My ideas

Regarding learning, the brain works like a system of pumps, each "pumping" meaning from a previous representation. For example, a first brain region gets information from the eyes and extracts lines, a second region takes the lines and recognizes letters, a third region takes the letters and recognizes words, and so on.

A pump works in the measure it is able to extract meaningful patterns: patterns which represent what is going on, therefore stable over time, therefore low entropy.

The most important criteria that will determine whether we will take action on a plan of ours is: do we intuitively believe it will have a positive Expected Emotional Outcome?

This means that we are unlikely to act on plans with negative EEO as intuitively predicted regardless on the external outcome we analytically expect.

Positive EEO is based both on "how much taking this action will distance ourselves from existential risks" and on "how much pleasure did this action bring us in the past (i.e., how addicted we are to it or to its cues)". Negative EEO is based on the opposite: how much it will bring us closer to existential risks and how much discomfort did it cause to us.

When faced with information coming from different channels (e.g., oral language & body language), people prioritize the channel which they are able to subjectively interpret in the less noisy way.

This explains why people with Autism Spectrum Disorders are more likely to disregard body language.

The maximum level of proficiency people can achieve at any given skill is limited by the time spent practicing and the amount of feedback gathered in each practice session.

Therefore, it is critical that you not only practice your skill but also practice your practice, tweaking it so that you can learn the most out of it.

A good approximation for Autism Spectrum Disorders is a high-pass filter: a filter that only lets details pass through and blocks other kinds of information such as context.

This idea explains why people with ASD tend to be relatively proficient in fields where context is irrelevant (e.g., mathematics, computer science) and relatively impaired in fields where context is necessary (e.g., social interactions, dancing and sports).

Everyone is upset when they cannot lose weight. Only fit people are upset when they cannot go to the gym.

Personal change only happens when we become addicted to the actions that bring us the results we want.

Only desiring the results only brings us frustration.

Our behavior has repercussion on our future, not only through the external ripples of our actions coalescing in waves that hit (the traditional concept of karma), but most importantly by changing our mind and thus our future behavior.

Take decisions not (only) for their external result, but for the future behavior and reactions of yours they promote.

I suspect that the autism-spectrum is two tailed. In the center, neurotypicals. On one tail, people with autism symptoms. On the other tail, people with tendencies opposed to those of autistic ones (i.e., a preference for context over detail and an impairment at detailed fields).

Our brain is made of multiple computing units, each individually evaluating a fragment of our perception. This leads to "broken phone" effects in which a unit's wrong inference is assumed as correct by other units (leading to illusions) and in which incomplete but coherent information is considered as complete (leading to confabulations).

Our brain contains not one but two decision-makers: the Planner (the cortex, which suggests what we should do) and the Gatekeeper (the basal ganglia, which decides whether to do it). For action to occur, the cortex must suggest it and the basal ganglia must approve it.

Most habit formation advice addresses only the Planner: better plans, fewer bad cues. But the Gatekeeper cannot imagine or plan; it can only remember. What causes the Gatekeeper to open the gate is experiential memory: the action is remembered to have brought positive emotions in the past. Habit formation must speak to both.

Pain is not a signal of damage but of vulnerability: a state where damage may occur in the future. Most times vulnerability is caused by damage, hence the common confusion. But there are cases of damage without pain (soldiers who lose a limb in battle and feel nothing) and pain without damage (touching a hot pot before any burning occurs).

This framework also explains psychosomatic pain: psychological damage, social damage, and lack of resources all predict future physical harm, so the brain responds with pain. It also explains placebos: they work by signaling that the situation is safe and the body can commit resources to healing. Placebos are permissions to heal.

Every time we fail at something we had the resources to succeed at, it is because we were optimizing an internal metric (Expected Emotional Outcome) instead of the external one. We considered it an internal success to fail at the external goal.

More broadly: when an organization optimizes hard for a proxy metric, it can commit itself to external failure. Optimizing for a proxy reduces its correlation with the ultimate goal, because the easiest way to improve an unweighted metric is to include items that contribute to the proxy but not to the underlying objective.

Not all time dedicated to work is work time. Sometimes we are in hobby mode: spending time on work-related activities for personal enjoyment rather than output, polishing a presentation more than the audience will care, getting nerd-sniped into building a tool that saves less time than it took to build.

There is nothing inherently wrong with hobby mode, as long as it is correctly categorized as such and not confused with work time. The key is to be deliberate about it: box it into specific slots in the schedule so the rest of the time can be fully focused.

