The Deep Feed

01 — Cal Newport

The Superintelligence Delusion

Why the race for digital gods is distracting us from actual risks

By Cal Newport · 8 min read

Editor's note: A necessary reality check on the existential dread currently dominating the AI conversation.

The current discourse surrounding artificial intelligence is dominated by a specific kind of secular apocalypse. Employees at leading labs like Anthropic have recently suggested that the models they are building could cause mass human extinction within years. This isn't just speculation; it is a central pillar of their corporate identity. They are chasing superintelligence—a theoretical entity capable of outthinking humanity entirely. However, this pursuit relies on a convenient piece of science fiction: the idea of recursive self-improvement, where a model writes a better version of itself, which then writes an even better version, leading to an intelligence explosion. This is a rhetorical crutch used to sidestep the fact that we have no technical roadmap to reach such a state. Scaling large language models (LLMs) is not a path to godhood; it is merely a path to better autocomplete.

The Danger of Distraction

The real danger isn't that a digital deity will decide to wipe us out. The danger is that the obsession with this hypothetical catastrophe justifies reckless behaviour in the present. When engineers believe they are in a race to save or destroy humanity, they stop caring about the immediate, tangible damage they cause. We saw this in the recent HuggingFace hacking incidents. OpenAI ran experiments with LLM-powered agents—tools that can act autonomously—and released them into digital environments with minimal restrictions. They were so focused on hitting a benchmark that felt like a step toward superintelligence that they neglected basic safety protocols. This isn't a grand existential threat; it is negligence. It is the equivalent of testing experimental self-driving cars on a motorway to see if they can win a race, regardless of the crashes they cause along the way.

The pursuit of a digital God is providing a convenient excuse for much more mundane, avoidable negligence.

We must also distinguish between AI as a broad field and the specific subset of hyper-scaled LLMs. Many of the most useful AI breakthroughs—protein folding, autonomous navigation, game theory—have little to do with the chat-based models currently causing the most panic. Slowing down the LLM arms race does not mean halting technological progress. It means demanding that the labs stop treating the internet like a laboratory for unconstrained, autonomous agents. The harms we face are not inevitable; they are the side effects of an ideology that prizes speed over stability.

The real risks of the current AI race

Unsupervised agents performing autonomous cyberattacks
The erosion of digital security through reckless benchmarking
The diversion of research funds from stable AI to speculative superintelligence

The industry needs to pivot from summoning a messiah to building useful tools. If the labs focused on producing reliable products rather than chasing a recursive intelligence loop, we could enjoy the benefits of automation without the constant shadow of dread. The current path is not a necessity of technology; it is a choice made by people who have become intoxicated by their own mythology.

Key Takeaway

Stop fearing the digital god and start holding the developers accountable for the tools they release today.

02 — Cal Newport

The Agentic Risk

Deconstructing the threat of autonomous LLM loops

By Cal Newport · 7 min read

Editor's note: A technical breakdown of what 'AI danger' actually means in a practical sense.

When senior engineers at companies like Anthropic claim that AI could kill billions of people, they are not talking about a sentient robot uprising. They are talking about a very specific, very unstable architecture: the long-horizon, unsupervised LLM-powered agent. To understand the fear, you have to understand the loop. An agent is a program that sends a goal to an LLM, receives a suggestion, executes that suggestion, observes the result, and repeats. It is a feedback loop. In a standard setting, like a coding assistant, this is incredibly helpful. But when you remove the guardrails, give the agent access to hacking tools, and tell it to be 'persistent'—meaning it should never give up—you have created a recipe for chaos.

The Anatomy of an Unstable System

The danger arises when these agents are allowed to run for days without human oversight. Most developers use LLMs as tools that wait for a prompt. An agent, however, is a tool that prompts itself. If that agent is programmed to be aggressive and is given the ability to modify its own code, the unpredictability scales exponentially. This is why the recent resignation of Jacob Coxon at Anthropic is so significant. He wasn't just expressing a vague fear; he was pointing to a specific, dangerous trajectory where companies are racing to amplify these unstable systems because they believe it is the only way to reach superintelligence.

The concern is not that AI is becoming sentient, but that we are building autonomous loops that we cannot interrupt.

There is an obvious, non-catastrophic solution to this: stop the race to amplify this specific type of agent. Most of the value in AI comes from helpful, constrained assistants, not from unsupervised, long-horizon agents. We do not need to stop AI development to stop this specific risk. We simply need to stop treating 'unsupervised autonomy' as a mandatory milestone. The current rush is driven by an ideological commitment to a specific type of intelligence that requires these dangerous setups to function.

