A daily bite-size selection of top business content.

PM edition. Issue number 1257

Latest 10 stories. Click the button for more.

Term: Private Equity

"Private equity (PE) is capital invested in companies not listed on a public stock exchange, where firms raise funds from investors (like pensions, endowments) to buy, improve, and then sell these businesses for profit, often taking an active management role to boost performance." - Private Equity

Private equity represents a strategic investment approach where specialised firms raise capital from institutional investors to acquire ownership stakes in companies not listed on public stock exchanges, implement operational improvements, and subsequently exit through sale or initial public offering (IPO).^1,2

Core Investment Mechanism

Private equity operates through a structured fund model in which general partners (GPs)-the investment managers-raise capital from limited partners (LPs) such as pension funds, endowments, family offices, and insurance companies.² These LPs commit capital for extended periods, typically five to ten years, during which funds remain illiquid.⁵ Rather than funding commitments upfront, GPs execute "capital calls" to deploy investor money as investment opportunities emerge, usually within the first few years of the fund's lifecycle.¹

The investment targets span multiple company lifecycle stages: venture capital (startup companies), growth capital (established companies seeking expansion), and buyouts (mature companies).¹ Notably, private equity can invest in both private companies and publicly-traded firms seeking to be taken private.²

Value Creation and Active Management

A defining characteristic of private equity is the active involvement of fund managers in portfolio company operations.¹ Rather than passive ownership, GPs implement efficiency initiatives, growth strategies, and operational improvements to enhance shareholder value.¹ This hands-on approach typically spans three to ten years, with a standard holding period of three to five years.³ During this period, GPs oversee progress, make strategic adjustments, and prepare companies for exit.²

Exit Strategy and Returns

The ultimate objective involves realising gains through negotiated sale or IPO at valuations significantly higher than entry prices.⁴ Upon exit, limited partners typically receive 80% of proceeds whilst general partners retain 20% in exchange for management efforts and full liability acceptance.² This profit-sharing structure aligns GP incentives with LP returns, creating mutual interest in value creation.

Key Strategies Within Private Equity

Three primary strategies characterise the sector:⁴

• Buyout: Acquisition of mature companies, often through leveraged structures where debt finances a portion of the purchase price
• Growth Equity: Investment in established companies with expansion potential, providing capital and expertise for market growth
• Venture Capital: Early-stage investment in startup companies with high growth potential, typically involving smaller investment sizes

The Investment Cycle

Private equity funds progress through three distinct phases:⁵

• Portfolio Construction (Years 1-4): GPs identify and acquire target companies, deploying capital into identified opportunities whilst implementing initial efficiency measures
• Value Creation (Years 2-7): Continuous oversight and strategic adjustments to improve operational performance and cash flow generation
• Harvest (Years 3-10): Exit execution through sale or IPO, with profit realisation and distribution to investors

Henry Kravis and the Foundations of Modern Private Equity

Henry Kravis stands as the preeminent theorist and practitioner whose career fundamentally shaped modern private equity. Born in 1944, Kravis co-founded Kohlberg Kravis Roberts (KKR) in 1976 alongside Jerome Kohlberg Jr. and George Roberts, establishing what would become one of the world's most influential private equity firms.

Kravis's relationship to private equity extends beyond mere participation; he essentially architected the contemporary leveraged buyout (LBO) model that defines much of the sector today. During the 1980s and 1990s, KKR pioneered the use of debt financing to acquire large, mature companies-a strategy that transformed private equity from a niche investment vehicle into a dominant force in global capital markets. His most celebrated transaction, the 1988 acquisition of RJR Nabisco for $25 billion, remains emblematic of the scale and sophistication that Kravis brought to the industry.

Kravis's strategic philosophy centred on identifying undervalued or underperforming companies with strong cash flows, acquiring them through leveraged structures, implementing rigorous operational improvements, and subsequently exiting at substantial multiples. This approach-combining financial engineering with genuine operational value creation-became the template for modern private equity practice. His emphasis on active management and hands-on involvement in portfolio company operations established the expectation that PE firms would function as strategic partners rather than passive investors.

Beyond deal execution, Kravis demonstrated exceptional skill in fundraising and investor relations, building KKR into an institution capable of raising multi-billion-dollar funds. His ability to communicate investment theses and deliver consistent returns to limited partners established the institutional trust necessary for private equity's explosive growth. By the early 2000s, KKR had become synonymous with private equity excellence, managing assets exceeding $100 billion.

Kravis's influence extended to shaping industry standards around governance, transparency, and performance measurement. He advocated for alignment between GP and LP interests through carried interest structures-ensuring that fund managers bore meaningful financial risk alongside their investors. This alignment principle became foundational to private equity's legitimacy as an asset class.

His biography reflects the broader evolution of private equity itself: from a relatively obscure investment strategy in the 1970s to a dominant force reshaping global business by the 21st century. Kravis's career demonstrates how individual vision, combined with disciplined execution and institutional building, can create lasting market structures. Today, his legacy permeates private equity practice, with most major firms adopting operational frameworks, governance models, and value creation methodologies that trace their intellectual lineage directly to KKR's pioneering work under Kravis's leadership.

Quote: Kristalina Georgieva - Managing Director, IMF

"I myself took training on AI and became a master of Co-pilot because we all have to step forward." - Kristalina Georgieva - Managing Director, IMF

Kristalina Georgieva's statement underscores a pivotal moment in leadership amid artificial intelligence's rapid integration into economies worldwide. Delivered during a World Economic Forum Town Hall in Davos in 2026, addressing dilemmas around growth, her words reflect not only strategic foresight but a hands-on commitment to adaptation. As Managing Director of the International Monetary Fund (IMF), Georgieva has positioned herself at the forefront of navigating AI's dual potential for productivity gains and labour disruption^1,2.

Who is Kristalina Georgieva?

Born in 1953 in Bulgaria, Kristalina Georgieva rose through academia and public service to become one of the most influential economists globally. She holds a PhD in economic modelling and applied economics from Sofia University. Her career spans environmental economics at the World Bank, where she served as Chief Economist for Sustainable Development, to high-level European Union roles, including Commissioner for International Cooperation, Humanitarian Aid and Crisis Response, and Vice-President for Budget and Human Resources. Appointed IMF Managing Director in 2019, she navigated the institution through the COVID-19 pandemic, geopolitical tensions, and now AI-driven transformations. Georgieva's leadership emphasises resilience, equity, and proactive policy-making in uncertain times^1,2.