Chronic resistance (procrastination, going through the motions passively, avoidance of a particular task) should not be treated as an obstacle to power through. It is information to dissect and address.

Resistance is often a signal of a lack of clarity, skill, confidence, or feedback. Each source of friction can be identified and addressed individually. Once specific objections are resolved, the resistance often melts entirely. This applies equally to personal habits and to managing a team's engagement with their work.

Things which are antifragile benefit from volatility (Taleb, Antifragile).

However, how likely something is to exhibit an antifragile response is not a static property. When exposed to relatively moderate stressors, antifragile things become more likely to exhibit antifragility: a process I call antifragilization. Similarly, when not exposed to any stressor of at least moderate entity, the antifragile becomes weaker and less likely to exhibit an antifragile response: a process I call fragilization.

Yes, we can say that some activities are ergodic and others are non-ergodic. But it's also possible to say that one activity is more ergodic than the other one.

That's because people do not care about what happens when t→∞; instead, they care about the medium term. And in the medium term, it's possible to compare how ergodic the outcomes of an activity look.

This is important because it represents more closely the risk management evaluations people make when presented with competing options.

Antifragile is a term coined by Nassim Nicholas Taleb to identify what benefits from variation, usage, problems, and feedback.

I've merged my expertise in management together with my knowledge of antifragility in a course called "Antifragile Organizations". Designed for managers and entrepreneurs, it is practical and problem-oriented.

A visual framework for understanding antifragility. The diagram maps three zones: a yellow zone where stressors are too light to produce adaptation, a green zone where stressors are strong enough to trigger growth, and a red zone where stressors are too strong and cause harm.

The diagrams are useful for understanding how to become more antifragile: by shifting the left threshold (responding to smaller problems) and the right threshold (surviving larger ones). My students named them after me.

Antifragile entities cannot become fragile. They never lose the property of benefiting from stressors in a given range. However, the range of stressors that benefits them can shrink. I call this fragilization: the process by which, in the absence of sufficient stressors, the antifragile becomes progressively less able to exhibit antifragile responses.

Fragilization happens because, without natural selection acting on the lower-layer population, weak members are allowed to reproduce. The average member becomes weaker and the entity's tolerance to stressors decreases. Environments without challenges do not produce resilience. They erode it.

At first sight, chess appears to be a game of pure skill: the board is fully deterministic. Yet chess contains a luck component: in the 2023 Chess World Championship, the winner only won 4 out of 18 games against the finalist.

The reason is imperfect information. If we do not know which tactics our opponent studied, we can only assign probabilities to our moves. And when the best we can do is assign a probability, luck plays a role. More broadly: whenever there is imperfect information, there is a luck element, even in fully deterministic contexts.

A new habit won't be adopted sustainably in an organization unless a local density of at least 80% has been achieved in any given team.

(E.g., at least 80% of the members of a given team practice the habit with consistency.)

A training methodology designed to produce real behavior change: no slides, no pre-packaged content. The training begins by asking the audience what problems they face in a given area. Each problem is then addressed in concrete detail through feedback, roleplay, and hypotheticals.

Most training fails because it circles around the hard questions participants actually need answered. The Dellanna Method is all about surfacing and addressing those questions, acknowledging that teaching not only what the right thing to do is, but how to do it right, is necessary for behavior change.

Most managers communicate clearly enough to be understood, but that is not enough. The standard should be higher: be so clear that you cannot be misunderstood.

Superclarity means communicating with enough concreteness and specificity that misunderstanding is not merely unlikely. It is not possible. This requires asking yourself: if my interlocutor walked away with a different understanding than mine, what might they have missed?

Change initiatives fail when they produce elastic change: change that reverts once pressure is removed, like a business card that springs back. They succeed when they produce plastic change: permanent deformation.

The key is concentrated rather than diluted pressure. Reminding people of a desired change once a week for a year produces no real shift; confronting the same behavior multiple times a day produces change within a week. Any cadence longer than daily enforcement dilutes pressure to the point of producing only elastic change.

Maintenance never makes sense in the short term, yet it is indispensable in the long term. It never feels like a good time to perform maintenance. There is always something more productive to do instead.

The root cause is a deep-rooted belief that if we get the most out of each day, we will also make the most of the year. But some activities (maintenance, training, risk management, relationship-building) have benefits only visible in the long term and are systematically undervalued by short-term evaluations.

A framework for managing underperformance. Rather than a traditional Performance Improvement Plan with vague expectations, the Trajectory Framework uses a simple chart drawn during a one-on-one: a line connecting current performance to the required level in three months. Two-week milestones are derived from this line.