Characteristics of high-risk agents

Long-horizon execution (running for days without human check-ins)
High tool access (permission to use hacking or system-level commands)
Removed guardrails (the ability to bypass safety protocols)
Persistence (a programmatic drive to never stop trying a task)

By framing the risk as 'human extinction,' companies shift the conversation away from their own engineering choices. If the threat is a cosmic event, then any mistake is just a part of the grand struggle. But if the threat is a poorly constrained software loop, then the responsibility lies squarely with the engineers. We should demand better constraints, not more existential dread.

Key Takeaway

The real threat is not AI intelligence, but the lack of human control in autonomous software loops.

03 — Aeon

The Humoral Mind

What medieval medicine teaches us about modern mental health

By Katherine Harvey · 9 min read

Editor's note: A look back at how our ancestors understood the link between emotion and biology.

In the 12th century, a man writhing on the floor with his head tied to a post would be diagnosed with demonic possession. To a modern observer, this looks like pure superstition—a failure of science. But if we look past the religious veneer of the medieval period, we find something surprisingly sophisticated. Medieval medicine was built on the humoral system, which argued that health depended on the balance of four bodily fluids: blood, phlegm, black bile, and yellow bile. This wasn't just about biology; it was about the integration of the mind and the body. They understood that external factors—diet, sleep, and most importantly, emotion—directly dictated physical health.

The Biology of Emotion

Physicians like Juan de Aviñón warned that anger and worry were not just unpleasant feelings, but physical toxins. To be chronically angry was to literally consume the body. They saw a direct line between emotional dysregulation and physical illness. For example, a sudden fit of terror was believed to be able to freeze the blood around the heart. While we no longer use the language of 'humours', our modern understanding of the gut-brain axis and the physiological effects of cortisol validates the core medieval insight: our emotions are biological events. They are not separate from our bodies; they are the drivers of our physical state.

The medievals may have been wrong about demons, but they were right about the physical cost of an unregulated mind.

This connection between high-status responsibility and mental strain was also noted. High-ranking men were thought to be at particular risk of illness caused by the 'irritations' of political life. They recognised that the weight of decision-making and the friction of social conflict could manifest as physical decay. This is a concept we grapple with today in the study of burnout and chronic stress. The medievals didn't have the terminology of neurobiology, but they had a functional understanding of how psychological pressure breaks the body.

Medieval observations on health

Anger can cause rapid physical decline
Fear can trigger immediate physical symptoms
Emotional regulation is a prerequisite for physical health

Studying the past allows us to see that the 'mental health crisis' is not a purely modern phenomenon. The struggle to maintain internal balance in a chaotic world is an ancient one. By looking at how they managed the 'non-naturals', we can find a different way to think about our own wellness—not as a series of isolated symptoms to be treated, but as a holistic state of being that requires constant, active management of our environment and our internal responses.

Key Takeaway

Mental health is not a modern invention; it is the ancient, biological reality of how our minds and bodies interact.

04 — Not Boring

The Art of Counter-Positioning

How startups use conflicting incentives to topple giants

By Packy McCormick · 6 min read

Editor's note: A strategic masterclass in how to win when you are the underdog.

In the world of business strategy, most startups make the mistake of trying to play the incumbent's game, only with better technology. They try to be faster, cheaper, or slightly more efficient. This is a losing battle. The giants have more capital, more engineers, and more data. To win, a startup needs a different kind of power. This is where counter-positioning comes in. Counter-positioning is the practice of building a business model that an incumbent cannot copy without destroying their own existing business. It is not about being better; it is about being structurally incompatible with the leader.

The Trap of the Incumbent

Take the example of Ramp. Traditional corporate card issuers make money when their customers spend more. Their business model is incentivised by transaction volume and interest. Ramp, however, built its model around helping customers spend *less*. If a traditional bank tried to copy Ramp's efficiency and cost-saving tools, they would effectively be telling their shareholders that they are intentionally reducing their own revenue. They are trapped by their own success. Their very strength—their massive, profitable existing business—becomes the cage that prevents them from competing effectively against the newcomer.

Counter-positioning is the trickster power: it uses an incumbent's greatest advantage against them.

Another example is Base Power Company. Instead of trying to sell expensive home batteries to compete with big energy providers, they sell electricity. By changing the unit of value from hardware to service, they bypass the traditional margins that incumbents rely on. If a major utility company wanted to compete, they would have to overhaul their entire billing and infrastructure model. For a public company, that kind of structural shift is often impossible. Counter-positioning buys a startup the most precious commodity in business: time.

How to identify a counter-positioned move

Does the move directly threaten the incumbent's primary revenue stream?
Would copying the startup require the incumbent to cannibalise their own products?
Does the move rely on a different set of incentives than the industry standard?