Context of the Quote: AI's Tsunami on Global Jobs

Georgieva spoke at the WEF 2026 Town Hall on 'Dilemmas around Growth,' where she warned that AI will impact 40% of global jobs over the next few years - enhanced, eliminated, or transformed - rising to 60% in advanced economies. Entry-level positions face the brunt, described by her as a 'tsunami' hitting the labour market. This assessment draws from IMF research highlighting AI's uneven effects: productivity boosts in sectors like agriculture, healthcare, and translation services, yet risks of inequality if skills gaps persist, especially in emerging and low-income countries (20-26% exposure)^1,3,4. Her personal training in AI tools like Microsoft Copilot exemplifies the 'step forward' she advocates, urging leaders and workers to embrace reskilling for AI-enhanced roles¹.

Broader Economic Backdrop in 2026

Georgieva's remarks occur against a backdrop of subdued global growth (projected at 3.3% for 2026, below pre-pandemic 3.8% averages), geopolitical fragmentation, and technological shifts. AI offers a potential 0.1-0.8% annual productivity lift, capable of restoring pre-pandemic trajectories, but demands infrastructure, skills investment, and ethical regulation. She stresses flexibility - teaching 'how to learn' over specific jobs - with Northern Europe exemplifying success through historical education investments^1,2.

Leading Theorists on AI, Productivity, and Labour

Georgieva's views align with seminal thinkers on technology's economic impact:

• Erik Brynjolfsson and Andrew McAfee: MIT scholars and authors of The Second Machine Age, they argue AI marks a qualitative leap from prior automation, targeting cognitive tasks across skill levels. Without policy intervention, it risks widening inequality by favouring capital owners and high-skill workers while displacing middle-skill jobs¹.
• Shoshana Zuboff: Harvard professor and author of The Age of Surveillance Capitalism, Zuboff contends AI systems embed political choices on power and surveillance, urging ethical frameworks to prevent inequality concentration¹.
• Daron Acemoglu and Simon Johnson: MIT economists whose work on automation (e.g., Power and Progress) warns that technological choices determine whether AI drives shared prosperity or elite capture, echoing Georgieva's call for equitable distribution².

These theorists collectively reinforce Georgieva's message: AI's path depends on human agency - through training, regulation, and inclusive policies - rather than inevitability.

Implications for Leaders and Economies

Georgieva's example of mastering Copilot signals that leadership in the AI era requires personal adaptation alongside systemic reforms: upskilling workforces, bridging digital divides, and fostering 'together we are more resilient' collaboration. Her vision positions AI not as a divisive force but a 'miracle' for better jobs and lives, if harnessed proactively^1,2.

Term: Tree search

"Tree search is a fundamental problem-solving algorithm that systematically explores a state space structured as a hierarchical tree to find an optimal sequence of actions leading to a goal." - Tree search

Tree search represents a cornerstone methodology in artificial intelligence for navigating complex decision spaces and discovering optimal solutions. At its core, tree search operates by representing a problem as a hierarchical tree structure, where the root node embodies the initial state, internal nodes represent intermediate states or partial solutions, and leaf nodes denote terminal states or goal states. The algorithm systematically traverses this tree, evaluating different paths and branches to identify the most efficient route from the starting point to the desired objective.

Fundamental Principles

The architecture of tree search relies on several key components working in concert. A search tree is a tree representation of a search problem, with the root node corresponding to the initial condition. Actions describe all available steps, activities, or operations accessible to the agent at each node. The transition model conveys what each action accomplishes, whilst path cost assigns a numerical value to each path traversed. A solution constitutes an action sequence connecting the start node to the target node, and an optimal solution represents the path with the lowest cost among all possible solutions.

Tree search algorithms fundamentally balance two competing objectives: exploration (investigating new branches to discover potentially better solutions) and exploitation (focusing computational resources on promising branches already identified). This balance determines the efficiency and effectiveness of the search process.

Search Methodologies

Tree search encompasses two primary categories of approaches. Uninformed search (also called blind search) operates without domain-specific knowledge about the problem space. These algorithms traverse each tree node systematically until reaching the target, relying solely on the ability to generate successors and distinguish between goal and non-goal states. Uninformed search methods work through brute force, examining nodes without prior knowledge of proximity to the goal or optimal directions.

Conversely, informed search leverages domain knowledge to guide exploration more intelligently. A* search exemplifies this approach, combining the strengths of uniform-cost search and greedy search. A* evaluates potential paths by calculating the cost of each move using heuristic information, enabling the algorithm to prioritise branches most likely to lead toward optimal solutions.

Advanced Tree Search Techniques

Branch prioritisation represents a critical optimisation strategy wherein algorithms measure or predict which branches can lead to superior solutions, exploring these branches first to reach optimal or pseudo-optimal solutions more rapidly. Branch pruning complements this approach by identifying and skipping branches predicted to yield suboptimal solutions, thereby reducing computational overhead.

Branch and bound algorithms exemplify these principles by maintaining bounds or ranges of scoring values at each internal node, computing whether particular subbranches can improve upon the best solution discovered thus far. This systematic elimination of inferior search paths significantly reduces the search space requiring evaluation.

Monte Carlo tree search (MCTS) represents a sophisticated probabilistic variant that combines classical tree search with machine learning principles of reinforcement learning. Rather than exhaustively expanding the entire search space, MCTS performs random sampling through simulations and stores statistics of actions to make increasingly educated choices in subsequent iterations. This approach proves particularly valuable in domains with vast or infinite search spaces, such as board game artificial intelligence, cybersecurity applications, robotics, and text generation.

Practical Applications

Tree search algorithms address diverse problem domains. In chess, for instance, the search tree's root node represents the current board configuration, with each subsequent node describing potential moves by any piece. Since the unconstrained search space would be infinite, algorithms limit exploration to specific depths or numbers of moves ahead. Similarly, in molecular discovery and optimisation, tree search evaluates candidate solutions against reference criteria using scoring functions such as Tanimoto similarity measures.

