A daily bite-size selection of top business content.

PM edition. Issue number 1252

Latest 10 stories. Click the button for more.

Quote: Kevin Book - Clearview Energy Partners

"When analysts have looked at the things that could go wrong in global oil markets, [the Strait of Hormuz blockade] is about as wrong as things could go at any single point of failure." - Kevin Book - Clearview Energy Partners

Kevin Book's stark assessment captures the gravity of the Strait of Hormuz closure, a chokepoint through which approximately 20% of global crude oil and natural gas flows, now halted by an unprecedented insurance-driven shutdown triggered by the ongoing Iran war.¹ This event, unfolding since early 2026, has plunged world energy markets into turmoil, evoking memories of the 1970s oil embargo and threatening the most severe supply disruption at a single vulnerability point.¹

Who is Kevin Book?

Kevin Book serves as co-founder and managing partner of Clearview Energy Partners, a Washington, D.C.-based research firm specialising in energy markets, commodities, and geopolitical risk analysis.^1,2 With decades of experience, Book is a recognised authority frequently consulted by media outlets including NPR, Fox News, and industry podcasts for his insights on oil price volatility and supply chain disruptions.^1,2,3 His commentary on Fox News and YouTube discussions has highlighted the potential for Iranian retaliation to spike global oil prices through Hormuz interference, positioning him as a leading voice in navigating the intersection of warfare and energy economics.^2,3

Context of the Quote: The Iran War and Hormuz Shutdown

The quote arises from coverage of the Iran war's escalation, where drone strikes near the Strait of Hormuz prompted insurers to deem the narrow waterway uninsurable, effectively drying up tanker traffic without a formal blockade.¹ Typically, 20 million barrels of oil transit daily, but the closure has forced producers like Iraq to curtail output due to storage constraints, while attacks on infrastructure in Saudi Arabia, Qatar, and the UAE complicate rerouting efforts.¹ President Trump's response includes U.S. naval escorts and political risk insurance via the Development Finance Corporation (DFC), yet experts doubt its sufficiency given legal limits, finite budgets, and persistent risks to ships and crews.¹

Helima Croft of RBC Capital Markets describes this as the largest energy crisis since the 1970s, driven not by mines or missiles-as in the 1980s Tanker War-but by economical drone tactics that spooked commercial operators.¹ Shipping executives like Stamatis Tsantanis emphasise seafarer safety and environmental hazards in the strait's S-curve, underscoring why traffic remains stalled despite U.S. interventions.¹

Historical Backstory: The Strait of Hormuz as Global Oil's Achilles Heel

The Strait of Hormuz, a 33-kilometre-wide passage between Iran and Oman, has long been flagged as the world's most critical oil chokepoint by bodies like the U.S. Energy Information Administration (EIA). Iran has repeatedly threatened closure during tensions, but the 2026 war marks the first effective halt, amplifying fears realised in war games and risk models.¹

Precedents include the 1980s Iran-Iraq War's Tanker War, where attacks sank over 500 vessels, prompting U.S. reflagging and escorts of 2,500 tankers. That era saw oil prices double amid uncertainty, though global recessions tempered impacts. Earlier, the 1973 Arab oil embargo quadrupled prices via production cuts, not transit blocks, teaching lessons in strategic reserves now strained by current shortfalls.¹

Leading Theorists and Analysts on Oil Geopolitics

• Helima Croft (RBC Capital Markets): Global head of commodity strategy, Croft pioneered analysis of insurance-driven disruptions, predicting Hormuz risks from asymmetric threats like drones over conventional blockades.¹
• William Henagan (Council on Foreign Relations): Expert on maritime security, Henagan critiques DFC insurance limits in war zones, stressing financial and legal barriers to resuming trade.¹
• Daniel Yergin: Pulitzer-winning author of The Prize and vice chairman at S&P Global, Yergin theorised 'chokepoint vulnerabilities' in works like The New Map, forecasting Hormuz as a flashpoint where minimal action yields maximal disruption-a prophecy validated in 2026.¹
• Amy Myers Jaffe: Energy geopolitics professor at NYU, Jaffe's research on Middle East supply shocks emphasises alternate routes' inadequacies, aligning with current Gulf infrastructure hits.¹

These theorists collectively warn that Hormuz represents a 'single point of failure' in asymmetric warfare, where low-cost Iranian tactics exploit commercial risk aversion, outpacing military countermeasures and reshaping global energy security doctrines.¹

Term: Model density

"Model density" in AI, particularly regarding LLMs, is a performance-efficiency metric defined as the ratio of a model's effective capability (performance) to its total parameter size." - Model density

Model density represents a fundamental shift in how we measure artificial intelligence performance, moving beyond raw computational power to assess how effectively a model utilises its parameters. Rather than simply counting the number of parameters in a neural network, model density quantifies the ratio of effective capability to total parameter count, revealing how intelligently a model has been trained and architected.³

The Core Concept

At its essence, model density answers a critical question: how much useful intelligence does each parameter contribute? This metric emerged from the recognition that newer models achieve superior performance with fewer parameters than their predecessors, suggesting that progress in large language models stems not merely from scaling size, but from improving architecture, training data quality, and algorithmic efficiency.³

The concept can be understood through what researchers call capability density, formally defined as the ratio of a model's effective parameter count to its actual parameter count.³ The effective parameter count is estimated by fitting scaling laws to existing models and determining how large a reference model would need to be to match current performance. When this ratio exceeds 1.0, it indicates that a model performs better than expected for its size-a hallmark of efficient design.