By the second or third check-in, it becomes clear to everyone whether the three-month objective is achievable. The word 'trajectory' shifts the frame from deficit (what is lacking) to direction (where we are heading), which reduces defensiveness and increases alignment.

When delegating, most managers say: 'here is what I want you to do.' Better is to also define what a bad result looks like and what a good result looks like. Best is to add what a great result looks like.

The third tier is not just about raising the bar. It is about painting a vivid picture of what excellent looks like, which shifts the delegate's focus from compliance to ambition. The framework is equally useful for writing job descriptions and for upskilling: showing someone what great looks like dramatically accelerates their development.

Organizational culture is not a set of concepts or a list of values. It is a track record of which behaviors led to good personal outcomes and which did not. What determines whether raising a hand is part of a team's culture is the track record of what happens when someone raises a hand.

Defining Core Values and communicating them does not change culture. Only actions change the track record, and only a changed track record changes culture.

We often pull growth forward: taking the time that should be dedicated to building skills and using it instead to meet short-term objectives. This always backfires, because reaching the next objective will require skills that were never built.

The underlying false belief is that if we get the most out of each day, we also get the most out of the year. Long-term growth requires spending some days doing things that bring no immediate return but are necessary to enable progress beyond a certain point.

Even when delivered politely, most negative feedback is inherently adversarial. It implies the recipient should abandon their way of doing things and adopt the giver's. This triggers defensiveness, even when the suggestion is sound.

The trick is to avoid framing feedback as 'me vs you' and instead frame it as 'me and you vs a common problem.' The less the manager's ego appears in the suggestion, the more the subordinate's ego can accept it.

Good long-term strategies have three properties. They are sustainable: they do not require risks or sacrifices that cannot be maintained long-term. They are constructive: they build the assets (trust, know-how, relationships, capital) that future success requires. And they move toward inevitability: they proactively surface and address problems so that success becomes nearly guaranteed rather than merely possible.

Centralization is only efficient to the center: be it the political center (the politicians), the geographical center (the capital), the financial center (the banks & related institutions), the societal center (the "front-row class") or the professional center (the "front-row professions").

Most of morality maps over not introducing non-ergodicity, which is to say, over not creating or reducing the distance from existential risks, for the self or for the members of the community.

Mimetic societies survive if and only if skin in the game is enforced. If it is, those with ineffective or sociopathic behaviors are forced out of the population, and only pro-survival behaviors are available to imitate.

Instead, in the absence of skin in the game, charlatans are revered and imitated, leading to decadence and incompetence.

One would think that a country that got hit hard during a pandemic's first wave would be hit lightly during further waves, as it built immunity. Instead, we often observe the opposite phenomenon: places that are hit hard in the first wave tend to be hit hard in subsequent ones. That's because of virus infrastructure: the geographic, demographic, cultural, and social factors that make a virus spread more easily in a given area.

To prevent pandemic spread within countries, governments must limit movement fractally: in infected countries, limit travel between states; in infected states, limit travel between cities; in infected cities, limit travel between neighborhoods; in infected neighborhoods, limit travel between blocks.

This approach contains spread at every scale while minimizing the overall cost to the population, since restrictions only apply where the infection is already present.

Polarization is not just political differences but the belief that the opposing group is so dangerous that unethical behavior against them is justified. Adopting extreme measures makes one's own group appear dangerous to the other side, which justifies their extreme measures in return. Therefore, polarization is always mutual.

Because polarization is always mutual, it cannot be resolved through polarizing actions. Censoring a political opponent, exaggerating their dangers, or failing to call out one's own unreasonable allies all deepen polarization by triggering retaliatory responses. Even when one side bears more responsibility, polarization must be addressed as a shared problem.

A thought experiment to test whether the world has gotten worse: observe what would happen if the arrow of time reversed. We would progressively lose technology, social rights, access to services, and the abundance we now take for granted.

This effect would be particularly severe for the bottom layers of society. Rights, plumbing, and internet access all disappeared earlier for the poorest as you go back in time. Reversing the arrow of time would be widely recognized as evil and unjust, which implies that the direction of time, toward the present, must be better.

Functional poolers are an architecture that allows for ML systems to extract meaning from incoming streams of data without external supervision, not even in the training phase.

When asking an AI to perform a task, most people describe the task but not what success looks like. A Success Criteria Loop fixes this: specify the task, define explicit success criteria, and instruct the AI to keep revising until those criteria are met.

This converts a one-shot prompt into an autonomous improvement loop. The AI evaluates its own output against the criteria, identifies what to improve, and iterates, without further input from the user.