Key Theorist: Richard E. Korf

Richard E. Korf stands as a preeminent figure in tree search algorithm development and optimisation. Born in the mid-twentieth century, Korf earned his doctorate in computer science and established himself as a leading researcher in artificial intelligence, particularly in search algorithms and heuristic methods. His career, primarily conducted at the University of California, Los Angeles (UCLA), has profoundly shaped modern understanding of tree search efficiency.

Korf's most significant contribution emerged through his development of iterative deepening depth-first search (IDDFS), an algorithm that combines the memory efficiency of depth-first search with the optimality guarantees of breadth-first search. This innovation proved transformative for tree search applications where memory constraints posed critical limitations. His work demonstrated that by iteratively increasing search depth, algorithms could find optimal solutions whilst maintaining linear space complexity rather than exponential requirements.

Beyond IDDFS, Korf advanced the theoretical foundations of admissible heuristics-functions that never overestimate the cost to reach a goal, thereby guaranteeing optimal solutions when used with algorithms like A*. His research on pattern databases and abstraction techniques enabled more sophisticated heuristic development, allowing tree search algorithms to prune vastly larger search spaces. Korf's contributions to understanding the relationship between heuristic quality and search efficiency established principles still guiding algorithm design today.

Throughout his career, Korf has investigated optimal solutions to classic puzzles including the Fifteen Puzzle and Rubik's Cube using tree search methodologies, demonstrating both theoretical elegance and practical computational achievement. His publications have become foundational texts in artificial intelligence education, and his mentorship has influenced generations of researchers developing increasingly sophisticated tree search variants. Korf's work exemplifies how rigorous mathematical analysis of search algorithms can yield practical improvements with profound implications for artificial intelligence applications.

Quote: Dara Khosrowshahi - CEO, Uber

"Where investors can do well is in finding companies that are truly looking to transform themselves using AI versus companies that are 'play-acting' their way into a pretend transformation." - Dara Khosrowshahi - CEO, Uber

Dara Khosrowshahi, CEO of Uber, delivered this pointed observation during a session at the World Economic Forum (WEF) Annual Meeting 2026 in Davos, titled An Honest Conversation on the Hopes and Anxieties of the (New) Economy. Speaking amid discussions on AI's role in reshaping industries, he highlighted the gap between superficial AI initiatives and profound operational overhauls.^1,5

Who is Dara Khosrowshahi?

Born in 1969 in Tehran, Iran, Dara Khosrowshahi fled the Iranian Revolution with his family at age nine, settling in the United States. He graduated from Brown University with a double major in electrical engineering and computer science. Khosrowshahi began his career at Credit Suisse First Boston before joining IAC/InterActiveCorp in 1998, where he rose to lead Expedia as CEO from 2005 to 2017, transforming it into a travel industry powerhouse amid the digital shift.¹ Appointed Uber's CEO in 2017, he navigated the company through scandals, regulatory battles, and the COVID-19 pandemic, achieving profitability in 2023 and expanding into autonomous vehicles, delivery, and freight. Under his leadership, Uber has aggressively integrated AI, using tools like Anthropic's Claude and Anysphere's Cursor to rebuild processes such as customer service from rigid policy adherence to goal-oriented AI reasoning.^1,2

Context of the Quote at Davos 2026

The quote emerged from Khosrowshahi's Davos remarks on genuine versus performative AI adoption. He critiqued companies for 'saying the right words' and applying an 'AI veneer' - tasks like summarising pitches that offer no competitive edge. True transformation demands discarding legacy policies, which he likened to a company's essence, and rebuilding workflows around AI agents with clear objectives, such as enhancing customer satisfaction.^1,2,3 Uber's breakthrough came in customer service: initial AI efforts followed old rules with modest gains, but a ground-up redesign enabled AI to reason dynamically, yielding superior results. Khosrowshahi warned of 'car crashes' - internal failures - en route to success, echoing broader WEF themes of productivity promises versus organisational disruption.^1,2

At Davos, discussions contrasted marginal AI tweaks (e.g., speeding loan approvals by minutes) with radical redesigns compressing cycles from days to minutes via agentic workflows, where humans oversee exceptions.² IMF Managing Director Kristalina Georgieva noted labour markets' unreadiness, with one in ten advanced-economy jobs needing new skills, advocating 'T-shaped' talent: broad AI literacy plus deep expertise.²

Leading Theorists on AI-Driven Corporate Transformation

Erik Brynjolfsson, Director of Stanford's Digital Economy Lab, pioneered research on AI's productivity impacts. His work with MIT's Andrew McAfee in The Second Machine Age (2014) argued digital technologies enable exponential growth but demand complementary innovations like process redesign. Brynjolfsson's recent studies quantify 'AI plus' effects: firms redesigning workflows see 2-3x productivity gains over mere tool adoption, aligning with Khosrowshahi's call to 'throw away old policies'.²

Carl Benedikt Frey and Michael Osborne (2013 Oxford study) quantified automation risks but evolved to emphasise reskilling. Frey's later research stresses 'augmentation' over replacement, advocating workflow redesign for human-AI symbiosis - humans for judgement, AI for execution - mirroring Uber's agentic shift.²

Thomas Davenport, analytics expert and author of The AI Advantage (2018), distinguishes 'cognitive' AI pilots from enterprise-scale integration. He identifies top performers as those pursuing 'top-down workflow redesign', measuring success by cycle-time reductions and throughput, not tool usage metrics - precisely Khosrowshahi's differentiator between 'play-acting' and transformation.²

McKinsey Global Institute theorists, including James Manyika, model AI's $13 trillion GDP boost by 2030 via diffusion into operations, not isolated projects. Their frameworks highlight 'organisational capital' - redesigned roles and governance - as the binding constraint, urging firms to rebuild talent ladders around oversight and innovation.²

Implications for Investors and Strategy

Khosrowshahi's insight guides investors to probe beyond AI announcements: seek evidence of workflow rewiring, policy discards, and measurable outcomes like decision speed. Success stories include Tech Mahindra's multilingual AI handling 3.8 million queries at 92% accuracy, and Uber's service agents.² Challenges persist: 90% of firms plan AI spend increases, yet many face hype disillusionment and skill erosion.¹ Forward-thinking strategies include agentic systems as 'co-workers', redesigned apprenticeships for judgement, and metrics focused on automation depth.²

Term: REPL (Read-Eval-Print Loop)

"REPL (Read-Eval-Print Loop) acts as an external, interactive programming environment-specifically Python-that allows an AI model to manage, inspect, and manipulate massive, complex input contexts that exceed its native token window." - REPL (Read-Eval-Print Loop)

A Read-Eval-Print Loop (REPL) is a simple interactive computer programming environment that takes single user inputs, executes them, and returns the result to the user, with a program written in a REPL environment executed piecewise. The term usually refers to programming interfaces similar to the classic Lisp machine interactive environment or to Common Lisp with the SLIME development environment.