Information Compression and the "Great Squeeze"

Model density becomes particularly illuminating when examined through the lens of information compression. Modern large language models achieve remarkable density through what has been termed "the Great Squeeze"-the process of compressing vast training datasets into mathematical representations.¹

Consider the Llama 3 family as a concrete example. During training, the model encountered approximately 15 trillion tokens of information. If stored in a traditional database, this would require 15 to 20 terabytes of raw data. The resulting Llama 3 70B model, however, contains only 70 billion parameters with a final weight of roughly 140 gigabytes-representing a 100:1 reduction in physical size.¹ This translates to a squeeze ratio where each parameter has "seen" over 200 different tokens of information during training.¹

The smaller Llama 3 8B model demonstrates even more extreme density, compressing 15 trillion tokens into 8 billion parameters-a ratio of nearly 1,875 tokens per parameter.¹ This extreme over-training paradoxically enables superior reasoning capabilities, as the higher density of learned experience per parameter allows the model to extract more nuanced patterns from its training data.

Semantic Density and Output Reliability

Beyond parameter efficiency, model density extends to the quality and consistency of outputs. Semantic density measures the confidence level of an LLM's response by analysing how probable and semantically consistent the generated answer is.² This metric evaluates how well each answer aligns with alternative responses and the query's overall context, functioning as a post-processing step that requires no retraining or fine-tuning.²

High semantic density indicates strong understanding of a topic and internal consistency, resulting in more reliable outputs.² This proves particularly valuable given that LLMs lack built-in confidence measures and can produce outputs that sound authoritative even when incorrect or misleading.⁵ By generating multiple responses and computing confidence scores between 0 and 1, semantic density identifies responses located in denser regions of output semantic space-and therefore more trustworthy.⁵

Intelligence Density in Practical Application

Beyond parameter ratios, practitioners increasingly focus on intelligence density as the amount of useful intelligence produced per unit of time or computational resource.⁴ This reframing acknowledges that once models achieve sufficient peak intelligence for their intended tasks, the primary constraint shifts from maximum capability to the density of intelligence they can produce.⁴ In customer support and similar domains, this means optimising the quantity of intelligence produced per second becomes more valuable than pursuing ever-higher peak performance.⁴

This principle reveals that high-enough peak intelligence is necessary but not sufficient; once achieved, value creation moves towards latency and density optimisation, where significant opportunities for differentiation remain under-explored and are cheaper to capture.⁴

The Exponential Progress Trend

Research indicates that the best-performing models at each time point show rising capability density, with newer models achieving given performance levels with fewer parameters than older models.³ This trend appears approximately exponential over time, suggesting that progress in large language models is fundamentally about improving efficiency rather than simply scaling up.³ This observation underscores that tracking parameter efficiency is essential for understanding future directions in natural language processing and machine learning.

Related Theorist: Ilya Sutskever and Scaling Laws

The theoretical foundations of model density connect deeply to the work of Ilya Sutskever, Chief Scientist at OpenAI and a pioneering researcher in understanding how neural networks scale. Sutskever's research on scaling laws-particularly his work demonstrating predictable relationships between model size, data size, and performance-provided the mathematical framework upon which modern density metrics rest.

Born in 1986 in Yegoryevsk, Russia, Sutskever emigrated to Canada as a child and developed an early passion for artificial intelligence. He completed his PhD at the University of Toronto under Geoffrey Hinton, one of the founding figures of deep learning, where he focused on understanding the principles governing neural network training and optimisation.

Sutskever's seminal work on scaling laws, conducted whilst at OpenAI alongside researchers including Jared Kaplan, revealed that model performance follows predictable power-law relationships with respect to compute, data, and model size.³ These discoveries fundamentally changed how the field approaches model development. Rather than viewing larger models as inherently better, Sutskever's work demonstrated that the efficiency with which a model uses its parameters matters profoundly.

His research established that progress in AI is not merely about building bigger models, but about understanding and optimising the relationship between parameters and capability-the very essence of model density. Sutskever's theoretical contributions directly enabled the concept of capability density, as researchers could now quantify how much "effective" capacity a model possessed relative to its actual parameter count. His work demonstrated that architectural innovations, superior training algorithms, and higher-quality data could yield models that achieve better performance with fewer parameters, validating the principle that density-not size-drives progress.

Sutskever's influence extends beyond scaling laws to shaping how the entire field conceptualises model efficiency. His emphasis on understanding the mathematical principles underlying neural network training rather than pursuing brute-force scaling has become increasingly relevant as computational costs and environmental concerns make parameter efficiency paramount. In this sense, model density represents the practical realisation of Sutskever's theoretical insights: the recognition that intelligent design and efficient parameter utilisation outweigh raw computational scale.

Quote: Larry Page - Google co-founder

"Sometimes it is important to wake up and stop dreaming." - Larry Page - Google co-founder

This deceptively simple observation emerged from one of the most consequential moments in technology history. In 2009, speaking at his alma mater's commencement ceremony, Larry Page shared the origin story of Google-a company that would fundamentally reshape how humanity accesses information. The quote encapsulates a philosophy that has defined not only Page's career but also influenced an entire generation of entrepreneurs and innovators: the critical distinction between idle dreaming and purposeful action.

The Midnight Revelation

Page's reflection was rooted in a specific, transformative experience. At age 23, whilst a doctoral student at Stanford University, he awoke in the middle of the night with a vivid idea: what if one could download the entire web, extract and preserve only the hyperlinks, and use that structure to understand information relationships? ⁴ Rather than allowing this vision to fade-as most midnight inspirations do-Page immediately grabbed a pen and began writing down the details, spending the remainder of that night scribbling out technical specifications and convincing himself the concept would actually work. ⁴

This moment crystallises the essence of his message. The dream itself was merely the starting point. What transformed it into Google was the immediate, deliberate action: the pencil, the paper, the rigorous thinking, and ultimately, the decision to pursue what seemed at the time like an audacious, even foolish, ambition.

The Philosophy Behind the Words

Page's philosophy rests on a paradox that challenges conventional wisdom about dreaming and aspiration. Whilst motivational culture often celebrates the importance of dreaming big, Page argues for something more nuanced: dreams are valuable only insofar as they catalyse action. The act of "waking up and stopping dreaming" is not a rejection of ambition but rather a call to transition from imagination to implementation.