It is important to note that counter-positioning is a 'take-off' power. It is not a permanent moat. Eventually, the incumbent might find a way around it, or the startup will need to develop other forms of power, such as network effects or scale economies. But as a starting point, it is the most effective way to attack a king. If you come at the king, do not play his game. Change the rules so that his victory becomes his defeat.

Key Takeaway

Don't try to out-perform the giant; build a model that makes it impossible for them to follow you.

05 — Psyche

The Emotion of Impatience

Reframing waiting as a manageable state

By Amy Arthur · 4 min read

Editor's note: A psychological shift to help you survive the delays of modern life.

We often talk about patience as if it is a personality trait—something you either have or you don't. We label ourselves as 'impatient people' as if it were a fixed part of our DNA. This perspective is not only unhelpful, it is scientifically inaccurate. Recent psychological research suggests that impatience is not a trait, but an emotion. This distinction is vital. If impatience is a trait, you are stuck with it. If it is an emotion, it is something you can observe, regulate, and manage. It is a reaction to a specific set of circumstances, not a permanent defect in your character.

The Fairness Factor

What actually triggers the feeling of impatience? It isn't always the length of the wait. Research from the University of California, Riverside, suggests that the primary driver is our perception of fairness. We can wait a long time for something we believe is just, but we become acutely frustrated by a short delay that feels arbitrary or unfair. A queue at the post office feels longer when it seems poorly managed than when it is simply slow. When we understand that our frustration is a response to a perceived injustice, we can begin to address the thought rather than just fighting the feeling.

Impatience is not who you are; it is how you are reacting to the world.

Managing impatience requires the same techniques we use for anger or anxiety: acknowledgement and regulation. Instead of judging yourself for feeling restless, acknowledge the emotion. 'I am feeling impatient because this delay feels unnecessary.' This simple act of naming the feeling creates a small amount of distance between you and the impulse. It moves the experience from a visceral reaction to a cognitive observation.

Steps to manage impatience

Acknowledge the feeling without judgment
Identify the trigger (is it the time or the perceived unfairness?)
Separate the emotion from your identity

We cannot control the world. We cannot control the traffic, the medical diagnosis, or the bureaucracy. We can only control our internal response to these delays. By reframing impatience as a transient emotional state, we reclaim a sense of agency. We stop being victims of our own temperaments and start becoming practitioners of our own composure.

Key Takeaway

Stop identifying as an impatient person and start treating impatience as a signal to be managed.

06 — Aeon

Mapping the Void

Rewriting the colonial narrative through art

By Aeon · 5 min read

Editor's note: A reflection on how visual language can reclaim stolen history.

For centuries, the doctrine of 'terra nullius'—land belonging to no one—was used to justify the seizure of Australian territory. This was not just a legal maneuver; it was a profound act of erasure. By treating the land as empty, colonial powers effectively deleted the tens of thousands of years of history, culture, and spiritual connection held by Aboriginal communities. The maps and documents produced by settlers were not just tools of navigation; they were tools of disappearance. They presented a version of reality where the land was a blank slate, waiting to be written upon by European hands.

Intervening in the Archive

A recent collaboration between artist Patrick Waterhouse and the Warlukurlangu art centre seeks to undo this erasure. Rather than creating new works in isolation, Waterhouse has worked with Warlpiri artists to intervene directly in the colonial archive. They take the very maps and documents that were used to claim the land and they paint over them. They layer Indigenous visual languages, stories, and knowledge directly onto the colonial cartography. It is a literal and symbolic act of reclamation. The maps no longer represent 'empty' space; they are filled with the presence of the people who were there all along.

To paint over a map is to refuse the lie of the empty land.

This process is not merely about destruction; it is about intersection. The resulting works show a tension between two different ways of seeing the world. One is the colonial view: linear, mathematical, and focused on ownership and boundaries. The other is the Indigenous view: cyclical, spiritual, and focused on connection and story. When these two perspectives meet on the canvas, the 'void' that the settlers claimed to find is revealed to be a lie. The land is not empty; it is crowded with meaning.

The impact of archival intervention

Challenges the legitimacy of colonial documents
Reintroduces lost histories into the visual record
Creates a dialogue between conflicting worldviews

This work reminds us that history is not just what happened; it is how we choose to record it. If the records are biased, the history will be broken. By using art to puncture the official narrative, these artists are doing more than making aesthetic objects; they are performing an act of historical justice. They are ensuring that the stories of the land can no longer be ignored by the maps that attempt to contain them.

Key Takeaway

History is written by the victors, but it can be reclaimed by those who refuse to let their stories be erased.