How REPL Works

The REPL cycle consists of four fundamental stages:

• Read: The REPL environment reads the user's input, which can be a single line of code or a multi-line statement.
• Evaluate: It evaluates the code, executes the statement or expression, and calculates its result.
• Print: This function prints the evaluation result to the console. If the code doesn't produce an output, like an assignment statement, it doesn't print anything.
• Loop: The REPL loops back to the start, ready for the next line of input.

The name derives from the names of the Lisp primitive functions which implement this functionality. In Common Lisp, a minimal definition is expressed as:

(loop (print (eval (read))))

where read waits for user input, eval evaluates it, print prints the result, and loop loops indefinitely.

Key Characteristics and Advantages

REPLs facilitate exploratory programming and debugging because the programmer can inspect the printed result before deciding what expression to provide for the next read. The read-eval-print loop involves the programmer more frequently than the classic edit-compile-run-debug cycle, enabling rapid iteration and immediate feedback.

Because the print function outputs in the same textual format that the read function uses for input, most results are printed in a form that could be copied and pasted back into the REPL. However, when necessary to print representations of elements that cannot sensibly be read back in-such as a socket handle or a complex class instance-special syntax is employed. In Python, this is the <__module__.class instance> notation, and in Common Lisp, the #<whatever> form.

Primary Uses

REPL environments serve multiple purposes:

• Interactive prototyping and algorithm exploration
• Mathematical calculation and data manipulation
• Creating documents that integrate scientific analysis (such as IPython)
• Interactive software maintenance and debugging
• Benchmarking and performance testing
• Test-driven development (TDD) workflows

REPLs are particularly characteristic of scripting languages, though their characteristics can vary greatly across programming ecosystems. Common examples include command-line shells and similar environments for programming languages such as Python, Ruby, JavaScript, and various implementations of Java.

State Management and Development Workflow

In REPL environments, state management is dynamic and interactive. Variables retain their values throughout the session, allowing developers to build and modify the state incrementally. This makes it convenient for experimenting with data structures, algorithms, or any code that involves mutable state. However, the state is confined to the REPL session and does not persist beyond its runtime.

The process of writing a new function, compiling it, and testing it on the REPL is very fast. The cycle of writing, compiling, and testing is notably short and interactive, allowing developers to preserve application state during development. It is only when developers choose to do so that they run or compile the entire application from scratch.

Advanced REPL Features

Many modern REPL implementations offer sophisticated capabilities:

• Levels of REPLs: In many Lisp systems, if an error occurs during reading, evaluation, or printing, the system starts a new REPL one level deeper in the error context, allowing inspection and potential fixes without restarting the entire program.
• Interactive debugging: Common Lisp REPLs open an interactive debugger when certain errors occur, allowing inspection of the call stack, jumping to buggy functions, recompilation, and resumption of execution.
• Input editing and context-specific completion over symbols, pathnames, and class names
• Help and documentation for commands
• Variables to control reader and printer behaviour

Historical Context and Key Theorist: John McCarthy

John McCarthy (1927-2011), the pioneering computer scientist and artificial intelligence researcher, is fundamentally associated with the development of REPL concepts through his creation of Lisp in 1958. McCarthy's work established the theoretical and practical foundations upon which modern REPL environments are built.

McCarthy's relationship to REPL emerged from his revolutionary approach to programming language design. Lisp, which McCarthy developed at MIT, was the first language to embody the principles that would later be formalised as the read-eval-print loop. The language's homoiconicity-the property that code and data share the same representation-made interactive evaluation a natural and elegant feature. McCarthy recognised that programming could be fundamentally transformed by enabling programmers to interact directly with a running interpreter, rather than following the rigid edit-compile-run cycle that dominated earlier computing paradigms.

McCarthy's biography reflects a career dedicated to advancing both theoretical computer science and artificial intelligence. Born in Boston, he studied mathematics at Caltech before earning his doctorate from Princeton University. His academic career spanned MIT, Stanford University, and other leading institutions. Beyond Lisp, McCarthy made seminal contributions to artificial intelligence, including pioneering work on symbolic reasoning, the concept of time-sharing in computing, and foundational theories of computation. He was awarded the Turing Award in 1971, the highest honour in computer science, recognising his profound influence on the field.

McCarthy's vision of interactive programming through Lisp's REPL fundamentally shaped how developers approach problem-solving. His insistence that programming should be a dialogue between human and machine-rather than a monologue of compiled instructions-anticipated modern interactive development practices by decades. The REPL concept, emerging directly from McCarthy's Lisp design philosophy, remains central to contemporary programming education, exploratory data analysis, and rapid prototyping across numerous languages and platforms.

McCarthy's legacy extends beyond the technical implementation of REPL; he established the philosophical principle that programming environments should support human cognition and iterative refinement. This principle continues to influence the design of modern development tools, interactive notebooks, and AI-assisted coding environments that prioritise immediate feedback and exploratory interaction.

Quote: Bertrand Russell - Analytical philosopher

"Do not fear to be eccentric in opinion, for every opinion now accepted was once eccentric." - Bertrand Russell - Analytical philosopher

Bertrand Russell's exhortation captures the essence of intellectual progress, reminding us that groundbreaking ideas often begin as outliers dismissed by the mainstream. This perspective stems from his own revolutionary contributions to philosophy and mathematics, where he fearlessly challenged established doctrines to forge new paths in human thought^1,4.