This perspective is intimately connected to another of Page's core beliefs: that "mega-ambitious dreams" are often easier to pursue than incremental improvements. ⁵ His reasoning is counterintuitive but compelling-when one pursues truly revolutionary goals, competition is minimal because few people possess both the audacity and the capability to attempt them. ⁵ The barrier to entry is not market saturation but rather the psychological courage required to commit to something genuinely transformative.

Formative Influences: The Leadershape Programme

Page's approach to turning dreams into reality was significantly shaped by his participation in Leadershape, a summer programme at the University of Michigan that he attended during his undergraduate years. ⁴ The programme's central philosophy-to maintain a "healthy disregard for the impossible"-became a guiding principle throughout his career. ⁴ This concept proved instrumental in Page's willingness to pursue Google despite the significant risk of abandoning his doctoral studies at Stanford, a decision he and co-founder Sergey Brin initially hesitated to make.

The Leadershape ethos represents a deliberate cultivation of what might be called "productive audacity"-the ability to envision solutions to major problems without being paralysed by conventional limitations or established market structures. For Page, this was not mere motivational rhetoric but a practical framework for identifying where leverage exists in the world, allowing one to accomplish more with less effort.

The Broader Context: Pragmatism Meets Vision

Page's philosophy sits at the intersection of two seemingly opposed traditions in American thought: the visionary idealism of entrepreneurship and the pragmatic engineering mindset. His father, Carl Victor Page Sr., was a computer scientist and artificial intelligence pioneer; his mother, Gloria, was a programmer. ⁴ This intellectual heritage meant that Page was raised in an environment where ambitious thinking was paired with rigorous technical problem-solving.

The quote also reflects a distinctly Silicon Valley perspective that emerged in the 1990s and early 2000s-the belief that technological progress requires not incremental refinement but revolutionary reimagining. Page has stated explicitly: "Especially in technology, we need revolutionary change, not incremental change." ¹ This conviction shaped Google's approach to search, which fundamentally departed from existing search engine methodologies by leveraging the link structure of the web itself.

The Tension Between Dreaming and Doing

What makes Page's observation particularly insightful is its acknowledgement of a genuine psychological tension. Dreams are ephemeral; they dissolve upon waking unless captured and acted upon immediately. ⁴ Yet dreams are also essential-they provide the imaginative substrate from which genuine innovation emerges. The challenge is not to choose between dreaming and doing but to recognise that the transition between them must be swift and decisive.

This philosophy stands in contrast to certain strands of motivational thinking that emphasise visualisation and positive thinking as ends in themselves. For Page, these are merely preliminary steps. The real work begins when one "wakes up"-when the dream encounters reality and must be tested, refined, and implemented through sustained effort and technical rigour.

Legacy and Contemporary Relevance

Page's perspective has proven remarkably durable. In an era of increasing technological disruption, his insistence on the importance of "mega-ambitious dreams" combined with immediate, purposeful action remains profoundly relevant. The quote speaks to entrepreneurs, innovators, and anyone confronting the gap between aspiration and achievement.

The statement also carries an implicit warning: in a world saturated with motivational content and self-help rhetoric, the ability to distinguish between genuine vision and mere fantasy-and more importantly, the discipline to act decisively when a truly significant opportunity emerges-remains rare and valuable. Page's life and work suggest that this rarity is precisely what creates competitive advantage.

Ultimately, the quote represents Page's mature reflection on a principle that guided the creation of one of history's most consequential companies: that the space between dreaming and doing is not a chasm but a threshold, and that crossing it requires both the courage to recognise a genuinely transformative idea and the discipline to act upon it immediately and relentlessly.

Term: Model weights

"Model weights are the crucial numerical parameters learned during training that define a model's internal knowledge, dictating how input data is transformed into outputs and enabling it to recognise patterns and make predictions." - Model weights

Model weights represent the learnable numerical parameters within a neural network that determine how input data is processed to generate predictions, functioning similarly to synaptic strengths in a biological brain.^1,2,4 These values control the influence of specific features on the output, such as edges in images or tokens in language models, through operations like matrix multiplications, convolutions, or weighted sums across layers.^1,2,3 Initially randomised, weights are optimised during training via algorithms like gradient descent, which iteratively adjust them to minimise a loss function measuring the difference between predictions and actual targets.^1,2,5

In practice, for a simple linear regression model expressed as y = wx + b, the weight w scales the input x to predict y, while b is the bias term.² In complex architectures like convolutional neural networks (CNNs) or large language models (LLMs), weights include filters detecting textures and fully connected layers combining features, often numbering in billions.^1,2,5 This enables tasks from image classification to real-time translation, with pre-trained weights facilitating transfer learning on custom datasets.¹

Weights are distinct from biases, which add normalisation and extra characteristics to the weighted sum before activation functions, aiding forward and backward propagation.^3,6 Protecting these parameters is vital, as they encode the model's performance, robustness, and decision logic; unauthorised changes can lead to malfunction.⁵ In LLMs, weights boost emphasis on words or associations, shaping generative outputs.³

Key Theorist: Geoffrey Hinton

The preeminent theorist linked to model weights is **Geoffrey Hinton**, often called the 'Godfather of Deep Learning' for pioneering backpropagation and neural network training techniques that optimise these parameters.^1,2 Hinton's seminal 1986 paper with David Rumelhart and Ronald Williams popularised backpropagation, the cornerstone algorithm for adjusting weights layer-by-layer based on error gradients, revolutionising machine learning.^2,4

Born in 1947 in Wimbledon, London, Hinton descends from a lineage of scientists: his great-great-grandfather George Boole invented Boolean logic, his grandfather Charles Howard Hinton coined 'hyperspace', and his great-uncle was logician Bertrand Russell. Initially studying experimental psychology at Cambridge (BA 1970), Hinton earned a PhD in AI from Edinburgh in 1978, focusing on Boltzmann machines-early stochastic neural networks with learnable weights. Disillusioned with symbolic AI, he championed connectionism, simulating brain-like learning via weights.

