be inaccurate or misleading. In contrast, for low-stakes tasks like summarising blog posts or generating social media content, the "good enough" quality provided by AI may suffice. However, when accuracy, creativity, or critical thinking are required, human expertise remains superior.
‍

Non-Deterministic Nature Leads to Inconsistent Outputs

AI models, particularly LLMs, generate outputs probabilistically. This non-deterministic nature means that the same prompt can produce different results each time it is run. While this can be useful for creative or brainstorming tasks, it complicates workflows where consistency is critical.

In industries like finance, healthcare, or legal services, inconsistency can be a major problem. A non-deterministic response can lead to varying conclusions based on the same input, which is unacceptable in high-stakes environments where predictability and reliability are paramount. For instance, generating inconsistent legal advice or medical diagnoses can have serious consequences.
‍
The unpredictable nature of AI output can erode user confidence in its reliability. Without clear, consistent reasoning behind its answers, businesses and individuals may struggle to trust AI systems in decision-making processes, particularly when stakes are high. In comparison, human experts, while not infallible, provide more explainable and consistent logic in their decisions.
‍

Difficulty in Moving AI Demos to Production

AI demos often showcase highly polished, impressive capabilities in controlled environments, but translating these into robust, scalable production systems is much more complex. Demos usually operate on well-curated datasets, with models fine-tuned for specific tasks, while real-world applications require handling diverse, noisy, and incomplete data.
‍
Moving from a demo to production often reveals issues related to performance at scale. AI systems that work well in small, controlled environments can struggle with the increased computational demands, data volumes, and integration requirements of enterprise-level deployment. Moreover, fine-tuning AI to specific business needs requires significant technical expertise and resources, which may be outside the capabilities of many organisations.
‍
AI systems are highly sensitive to variations in input data that weren’t encountered during training. In production, data often changes over time, or comes with missing or incorrect entries, which can degrade AI performance. For example, an AI demo for customer service may perform well in idealised conditions but fail when faced with real-world complexities such as regional dialects, incomplete information, or unexpected queries.
‍
Deploying AI into production environments requires ongoing infrastructure support, including regular model retraining, updating, and monitoring to ensure performance remains optimal. Many businesses underestimate the continuous operational and maintenance overhead involved in keeping AI systems reliable and accurate post-deployment.
‍

The Path Forward

While AI has shown enormous potential, its current limitations—particularly in terms of output quality, consistency, and scalability—mean that businesses must carefully evaluate where and how to implement it. For certain tasks, especially those requiring high accuracy, deep expertise, or consistency, AI falls short compared to human experts. Moreover, the transition from impressive AI demos to fully functioning, reliable business systems remains a significant hurdle. However, in lower-stakes, large-scale, or repetitive tasks, AI is already proving to be highly useful, and as the technology evolves, solutions to these limitations may emerge.
‍

Where will new research take AI/LLMs next?

Companies with the ability to invest in R&D have the opportunity to extend the current capability of AI/LLMs by uncovering new ways to leverage large language models (LLMs) beyond the limits of existing use cases.
‍
Here we consider three areas of interest for companies wishing to invest in research to develop new opportunities with AI/LLMs.
‍

1. Unlocking New Business Models

‍To uncover new opportunities for large language models (LLMs) beyond standard applications, businesses can invest in R&D to explore niche use cases where AI potential remains largely untapped. By focusing on highly specialised areas, such as healthcare diagnostics, legal research, or complex systems modelling, companies can develop tailored AI solutions that address specific industry challenges. This allows for creating competitive advantages and unique offerings, particularly as industries increasingly look to AI for deeper insights and improved efficiency.
‍
Investing in R&D not only helps companies gain the potential for a first-mover advantage in emerging AI applications but also enables the possibility of cost savings and process optimisation. Fine-tuning LLMs on domain-specific data can enhance their accuracy and relevance, turning them into valuable tools for tasks that currently lack efficient solutions. For example, LLMs can be adapted for industrial IoT systems to interpret sensor data for predictive maintenance or for niche creative tasks, like composing music in highly specific genres.
‍
To maximise the impact of R&D efforts, businesses should use strategies like pilot projects and interdisciplinary collaboration, which help test and refine new AI applications before full-scale implementation. Additionally, leveraging open-source models can lower the costs and risks associated with experimenting with LLMs, enabling quicker proofs of concept. By navigating challenges such as data quality and ethical considerations, companies can responsibly harness the power of AI, discovering new growth areas and transforming specialised domains through innovation.
‍

2. Vector Search Expansion

‍Vector search is transforming information retrieval by enabling semantic understanding, which goes beyond traditional keyword matching. This is particularly valuable in industries like healthcare, legal, and education, where vast content libraries require precise indexing and retrieval. In healthcare, vector search can enhance clinical research by quickly identifying relevant medical records or studies based on context rather than exact terms. Additionally, it aids radiologists by enabling the retrieval of similar medical images, improving diagnostic accuracy.
‍
In the legal sector, vector search enhances case law research by identifying cases with similar legal principles, even if they don't share exact keywords. This capability is invaluable for thorough legal research and contract analysis, where lawyers need to locate documents with specific clauses or compliance terms efficiently. In education, vector search supports academic research and personalised learning by helping users find relevant papers based on themes or recommend resources aligned with students' interests.
‍
As vector search technology advances, it offers new opportunities for enhanced accuracy and relevance, particularly when combined with domain-specific data and integrated with other AI technologies like NLP and knowledge graphs. By investing in vector search, companies can improve decision-making, streamline content retrieval, and offer personalised experiences, unlocking significant value across content-rich industries.
‍