The Man Behind the Quote: Bertrand Russell's Extraordinary Life

Born on 18 May 1872 at Ravenscroft, a countryside estate in Trellech, Monmouthshire, Bertrand Arthur William Russell hailed from an aristocratic British family renowned for its progressive values and political involvement. Despite his privileged origins, his childhood was shadowed by profound emotional isolation following the early deaths of his parents. Raised by stern grandparents, young Bertrand grappled with loneliness and even contemplated suicide during his teenage years. Mathematics and the natural world became his refuge, providing solace and direction amid personal turmoil⁴.

Russell's academic brilliance secured him a scholarship to Trinity College, Cambridge, in 1890, where he studied the Mathematical Tripos under Robert Rumsey Webb. This period honed his analytical prowess and ignited his lifelong quest to unify mathematics with logic. His career spanned authorship, activism, and academia, marked by bold stances on pacifism during the First World War - which cost him his Trinity fellowship - and later campaigns against nuclear weapons. In 1950, he received the Nobel Prize in Literature for his defence of humanitarian ideals and freedom of thought. Russell died on 2 February 1970 at age 97, his ashes scattered in the Welsh mountains per his secular wishes⁴.

Context of the Quote: A Liberal Decalogue for Free Thinkers

The quote originates from Russell's A Liberal Decalogue, a set of ten commandments for liberals published in 1951. It encapsulates his belief in the value of independent thought, urging readers not to shy away from unconventional views. In an era of ideological conformity, Russell drew from his experiences rejecting idealism and embracing logical rigour. The full decalogue promotes virtues like originality and scepticism, reflecting his view that societal advancement hinges on tolerating - and encouraging - eccentricity⁵.

Russell embodied this principle: his work On Denoting (1905) revolutionised philosophical analysis, while his pacifism and critiques of totalitarianism often positioned him as an intellectual maverick. The quote underscores a historical truth - from heliocentrism to evolution, paradigm shifts begin with 'eccentric' ideas that gain acceptance through evidence and debate^2,3.

Leading Theorists and the Rise of Analytic Philosophy

Russell was a founding architect of **analytic philosophy**, a tradition emphasising clarity, logic, and language analysis over metaphysics. This movement transformed Western philosophy in the early twentieth century, rejecting vague idealism for precision⁴.

Key figures include:

• Gottlob Frege (1848-1925): German logician and mathematician whose Begriffsschrift (1879) invented modern predicate logic, providing tools Russell used to dissect meaning and reference.
• G. E. Moore (1873-1958): Russell's Cambridge contemporary who, alongside him, led the revolt against British idealism. Moore's Principia Ethica (1903) prioritised common-sense realism and ethical non-naturalism.
• Alfred North Whitehead (1861-1947): Russell's collaborator on Principia Mathematica (1910-1913), a Herculean effort to derive all mathematics from logical axioms, influencing foundational studies despite Godel's later incompleteness theorems.
• Ludwig Wittgenstein (1889-1951): Russell's student whose Tractatus Logico-Philosophicus (1921) built on Russell's ideas, shifting focus to language's limits, though he later critiqued early analytic positivism.

These thinkers formed an intellectual lineage that prioritised verifiable truth over speculation, aligning with Russell's quote by validating once-eccentric notions like logical atomism through rigorous scrutiny⁴.

Enduring Relevance: Eccentricity as the Engine of Progress

Russell's words resonate in fields from science to social reform, where dissent drives innovation. His legacy - over 40 books, Nobel acclaim, and activism - affirms that fearing eccentricity stifles discovery. As he navigated personal and political storms, Russell proved that accepted truths emerge from bold, once-marginalised opinions^1,3,4.

Term: Tool calling

"Tool calling (often called function calling) is a technical capability in modern AI systems-specifically Large Language Models (LLMs)-that allows the model to interact with external tools, APIs, or databases to perform tasks beyond its own training data." - Tool calling

Tool calling, also known as function calling, is a technical capability that enables Large Language Models (LLMs) to intelligently request and utilise external tools, APIs, databases, and services during conversations or processing tasks.^1,2 Rather than relying solely on information contained within their training data, LLMs equipped with tool calling can dynamically access real-time information, perform actions, and interact with external systems to provide more accurate, current, and actionable responses.^3,4

How Tool Calling Works

The tool calling process follows a structured flow that bridges the gap between language models and external systems:²

1. A user submits a prompt or query to the LLM that may require external data or functionality
2. The model analyses the request and determines whether a tool is needed to fulfil it
3. If necessary, the model outputs structured data specifying which tool to call and what parameters to use
4. The application executes the requested tool with the provided parameters
5. The tool returns results to the model
6. The model incorporates this information into its final response to the user

Critically, the model itself does not execute the functions or interact directly with external systems. Instead, it generates structured parameters for potential function calls, allowing your application to maintain full control over whether to invoke the suggested function or take alternative actions.⁸

Defining Tools and Functions

Tools are defined using JSON Schema format, which informs the model about available capabilities.³ Each tool definition requires three essential components:

• Name: A function identifier using alphanumeric characters, underscores, or dashes (maximum 64 characters)
• Description: A clear explanation of what the function does, which the model uses to decide when to call it
• Parameters: A JSON Schema object describing the function's input arguments and their types

For example, a weather function might be defined with the name get_weather, a description explaining it retrieves current weather conditions, and parameters specifying that it requires a location argument.²

Types of Tool Calling

Tool calling implementations vary in complexity depending on application requirements:¹

• Simple: One function triggered by a single user prompt, ideal for basic utilities
• Multiple: Several functions available, with the model selecting the most appropriate one based on user intent
• Parallel: The same function called multiple times simultaneously for complex requests
• Parallel Multiple: Multiple different functions executed in parallel within a single request
• Multi-Step: Sequential function calling within one conversation turn for data processing workflows
• Multi-Turn: Conversational context combined with function calling, enabling AI agents to interact with humans in iterative loops

Primary Use Cases

Tool calling enables two fundamental categories of functionality:⁴

Fetching Data: Retrieving up-to-date information for model responses, such as current weather conditions, currency conversion rates, or specific data from knowledge bases and APIs. This approach is particularly valuable for Retrieval-Augmented Generation (RAG) systems that require access to external knowledge sources.⁴

Taking Action: Performing external operations such as submitting forms, updating application state, scheduling appointments, controlling smart home devices, or orchestrating agentic workflows including conversation handoffs.^4,5

Practical Applications

Tool calling transforms LLMs from passive information providers into active agents capable of real-world interaction. Common implementations include:⁵