In the 1980s, amid the first AI winter, Hinton persisted at Carnegie Mellon and Toronto, developing restricted Boltzmann machines for unsupervised pre-training of weights, addressing vanishing gradients. His 2006 breakthrough with Alex Krizhevsky and Ilya Sutskever-training deep belief networks on ImageNet-proved deep nets with billions of weights could excel, sparking the deep learning revolution.¹ At Google Brain (2013-2023), he advanced capsule networks and transformers indirectly influencing LLMs. Hinton quit Google in 2023, warning of AI risks, and won the 2018 Turing Award with Yann LeCun and Yoshua Bengio. His work directly underpins how modern models, including LLMs, learn weights to recognise patterns and predict outcomes.^3,5

Quote: Abraham Lincoln - American president

"I'm a success today because I had a friend who believed in me and I didn't have the heart to let him down" - Abraham Lincoln - American president

Abraham Lincoln's reflection on success reveals a fundamentally relational understanding of achievement-one that stands in stark contrast to the individualistic narratives that often dominate discussions of personal accomplishment. By attributing his success not to his own talents or efforts, but to a friend's belief in him, Lincoln articulates a philosophy that places human connection and moral accountability at the centre of meaningful achievement.¹

The Context of Lincoln's Philosophy

Lincoln's words carry particular weight when considered against the trajectory of his own life. Born on 12 February 1809 in a log cabin in Kentucky, he emerged from profound poverty with minimal formal education.¹ His early years were marked by repeated failures and setbacks-experiences that might have extinguished ambition in lesser individuals. Yet Lincoln persisted, working as a postmaster, surveyor, shopkeeper, and eventually lawyer, roles that kept him intimately connected to ordinary people and their struggles.¹ This grounding in common experience proved formative to his character and his understanding of what success truly meant.

When Lincoln rose to the presidency in 1861, he inherited a nation fractured by the slavery question and on the precipice of civil war. The crucible of the American Civil War would test his definition of success in the most severe manner imaginable. In this context, success could not be measured by personal acclaim or political victory alone. Instead, it demanded the preservation of the Union, the abolition of slavery, and the maintenance of democratic principles-objectives that required extraordinary moral courage and an unwavering commitment to principles despite immense personal and political cost.¹

The Philosophy Behind the Quote

Lincoln's statement reveals several interconnected philosophical commitments. First, it emphasises the role of encouragement and moral support in sustaining perseverance through hardship.¹ The friend who believed in him functioned not merely as a cheerleader, but as a source of validation that made continued effort possible when circumstances might otherwise have counselled surrender.

Second, the phrase "I didn't have the heart to let him down" points to something deeper than mere gratitude. It speaks to accountability, loyalty, and character as the true drivers of achievement.¹ For Lincoln, success was not primarily about personal gain or self-realisation; it was about honouring the trust that others had placed in him. This transforms success from an individual metric into a shared responsibility-a covenant between the person striving and those who have invested belief in their potential.

Third, Lincoln's formulation suggests that success is fundamentally a shared journey, built on belief, responsibility, and the strength drawn from knowing someone stood by you when it mattered most.¹ This perspective inverts the typical hierarchy of achievement. Rather than the successful individual standing alone at the summit, Lincoln positions himself as part of a web of mutual obligation and interdependence.

Intellectual Foundations and Related Thought

Lincoln's philosophy of relational success anticipated themes that would become central to later philosophical and psychological inquiry. His emphasis on the role of belief and encouragement in human development prefigures contemporary research in social psychology and developmental theory, which has consistently demonstrated that external validation and social support are crucial factors in determining whether individuals persist through challenges or abandon their aspirations.

The concept of accountability to others as a motivating force also resonates with virtue ethics traditions, which emphasise character development through relationships and community. Rather than viewing morality and achievement as matters of individual will or rational calculation, virtue ethics-rooted in Aristotelian philosophy-understands human flourishing as inherently social, developed through habituation within communities of practice and mutual accountability.

Lincoln's thinking also aligns with what later thinkers would call the "relational self"-the understanding that identity and capability are not fixed, autonomous properties but are continually constituted through relationships with others. This stands in contrast to the Enlightenment emphasis on the autonomous, rational individual that dominated much nineteenth-century thought.

The Broader Context of Lincoln's Thought on Character

This quote sits within a larger body of Lincoln's reflections on character, responsibility, and human nature. His statement that "Character is like a tree and reputation its shadow" suggests a similar philosophy: what matters is the inner reality of one's character, not the external appearance of success.⁶ His observation that "Nearly all men can stand adversity, but if you want to test a man's character, give him power" reveals his conviction that true character is revealed not in comfortable circumstances but in how one exercises authority and influence.⁴

Lincoln's emphasis on the moral dimensions of success also appears in his assertion that "You cannot escape the responsibility of tomorrow by evading it today."⁴ This captures his understanding that success requires not merely present effort but a sustained commitment to future obligations-a temporal extension of the accountability he emphasises in the quote about his friend.

The Enduring Relevance

Lincoln's philosophy of success remains profoundly relevant in contemporary contexts that often celebrate individual achievement and self-made narratives. His insistence that success is relational-that it depends fundamentally on the belief and support of others-offers a corrective to narratives that obscure the social foundations of individual accomplishment. In doing so, it invites reflection on the networks of support, privilege, and mutual obligation that enable any individual's rise, and on the reciprocal responsibilities that success entails.