3. Industry-Specific Fine-Tuning

‍Fine-tuning large language models (LLMs) on private, domain-specific data enables companies to achieve competitive advantages, particularly in industries where specialised knowledge and high accuracy are crucial. By customising LLMs with proprietary data, businesses can enhance the model’s relevance and accuracy for industry-specific tasks. This is especially valuable in fields like healthcare, finance, and legal services, where standard LLMs may struggle with nuanced terminology or complex regulatory language. Fine-tuning ensures the model understands the unique context and requirements of these sectors, resulting in more precise and useful outputs.
‍
This approach also allows businesses to create tailored applications, such as specialised customer service solutions, that align closely with their products or services. By integrating proprietary knowledge directly into the AI, companies can safeguard their intellectual property and maintain a competitive edge. Furthermore, in high-stakes environments like healthcare or finance, fine-tuned models can help meet strict accuracy and compliance standards, supporting better decision-making and operational reliability.
‍
As a long-term strategy, fine-tuning provides scalability and adaptability across various organisational functions. Companies can continually update their models with new data, ensuring they remain current and aligned with industry developments. This ongoing refinement supports the development of AI-driven services that are uniquely suited to a company’s specific needs, establishing a sustained advantage in markets where precision and expertise are paramount.
‍

Practical Guidance on How to Harness AI Today

Adopting AI in a meaningful way starts with identifying real business challenges and aligning AI solutions to address them.
‍
This section offers a step-by-step approach to help integrate AI, focusing on problem-solving over technology for its own sake. By beginning with organisational pain points, selecting the right tools, and validating AI applications in real-world contexts, companies can harness AI to deliver measurable value. Leveraging open-source models enhances flexibility and control, and embedding AI into the broader problem-solving framework ensures its long-term impact. This guide aims to simplify AI adoption, making it a practical tool for driving efficiency and innovation across industries.
‍

1. Start with Pain Points

• Identify the biggest challenges or inefficiencies in the organisation. For example:
  • Where are the operational bottlenecks? Are manual tasks slowing down workflows?
  • What customer support challenges won’t go away? Are customers facing delays or getting insufficient support?
  • Do you have data overload? Are teams struggling to find relevant insights in large datasets?
  • Prioritise these by impact. Assess each problem’s impact on business goals (e.g., revenue growth, customer satisfaction) and prioritise issues that, when solved, will yield high-value results
    ‍

2. Map AI Solutions to Specific Problems

• Align AI use cases (from Section 5 of this document) with identified pain points. For example:
  • Automate high-frequency, low-complexity tasks like data entry, report generation, and scheduling.
  • Deploy chatbots or virtual assistants to handle routine customer inquiries, freeing up human agents for complex queries.
  • Use AI-powered search (e.g., vector search) to improve access to information in content-heavy fields like legal, healthcare, and education.

• Select the Right Tools for the Job:
  • Choose off-the-shelf AI solutions or open-source models that directly address the problem. For example, LLMs for knowledge management or predictive analytics for supply chain optimisation.
    ‍

3. Test and Validate AI Solutions in Real-World Scenarios

• Pilot Programs with Measurable Goals
  • Set up pilot projects that target specific, measurable outcomes. For example, in customer service, measure metrics like response time or customer satisfaction improvement.

• Iterate Based on Feedback
  • Collect feedback from users (e.g., employees, customers) to refine the AI solution, make adjustments to better solve the problem.

• Evaluate Success Criteria:
  • Assess whether the AI solution effectively addresses the problem, using KPIs aligned with the initial pain points (e.g., reduced processing time, increased accuracy).
    ‍

4. Leveraging Open-Source AI to Enhance Flexibility and Control

• Use Open-Source Models for Custom Solutions:
  • For complex or highly specific problems, consider fine-tuning open-source models (e.g., Hugging Face transformers, AWS Bedrock) to create tailored solutions without relying on proprietary platforms (e.g. ChatGPT or Vertex AI).

• Open-source solutions allow businesses to keep data in-house, which is beneficial for privacy-sensitive industries like healthcare and finance.

• Open-source models can be customised or adapted for specific problem spaces, enabling adaptive evolution to changing business needs.
  ‍

5. Integrating AI into the Broader Problem-Solving Framework

• Ensure AI is fully integrated into daily workflows to continuously address the identified challenges, rather than being a standalone tool.

• Create frameworks to monitor AI usage, ensuring ethical, regulatory, and performance standards are met, especially in critical areas like customer data handling.

• Encourage teams to look for new challenges that AI can address, ensuring ongoing value creation and fostering a culture of innovation.
  ‍

AI's Real Impact Today

In 2024, AI stands as a transformative technology at the forefront of industry innovation, much like the internet in its early days. While powerful models like GPT, Claude, and LLaMA have introduced unprecedented capabilities, AI's full potential remains largely unexplored. This whitepaper has examined AI's current applications, comparing its evolution to past technological revolutions, and highlighting the challenges businesses face in distinguishing genuine innovations from overhyped promises.

By focusing on practical, measurable applications, businesses can harness AI’s power to improve productivity, automate complex tasks, and create competitive advantages. However, AI’s limitations—such as scalability, non-deterministic outcomes, and data privacy issues—require careful consideration. Through a clear, practical approach to adoption, organisations can leverage AI to drive meaningful results today while preparing for the technology's ongoing evolution.

AI's trajectory is still unfolding, and its most transformative uses may yet be on the horizon. By approaching AI with clarity and adaptability, businesses can navigate the hype and seize valuable opportunities, positioning themselves for sustained success in an increasingly AI-driven world.
‍