• Conversational agents that answer questions by accessing current data
• Voice AI bots that check weather, look up stock prices, or query databases
• Automated systems that schedule appointments or control connected devices
• Agentic AI workflows that perform complex multi-step tasks

Key Distinction: Tools vs Functions

Whilst the terms are often used interchangeably, a subtle distinction exists. A function is a specific kind of tool defined by a JSON schema, allowing the model to pass structured data to your application. A tool is the broader concept encompassing any external capability or resource-including functions, custom tools with free-form text inputs and outputs, and built-in tools such as web search, code execution, and Model Context Protocol (MCP) server functionality.^2,8

Related Strategy Theorist: Andrew Ng

Andrew Ng (born 1976) is a pioneering computer scientist and AI researcher whose work has profoundly influenced how modern AI systems are designed and deployed, including the development of tool-augmented AI architectures. As a co-founder of Coursera, Chief Scientist at Baidu, and founder of Landing AI, Ng has consistently advocated for practical, production-oriented approaches to artificial intelligence that extend model capabilities beyond their training data.

Ng's relationship to tool calling stems from his broader philosophy that effective AI systems must be grounded in real-world applications. Rather than viewing LLMs as isolated systems, Ng has championed the integration of language models with external tools, databases, and domain-specific systems-an approach that directly parallels modern tool calling implementations. His work on machine learning systems design emphasises the importance of connecting AI models to actionable data and external services, enabling them to operate effectively in production environments.

In his influential writings and lectures, particularly through his "AI for Everyone" initiative and subsequent work on AI transformation, Ng has stressed that the future of AI lies not in larger models alone, but in intelligent systems that can leverage external resources and tools to solve real problems. This perspective aligns precisely with tool calling's core principle: extending LLM capabilities by enabling structured interaction with external systems.

Ng's background includes a PhD in Computer Science from UC Berkeley, where he conducted research in machine learning and robotics. He served as Director of the Stanford Artificial Intelligence Laboratory and has held leadership positions at major technology companies. His contributions to deep learning, transfer learning, and practical AI deployment have shaped industry standards for building intelligent systems that operate beyond their training data-making him a foundational figure in the theoretical and practical development of tool-augmented AI systems like those enabled by tool calling.

Quote: Rudyard Kipling - English writer

"If you can keep your head when all about you are losing theirs and blaming it on you..." - Rudyard Kipling - English writer

This iconic opening line from Rudyard Kipling's poem If-, first published in 1910, encapsulates a timeless blueprint for navigating life's tempests with composure and integrity.^1,3 Written as a paternal exhortation, the poem distils hard-won virtues into a series of conditional challenges, urging the reader - ostensibly Kipling's son John - to cultivate self-mastery amid chaos, doubt, and reversal.^2,5

Rudyard Kipling: The Man Behind the Verse

Joseph Rudyard Kipling (1865-1936), born in Bombay during the British Raj, was a prolific English writer whose works vividly captured imperial India and the human spirit's indomitable core.¹ Educated in England but returning to India as a journalist, Kipling rose to fame with Plain Tales from the Hills (1888) and The Jungle Book (1894), earning the Nobel Prize in Literature in 1907 - the first English-language recipient.³ His life, however, was shadowed by tragedy: the death of his daughter Josephine in 1899 and his son John in 1915 during the First World War, events that infused his later poetry with poignant depth.⁵ If- emerged from this crucible, reportedly inspired by Leander Starr Jameson, leader of the failed Jameson Raid (1895-1896), a botched incursion into the Transvaal that symbolised British imperial overreach and personal fortitude under scrutiny.^1,7

The Context of 'If-': A Poem for Perilous Times

Published in Kipling's collection Rewards and Fairies, If- appeared amid Edwardian Britain's fading imperial certainties and the looming Great War.¹ Framed as 'Brother Square-Toes', it retells the life of George Washington through a father's voice, blending historical homage with universal counsel.¹ The poem addresses adversity head-on: maintaining poise when blamed unjustly, balancing self-trust with humility, enduring lies and hatred without reciprocation, and treating triumph and disaster as 'impostors'.^3,5 It culminates in a vision of mastery - 'Yours is the Earth and everything that's in it, / And - which is more - you'll be a Man, my son!' - championing willpower, humility, and relentless effort over sixty seconds of an 'unforgiving minute'.⁴

Core Themes: Virtues for the Stoic Soul

Kipling's verse extols:

• Composure and Self-Reliance: Retain clarity amid panic and false accusation.^1,2
• Balance in Extremes: Dream without enslavement, think without obsession, and equate success with failure.^3,5
• Resilience and Sacrifice: Rebuild from ruins, risk all without complaint, and persevere through exhaustion via sheer will.⁴
• Humility and Integrity: Engage crowds and kings without losing virtue or common touch; value all but depend on none.⁷

Educators often parse it as paternal wisdom, emphasising patience, honesty, self-belief, and stoic endurance.²

Leading Theorists on Stoicism and Resilience

Kipling's precepts echo ancient Stoicism, the philosophical school founded by Zeno of Citium (c. 334-262 BCE), which teaches virtue as the sole good and equanimity amid externals.⁵ Key figures include:

• Marcus Aurelius (121-180 CE): Roman Emperor and author of Meditations, who advocated treating fortune's reversals with indifference: 'You have power over your mind - not outside events'. His emphasis on rational self-control mirrors Kipling's call to 'keep your head'.⁵
• Epictetus (c. 50-135 CE): Former slave turned philosopher, whose Enchiridion insists: 'It's not what happens to you, but how you react to it that matters'. This aligns with trusting oneself amid doubt and rebuilding with 'worn-out tools'.⁵
• Seneca (c. 4 BCE-65 CE): Statesman and tragedian, who in Letters to Lucilius praised enduring hardship silently, much like Kipling's stoic gambler who loses all yet starts anew without murmur.⁵

Modern interpreters, such as C.S. Lewis in his concept of 'men without chests' from The Abolition of Man (1943), reinforce Kipling's virtues of courage and principled action against emotional excess - virtues Kipling deemed essential for manhood.⁵

Over a century on, If- resonates in boardrooms, sports arenas, and crises, its counsel a lodestar for leaders facing volatility with grace.⁷

Quote: Helima Croft - RBC Capital Markets

"We're now facing what looks like the biggest energy crisis since the oil embargo in the 1970s." - Helima Croft - RBC Capital Markets

The comparison to the 1970s oil embargo carries profound weight in energy markets, and understanding why requires examining both historical precedent and the distinctive characteristics of the current crisis.