The quote also speaks to the question of motivation and meaning. In a culture that often measures success by external markers-wealth, status, power-Lincoln's definition redirects attention to internal measures: the integrity of honouring trust, the dignity of loyalty, and the satisfaction of living up to the belief others have placed in you. This reframing suggests that the deepest forms of success are those that align personal achievement with relational responsibility.

Term: Recursive Language Model (RLM)

"A Recursive Language Model (RLM) is an AI inference strategy where a large language model (LLM) is granted the ability to programmatically interact with and recursively call itself or smaller helper models to solve complex tasks and process extremely long inputs." - Recursive Language Model (RLM)

A **Recursive Language Model (RLM)** is an innovative inference strategy that empowers large language models (LLMs) to treat input contexts not as static strings but as dynamic environments they can actively explore, decompose, and recursively process.^1,3,4 This approach fundamentally shifts AI from passive text processing to active problem-solving, enabling the handling of extremely long inputs, complex reasoning tasks, and structured outputs without being constrained by traditional context window limits.^1,6

At its core, an RLM operates within a Python Read-Eval-Print Loop (REPL) environment where the input context is stored as a programmable variable.^1,2,3 The model begins with exploration and inspection, using tools like string slicing, regular expressions, and keyword searches to scan and understand the data structure actively rather than passively reading it.¹ It then performs task decomposition, breaking the problem into smaller subtasks that fit within standard context windows, with the model deciding the splits based on its discoveries.^1,3

The hallmark is recursive self-calls, where the model invokes itself (or smaller helper models) on each subtask, forming a tree of reasoning that aggregates partial results into variables within the REPL.^1,4 This is followed by aggregation and synthesis, combining outputs programmatically into lists, tables, or documents, and verification and self-checking through re-runs or cross-checks for reliability.¹ Unlike traditional LLMs that process a single forward pass on tokenised input, RLMs grant the model 'hands and eyes' to query itself programmatically, such as result = rlm_query(sub_prompt), transforming context from 'input' to 'environment'.^1,3

RLMs address key limitations like 'context rot'-degradation in long-context performance-and scale to effectively unlimited lengths (over 10 million tokens tested), outperforming baselines by up to 114% on benchmarks without fine-tuning, via prompt engineering alone.^3,6,2 They differ from agentic systems by decomposing context adaptively rather than predefined tasks, and from reasoning models by scaling through recursive decomposition.⁶

Key Theorist: Alex L. Zhang and the MIT Origins

The primary theorist behind RLMs is **Alex L. Zhang**, a researcher affiliated with MIT, who co-authored the seminal work proposing RLMs as a general inference framework.^3,4,8 In his detailed blog and the arXiv paper 'Recursive Language Models' (published around late 2025), Zhang articulates the vision: enabling LLMs to 'recursively call themselves or other LLMs' to process unbounded contexts and mitigate degradation.^3,4 His implementation uses GPT-5 or GPT-5-mini in a Python REPL, allowing adaptive chunking and recursion at test time.³

Alex L. Zhang's biography reflects a deep expertise in AI scaling and inference innovations. Active in 2025 through platforms like his GitHub blog (alexzhang13.github.io), he focuses on practical advancements in language model capabilities, particularly long-context handling.³ While specific early career details are sparse in available sources, his work builds on MIT's disruptive ethos-echoed in proposals like 'why not let the model read itself?'-positioning him as a key figure in the 2026 paradigm shift towards recursive AI architectures.^1,8 Zhang's contributions emphasise test-time compute scaling, distinguishing RLMs from mere architectural changes by framing them as a 'thin wrapper' around standard LLMs that reframes them as stateful programmes.⁵

Experimental validations in Zhang's framework demonstrate RLMs' superiority, such as dramatically improved accuracy on pairwise comparison tasks (from near-zero to over 58%) and spam classification in massive prompts.^2,4 His ideas have sparked widespread discussion, with sources hailing RLMs as 'the ultimate evolution of AI' and a 'game-changer for 2026'.^1,2,7

Quote: George Bernard Shaw

"The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore, all progress depends on the unreasonable man." - George Bernard Shaw - Irish playwright

George Bernard Shaw (1856–1950), the Irish playwright, critic, and Nobel laureate, originated this quote in his 1903 play Man and Superman, specifically in the section "Maxims for Revolutionists."^1,3 Shaw, born in Dublin to a Protestant family amid economic hardship, moved to London in 1876, where he became a leading figure in the Fabian Society—a socialist group advocating gradual reform over revolution—and penned over 60 plays blending wit, philosophy, and social critique.³

Context of the Quote

The line appears in Man and Superman, a philosophical comedy subtitled "A Comedy and a Philosophy," which explores themes of human evolution, will, and societal progress through the character of John Tanner, a revolutionary dreamer pursuing (and fleeing) the spirited Ann Whitefield.¹ In "Maxims for Revolutionists," Shaw distills provocative ideas on human nature, arguing that progress requires challenging the status quo rather than conforming to it. The "reasonable man" accepts the world as is, ensuring stability but stagnation; the "unreasonable man" imposes his vision, driving innovation despite resistance.^1,2,3 Shaw, a Fabian socialist who favored incremental change via education and agitation, used the maxim to celebrate disruptive persistence as essential to societal advancement, echoing his belief in remolding the world "nearer to the heart’s desire."⁴

This idea resonated widely: it inspired sales leaders viewing "unreasonableness" as bold action against excuses²; marketers urging challenge over compromise amid populism⁴; and even Hacker News debates contrasting revolution with evolution⁵. It also titled John Elkington and Pamela Hartigan's 2008 book The Power of Unreasonable People, profiling social and environmental entrepreneurs who create markets for change.⁶

Shaw's Backstory

Shaw rejected conventional jobs, surviving as a music and theater critic under pseudonyms like "Corno di Bassetto" while writing novels that flopped. His breakthrough came with plays like Mrs. Warren's Profession (1893), censored for exposing prostitution's economic roots, and Pygmalion (1913), later adapted into My Fair Lady. A vegetarian, teetotaler, and spelling reformer, Shaw won the 1925 Nobel Prize in Literature but donated the money for translations of August Strindberg. Politically, he supported women's suffrage, Irish Home Rule, and eugenics (later controversial), and endorsed Soviet experiments while critiquing capitalism. At 94, he broke his hip falling from a ladder while pruning a tree, dying soon after. His works, blending Shavian wit with Nietzschean vitality, remain staples for dissecting power, class, and human drive.^3,4

Leading Theorists on Unreasonableness, Progress, and Adaptation

Shaw's maxim draws from and influenced thinkers on innovation, disruption, and social change. Key figures include:

• Fabian Society Influentials (Shaw's Circle): Shaw co-founded this gradualist socialist group in 1884, named after Roman general Quintus Fabius Maximus Verrucosus (the "Delayer"), who used attrition over direct battle. Sidney and Beatrice Webb advanced "permeation"—infiltrating elites for reform—while Annie Besant agitated for labor rights. Their motto, "educate, agitate, organize," embodied reasoned persistence against orthodoxy, mirroring Shaw's "unreasonable" drive within structured evolution.⁴

• Friedrich Nietzsche (1844–1900): The German philosopher's concepts of the Übermensch (overman) and will to power prefigure Shaw's rebel, urging transcendence of herd morality. In Thus Spoke Zarathustra (1883–1885), Nietzsche celebrates creators who affirm life against nihilistic conformity, influencing Shaw's evolutionary Superman.³ (Inferred link via shared themes in Shaw's play.)

• Social Entrepreneurs (Modern Applications): Elkington and Hartigan highlight "unreasonable" innovators like Muhammad Yunus (Grameen Bank microfinance) and Wendy Kopp (Teach For America), who built markets defying poverty and education norms. Their 2008 book frames Shaw's idea as a blueprint for systemic change via audacious markets.⁶

• Critics and Counter-Theorists: Hacker News commenter "vph" argues the quote overstates revolution, crediting evolution—incremental, "reasonable" adaptation—for true progress, citing Darwinian biology over rupture.⁵ Jim Carroll contrasts it with Fabian delay tactics, warning prudence yields modest fruit while unreasonableness risks chaos.⁴

Shaw's maxim endures as a rallying cry for visionaries, underscoring that all progress depends on the unreasonable man by forcing adaptation on a resistant world.^1,2

Quote: Jensen Huang - Nvidia CEO

"OpenClaw is probably the single most important release of software, probably ever. If you look at... the adoption of it, Linux took some 30 years to reach this level. OpenClaw has now surpassed Linux. It is now the single most downloaded open source software in history, and it took 3 weeks." - Jensen Huang - Nvidia CEO

In a striking declaration at the Morgan Stanley Technology, Media and Telecom Conference in San Francisco, Nvidia CEO Jensen Huang positioned OpenClaw as a revolutionary force in open source software, outpacing even the legendary Linux kernel in adoption speed and scale.⁵ This remark underscores Huang's vision for AI agents - autonomous systems capable of continuous operation and complex tasks - as the next frontier in artificial intelligence, with OpenClaw serving as their foundational framework.⁵

Context of the Quote

Delivered on 4 March 2026, Huang's comments came amid discussions on Nvidia's strategic investments in AI leaders like OpenAI and Anthropic, where he noted that recent deals, including a $30 billion stake in OpenAI, might represent the company's final major private investments before these firms pursue initial public offerings.^1,2,3,5,6 Amid this, Huang pivoted to OpenClaw's meteoric rise, contrasting its three-week dominance in downloads against Linux's three-decade journey to similar prominence.⁵ He highlighted its 'vertical' growth on semi-log charts, attributing this to the insatiable demand for AI agents that process a million times more tokens and run perpetually in enterprise environments.⁵

Who is Jensen Huang?

Jensen Huang co-founded Nvidia in 1993 alongside Chris Malachowsky and Curtis Priem, initially focusing on graphics processing units (GPUs) for gaming and visualisation.⁴ Under his leadership, Nvidia pivoted decisively to AI and high-performance computing, with breakthroughs like CUDA - a parallel computing platform that locks in developers through its ecosystem of software, interconnects like NVLink, and rack-scale systems.⁴ Huang's prescience in positioning GPUs as indispensable for AI training and inference has propelled Nvidia to a market leader, with hyperscalers committing over $660 billion in AI spending for 2026 alone.⁴ His conference appearances, including this one, blend investment insights with technological evangelism, reinforcing Nvidia's moat in the AI stack.^1,3,4,5

What is OpenClaw?

OpenClaw emerges as Nvidia's open source initiative tailored for AI agents - intelligent, persistent programmes that autonomously handle tasks such as software development, tool creation, and data processing.⁵ Unlike traditional software, these agents operate continuously, consuming vast token volumes (a measure of computational language processing) and integrating seamlessly into workflows.⁵ Huang's team deploys numerous OpenClaw instances internally, automating coding and innovation, which explains the explosive download figures: surpassing Linux - the cornerstone of servers, supercomputers, and embedded systems - in just three weeks.⁵ This positions OpenClaw not merely as code, but as infrastructure for the agentic AI era, where autonomy scales intelligence.

Backstory: Linux's Enduring Legacy

To grasp OpenClaw's feat, consider Linux's trajectory. Initiated in 1991 by Linus Torvalds as a hobby project, Linux evolved into the world's most ubiquitous operating system kernel, powering 96% of the top supercomputers, most cloud infrastructure, and Android devices.⁵ Its adoption spanned three decades, driven by open source principles, community contributions, and enterprise embrace from IBM to Google. Yet, as Huang noted, even this benchmark took 30 years to cement Linux as a download and deployment juggernaut.⁵ OpenClaw's subversion of this timeline signals a paradigm shift: AI-driven tools now accelerate adoption via immediate utility in high-stakes domains like enterprise AI.