The 1973 Oil Embargo: Historical Context

The 1973 Arab oil embargo, triggered by the Yom Kippur War, fundamentally reshaped global energy markets and geopolitics. The Organisation of Arab Petroleum Exporting Countries (OAPEC) imposed an embargo on oil shipments to nations supporting Israel, reducing global oil supplies by approximately 7% and causing crude prices to quadruple from $3 to $12 per barrel within months. The embargo lasted five months but exposed the vulnerability of Western economies to supply disruptions orchestrated through deliberate political action. Beyond the immediate price shock, the embargo triggered stagflation, fuel rationing, long queues at petrol stations, and a fundamental reassessment of energy security across industrialised nations. It demonstrated that energy markets were not merely economic systems but critical infrastructure vulnerable to geopolitical weaponisation.

The Current Crisis: Physical Disruption and Strategic Vulnerability

What distinguishes the current situation is that rather than a deliberate embargo imposed by suppliers, the disruption stems from active military conflict directly targeting energy infrastructure and choking critical shipping routes. The Strait of Hormuz, through which approximately 21% of global petroleum and 25% of liquefied natural gas (LNG) passes, has become what one analyst described as an "effective parking lot with very few tankers going through." This represents not a policy decision but a physical blockade created by military operations and the resulting insurance and security risks that make transit prohibitively dangerous or expensive.

The targeting of energy facilities compounds the supply shock. Qatar's LNG operations-critical to global gas supplies, particularly for Europe and Asia-have been directly targeted. The United Kingdom, which has weaned itself from Russian gas supplies, is heavily dependent on Qatari LNG imports, creating a two-fold vulnerability: the loss of Russian supplies combined with disruption to alternative sources. Europe faces what analysts describe as a "significant energy shock" precisely because it has systematically eliminated Russian energy dependence without securing alternative, stable sources.

Why This May Exceed the 1970s Crisis

Several factors suggest the current disruption could prove more severe than the 1973 embargo. First, the 1970s embargo was time-limited and politically negotiable; the current conflict has no clear endpoint and depends on military outcomes rather than diplomatic resolution. Second, the 1970s crisis affected primarily crude oil; the current crisis simultaneously disrupts both oil and natural gas markets, with LNG prices reflecting substantially higher risk premiums than crude oil. Third, alternative export routes are extremely limited. Whilst the 1973 embargo could theoretically be lifted through negotiation, producers such as Kuwait and southern Iraq lack viable alternative export routes if the Strait remains closed. These become, in the terminology of contemporary analysis, "stranded assets"-resources that cannot reach markets regardless of price.

The duration question remains critical. The 1973 embargo lasted five months; current assessments suggest this disruption could persist far longer, depending on military developments and the timeline that policymakers in Washington define as "success." Extended disruption would create cascading effects: shipping companies and insurers withdrawing from the region, alternative routes becoming congested, and prices remaining elevated not because of scarcity alone but because of the structural inability to move supplies through traditional channels.

Helima Croft and the Analysis of Energy Geopolitics

Helima Croft, Managing Director and Head of Global Commodity Strategy and Middle East and North Africa (MENA) Research at RBC Capital Markets, occupies a distinctive position in contemporary energy analysis. Her role encompasses not merely market forecasting but strategic assessment of how geopolitical events translate into energy market outcomes. As a member of the National Petroleum Council-a select advisory body that informs the U.S. Secretary of Energy on matters relating to oil and natural gas-Croft operates at the intersection of market analysis, policy influence, and strategic intelligence.

Her assessment that current conditions mirror the 1970s crisis reflects her expertise in recognising structural similarities across different historical periods. However, her analysis also emphasises what distinguishes the current moment: the role of drone and missile capabilities, the vulnerability of alternative export routes, and the question of whether security escorts through the Strait or political risk insurance will prove sufficient to incentivise shipping companies to resume normal operations. These are not merely economic questions but strategic ones about the credibility of security guarantees and the risk tolerance of commercial actors operating in conflict zones.

The Theoretical Framework: Energy Security and Geopolitical Risk

The analysis of energy disruption as a geopolitical weapon draws on several theoretical traditions. The concept of "energy security" emerged as a distinct field of study following the 1973 embargo, with scholars examining how nations could reduce vulnerability to supply shocks. Theorists such as Daniel Yergin, whose work on energy history and geopolitics has shaped policy thinking for decades, emphasised that energy markets are inherently political-that supply, pricing, and access reflect power relationships rather than purely economic forces.

More recent scholarship on "critical infrastructure" and "systemic risk" provides additional analytical frameworks. The Strait of Hormuz represents what security theorists call a "chokepoint"-a geographic location whose disruption creates disproportionate systemic effects. The concentration of global energy flows through a narrow maritime passage creates what economists term "tail risk": low-probability but catastrophic outcomes. The current situation represents the actualisation of this theoretical risk.

Contemporary analysis also draws on game theory and strategic studies, examining how military actors calculate the costs and benefits of targeting energy infrastructure. The targeting of Qatar's LNG facilities suggests a deliberate strategy to maximise economic disruption beyond immediate military objectives. This reflects what strategists call "economic coercion through infrastructure targeting"-using energy disruption as a tool of strategic pressure.

Market Implications and the Question of Price Responsiveness

Notably, Croft has observed that despite physical supply disruptions, the price reaction has been "pretty muted" relative to the risk involved. This apparent paradox reflects several dynamics. First, markets may be pricing in expectations of policy intervention-announcements of strategic petroleum reserve releases or diplomatic efforts to secure alternative routes. Second, the market may be discounting the probability of extended disruption, assuming that either military resolution or negotiated settlement will restore flows within a defined timeframe. Third, different commodities show different risk premiums: European natural gas prices, which reflect the region's acute vulnerability, have risen 4-6%, a more accurate reflection of systemic risk than crude oil prices alone.