Leading Theorists in AI Agents and Open Source AI

• Linus Torvalds: Architect of Linux, Torvalds pioneered collaborative open source development via Git, influencing every major software ecosystem. His 'benevolent dictator' governance model ensured Linux's stability and growth, principles echoed in modern AI repositories.⁵
• Ilya Sutskever: Co-founder of OpenAI and key figure in transformer models (the backbone of agents), Sutskever's work on scaling laws demonstrated how compute and data yield emergent intelligence, paving the way for agentic systems like those powered by OpenClaw.
• Andrej Karpathy: Former OpenAI and Tesla AI director, Karpathy advanced accessible AI through nanoGPT and LLM training tutorials, theorising agent swarms - multi-agent collaborations - that align with Huang's vision of continuous, token-hungry OpenClaw deployments.
• Yohei Nakajima: Creator of BabyAGI, an early agent framework, Nakajima theorised task decomposition and self-improvement loops, concepts central to OpenClaw's real-world utility in software engineering and beyond.
• Sam Altman: OpenAI CEO, Altman champions 'agentic AI' as the post-ChatGPT phase, where models act independently. Despite tensions in Nvidia partnerships, his firm's trajectory validates Huang's infrastructure bets.^1,2,3

Huang's endorsement frames OpenClaw as the synthesis of these ideas: open source velocity meets agentic scale, challenging developers to harness AI's full potential.

Implications for AI and Open Source

OpenClaw's ascent heralds a compression of innovation cycles, where AI agents bootstrap their own ecosystems faster than human-led projects like Linux.⁵ For investors and technologists, it reinforces Nvidia's centrality: not just in hardware, but in software that cements lock-in.⁴ As agents proliferate - writing code, optimising systems, and driving revenue - Huang's words invite scrutiny of whether this marks the true democratisation of AI, or Nvidia's deepening dominance in the field.^1,4,5

Term: Mixture of Experts (MoE)

"Mixture of Experts (MoE) is an efficient neural network architecture that uses multiple specialised sub-models (experts) and a gating network (router) to dynamically select and activate only the most relevant experts for a given input." - Mixture of Experts (MoE)

This architectural approach divides a large artificial intelligence model into separate sub-networks, each specialising in processing specific types of input data. Rather than activating the entire network for every task, MoE models employ a gating mechanism-often called a router-that intelligently selects which experts should process each input. This selective activation introduces sparsity into the network, meaning only a fraction of the model's total parameters are used for any given computation.^1,3

Core Architecture and Components

The fundamental structure of MoE consists of two essential elements:⁴

• Expert networks: Multiple specialised sub-networks, typically implemented as feed-forward neural networks (FFNs), each with its own set of learnable parameters. These experts become skilled at handling specific patterns or types of data during training.¹
• Gating network (router): A trainable mechanism that evaluates each input and determines which expert or combination of experts is best suited to process it. This routing function is computationally efficient, enabling the model to make rapid decisions about expert selection.^1,3

In practical implementations, such as the Mixtral 8x7B language model, each layer contains multiple experts-for instance, eight separate feedforward blocks with 7 billion parameters each. For every token processed, the router selects only a subset of these experts (in Mixtral's case, two out of eight) to perform the computation, then combines their outputs before passing the result to the next layer.³

How MoE Achieves Efficiency

MoE models leverage conditional computation to reduce computational burden without sacrificing model capacity.³ This approach enables several efficiency gains:

• Models can scale to billions of parameters whilst maintaining manageable inference costs, since not all parameters are activated for every input.^1,3
• Training can occur with significantly less compute, allowing researchers to either reduce training time or expand model and dataset sizes.⁴
• Experts can be distributed across multiple devices through expert parallelism, enabling efficient large-scale deployments.¹

The gating mechanism ensures that frequently selected experts receive continuous updates during training, improving their performance, whilst load balancing mechanisms attempt to distribute computational work evenly across experts to prevent bottlenecks.¹

Historical Development and Key Theorist: Noam Shazeer

Noam Shazeer stands as the primary architect of modern MoE systems in deep learning. In 2017, Shazeer and colleagues-including the legendary Geoffrey Hinton and Google's Jeff Dean-introduced the Sparsely-Gated Mixture-of-Experts Layer for recurrent neural language models.^1,4 This seminal work fundamentally transformed how researchers approached scaling neural networks.

Shazeer's contribution was revolutionary because it reintroduced the mixture of experts concept, which had existed in earlier machine learning literature, into the deep learning era. His team scaled this architecture to a 137-billion-parameter LSTM model, demonstrating that sparsity could maintain very fast inference even at massive scale.⁴ Although this initial work focused on machine translation and encountered challenges such as high communication costs and training instabilities, it established the theoretical and practical foundation for all subsequent MoE research.⁴

Shazeer's background as a researcher at Google positioned him at the intersection of theoretical machine learning and practical systems engineering. His work exemplified a crucial insight: that not all parameters in a neural network need to be active simultaneously. This principle has since become foundational to modern large language model design, influencing architectures used by leading AI organisations worldwide. The Sparsely-Gated Mixture-of-Experts Layer introduced the trainable gating network concept that remains central to MoE implementations today, enabling conditional computation that balances model expressiveness with computational efficiency.¹

Applications and Performance

MoE architectures have demonstrated faster training and comparable or superior performance to dense language models on many benchmarks, particularly in multi-domain tasks where different experts can specialise in different knowledge areas.¹ Applications span natural language processing, computer vision, and recommendation systems.²

Challenges and Considerations

Despite their advantages, MoE systems present implementation challenges. Load balancing remains critical-when experts are distributed across multiple devices, uneven expert selection can create memory and computational bottlenecks, with some experts handling significantly more tokens than others.¹ Additionally, distributed training complexity and the need for careful tuning to maintain stability and efficiency require sophisticated engineering approaches.¹

Term: AI harness

"A harness (often called an agent harness or agentic harness) is an external software framework that wraps around a Large Language Model (LLM) to make it functional, durable, and capable of taking actions in the real world." - AI harness

An AI harness is the external software framework that wraps around a Large Language Model (LLM) to extend its capabilities beyond text generation, enabling it to function as a persistent, tool-using agent capable of taking real-world actions. Without a harness, an LLM operates in isolation-processing a single prompt and generating a response with no memory of previous interactions and no ability to interact with external systems. The harness solves this fundamental limitation by providing the infrastructure necessary for autonomous, multi-step reasoning and execution.

Core Functions and Architecture

An AI harness performs several critical functions that transform a static language model into a dynamic agent. Memory management addresses one of the most significant constraints of raw LLMs: their fixed context windows and lack of persistent memory. Standard language models begin each session with no recollection of previous interactions, forcing them to operate without historical context. The harness implements memory systems-including persistent context logs, summaries, and external knowledge stores-that carry information across sessions, enabling the agent to learn from past experiences and maintain continuity across multiple interactions.

Tool execution and external action represents another essential function. Language models alone can only produce text; they cannot browse the web, execute code, query databases, or generate images. The harness monitors the model's output for special tool-call commands and executes those operations on the model's behalf. When a tool call is detected, the harness pauses text generation, executes the requested operation in the external environment (such as performing a web search or running code in a sandbox), and feeds the results back into the model's context. This mechanism effectively gives the model "hands and eyes," transforming textual intentions into tangible real-world actions.

Context management and orchestration ensure that information flows efficiently between the model and its environment. The harness determines what information is provided to the model at each step, managing the transient prompt whilst maintaining a persistent task log separate from the model's immediate context. This separation is crucial for long-running projects: even if an AI agent instance stops and a new one begins later with no memory in the raw LLM, the project itself retains memory through files and logs maintained by the harness.

Modular Design and Components

Contemporary harness architectures increasingly adopt modular designs that decompose agent functionality into interchangeable components. Research from ICML 2025 on "General Modular Harness for LLM Agents in Multi-Turn Gaming Environments" demonstrates this approach through three core modules: perception, which processes both low-resolution grid environments and visually complex images; memory, which stores recent trajectories and synthesises self-reflection signals enabling agents to critique past moves and adjust future plans; and reasoning, which integrates perceptual embeddings and memory traces to produce sequential decisions. This modular structure allows developers to toggle components on and off, systematically analysing each module's contribution to overall performance.

Performance Impact and Practical Benefits

The empirical benefits of harness implementation are substantial. Models operating within a harness achieve significantly higher task success rates compared to un-harnessed baselines. In gaming environments, an AI with a memory and perception harness wins more games than the same AI without one. In coding tasks, an AI with a harness that runs and debugs its own code completes programming tasks that a standalone LLM would fail due to runtime errors. The harness essentially compensates for the model's inherent weaknesses-lack of persistence, inability to access external knowledge, and propensity for errors-resulting in markedly improved real-world performance.

Perhaps most significantly, harnesses extend what an AI can accomplish without requiring model retraining. Want an LLM to handle images? Integrate a vision module or image captioning API into the harness. Need mathematical reasoning or complex logic? Add the appropriate tool or module. This extensibility makes harnesses economically valuable: two products using identical underlying LLMs can deliver vastly different user experiences based on the quality and sophistication of their respective harnesses.

Evolution and Strategic Importance

As AI capabilities have advanced, harness design has become increasingly critical to product success. The harness landscape is dynamic and evolving: popular agents like Manus have undergone five complete re-architectures since March 2024, and even Anthropic continuously refines Claude Code's agent harness as underlying models improve. This reflects a fundamental principle: as models become more capable, harnesses must be continually simplified, stripping away scaffolding and crutches that are no longer necessary.

The distinction between orchestration and harness is worth noting. Orchestration serves as the "brain" of an AI system-determining the overall workflow and decision logic-whilst the harness functions as the "hands and infrastructure," executing those decisions and managing the technical details. Both are critical for complex AI agents, and improvements in either dimension can dramatically enhance real-world performance.

Related Theorist: Allen Newell and Cognitive Architecture

Allen Newell (1927-1992) was an American cognitive scientist and computer scientist whose theoretical framework profoundly influences contemporary harness design. Newell's "Unified Theories of Cognition" (UTC), published in 1990, proposed that human cognition operates through integrated systems of perception, memory, and reasoning-three faculties that work in concert to enable intelligent behaviour. This theoretical foundation directly inspired the modular harness architectures now prevalent in AI research.

Newell's career spanned the emergence of cognitive science as a discipline. Working initially at the RAND Corporation and later at Carnegie Mellon University, he collaborated with Herbert Simon to develop the "Physical Symbol System Hypothesis," which posited that physical symbol systems (such as computers) could exhibit intelligent behaviour through the manipulation of symbols according to rules. This work earned Newell and Simon the Turing Award in 1975, recognising their foundational contributions to artificial intelligence.

Newell's UTC represented his mature synthesis of decades of research into human problem-solving, learning, and memory. Rather than treating perception, memory, and reasoning as separate cognitive modules, Newell argued they must be understood as deeply integrated systems operating within a unified cognitive architecture. This insight proved prescient: modern AI harnesses implement precisely this integration, with perception modules processing environmental information, memory modules storing and retrieving relevant context, and reasoning modules synthesising these inputs into coherent action sequences.

The connection between Newell's theoretical work and contemporary harness design is not merely coincidental. Researchers explicitly cite Newell's framework when justifying modular harness architectures, recognising that his cognitive science insights provide a principled foundation for engineering AI systems. In this sense, Newell's work from the 1980s and early 1990s anticipated the architectural requirements that AI engineers would discover empirically decades later when attempting to build capable, persistent, tool-using agents.