The question of whether security escorts or political risk insurance will prove sufficient to restore shipping through the Strait remains unresolved. This is not merely a technical question but a strategic one: will commercial actors trust security guarantees in an active conflict zone? The answer will determine whether the current disruption proves temporary or structural.

Conclusion: Historical Echoes and Contemporary Distinctiveness

The comparison to the 1970s oil embargo serves as a useful historical reference point, but the current crisis possesses distinctive characteristics that may render it more severe and more difficult to resolve. The 1973 embargo was a deliberate policy instrument that could be negotiated; the current disruption stems from active military conflict with no clear resolution mechanism. The 1970s crisis affected primarily crude oil; the current crisis simultaneously disrupts oil and natural gas markets. And whilst the 1973 embargo lasted five months, current assessments suggest this disruption could persist far longer, creating structural changes in energy markets, shipping patterns, and geopolitical alignments that will persist long after military operations cease.

Term: Diffusion models

"Diffusion models are a class of generative artificial intelligence (AI) models that create new data instances by learning to reverse a gradual, step-by-step process of adding noise to training data." - Diffusion models

Diffusion models are a class of generative artificial intelligence models that create new data instances by learning to reverse a gradual, step-by-step process of adding noise to training data. They represent one of the most significant advances in machine learning, emerging as the dominant generative approach since the introduction of Generative Adversarial Networks in 2014.

Core Mechanism

Diffusion models operate through a dual-phase process inspired by non-equilibrium thermodynamics in physics. The mechanism mirrors the natural diffusion phenomenon, where molecules move from areas of high concentration to low concentration. In machine learning, this principle is inverted to generate high-quality synthetic data.

The process consists of two complementary components:

• Forward diffusion process: Training data is progressively corrupted by adding Gaussian noise through a series of small, incremental steps. Each step introduces controlled complexity via a Markov chain, gradually transforming structured data into pure noise.
• Reverse diffusion process: The model learns to reverse this noise-addition procedure, starting from random noise and iteratively removing it to reconstruct data that matches the original training distribution.

During training, the model learns to predict the noise added at each step of the forward process by minimising a loss function that measures the difference between predicted and actual noise. Once trained, the model can generate entirely new data by passing randomly sampled noise through the learned denoising process.

Key Components and Architecture

Three essential elements enable diffusion models to function effectively:

• Forward diffusion process: Adds noise to data in successive small steps, with each iteration increasing randomness until the data resembles pure noise.
• Reverse diffusion process: The neural network learns to iteratively remove noise, generating data that closely resembles training examples.
• Score function: Estimates the gradient of the data distribution with respect to noise, guiding the reverse diffusion process to produce realistic samples.

A notable architectural advancement is the Latent Diffusion Model (LDM), which runs the diffusion process in latent space rather than pixel space. This approach significantly reduces training costs and accelerates inference speed by first compressing data with an autoencoder, then performing the diffusion process on learned semantic representations.

Advantages Over Alternative Approaches

Diffusion models offer several compelling advantages compared to competing generative models such as GANs and Variational Autoencoders (VAEs):

• Superior image quality: They generate highly realistic images that closely match the distribution of real data, outperforming GANs through their distinct mechanisms for precise replication of real-world imagery.
• Stable training: Unlike GANs, diffusion models avoid mode collapse and unstable training dynamics, providing a more reliable learning process.
• Flexibility: They can model complex data distributions without requiring explicit likelihood estimation.
• Theoretical foundations: Based on well-understood principles from stochastic processes and statistical mechanics, providing strong mathematical grounding.
• Simple loss functions: Training employs straightforward and efficient loss functions that are easier to optimise.

Applications and Impact

Diffusion models have revolutionised digital content creation across multiple domains. Notable applications include:

• Text-to-image generation (Stable Diffusion, Google Imagen)
• Text-to-video synthesis (OpenAI SORA)
• Medical imaging and diagnostic applications
• Autonomous vehicle development
• Audio and sound generation
• Personalised AI assistants

Mathematical Foundation

Diffusion models are formally classified as latent variable generative models that map to latent space using a fixed Markov chain. The forward process gradually adds noise to obtain the approximate posterior:

q(x_|x_0)

where x_1, \ldots, x_T are latent variables with the same dimensionality as the original data x_0. The reverse process learns to invert this transformation, generating new samples from pure noise through iterative denoising steps.

Theoretical Lineage: Yoshua Bengio and Deep Learning Foundations

Whilst diffusion models represent a relatively recent innovation, their theoretical foundations are deeply rooted in the work of Yoshua Bengio, a pioneering figure in deep learning and artificial intelligence. Bengio's contributions to understanding neural networks, representation learning, and generative models have profoundly influenced the development of modern AI systems, including diffusion models.

Bengio, born in 1964 in Paris and now based in Canada, is widely recognised as one of the three "godfathers of AI" alongside Yann LeCun and Geoffrey Hinton. His career has been marked by fundamental contributions to machine learning theory and practice. In the 1990s and 2000s, Bengio conducted groundbreaking research on neural networks, including work on the vanishing gradient problem and the development of techniques for training deep architectures. His research on representation learning established that neural networks learn hierarchical representations of data, a principle central to understanding how diffusion models capture complex patterns.

Bengio's work on energy-based models and probabilistic approaches to learning directly informed the theoretical framework underlying diffusion models. His emphasis on understanding the statistical principles governing generative processes provided crucial insights into how models can learn to reverse noising processes. Furthermore, Bengio's advocacy for interpretability and theoretical understanding in deep learning has influenced the rigorous mathematical treatment of diffusion models, distinguishing them from more empirically-driven approaches.

In recent years, Bengio has become increasingly focused on AI safety and the societal implications of advanced AI systems. His recognition of diffusion models' potential-both for beneficial applications and potential risks-reflects his broader commitment to ensuring that powerful generative technologies are developed responsibly. Bengio's continued influence on the field ensures that diffusion models are developed with attention to both theoretical rigour and ethical considerations.

The connection between Bengio's foundational work on deep learning and the emergence of diffusion models exemplifies how theoretical advances in understanding neural networks eventually enable practical breakthroughs in generative modelling. Diffusion models represent a maturation of principles Bengio helped establish: the power of hierarchical representations, the importance of probabilistic frameworks, and the value of learning from data through carefully designed loss functions.