Edition 7

NVIDIA, Microsoft, Emergence, and others signed an industry-wide letter arguing that open-weight models that anyone can download, inspect, modify, and run on their own infrastructure belong at the center of American AI leadership.

Jensen Huang made his first tweet ever to support the argument, framing further that the world needs both frontier closed models and frontier open models.

The letter's public case is that open weights expand access to the AI economy by reserving frontier-scale capability for genuine frontier problems and running efficient, specialized models everywhere else; they strengthen competition across chips, applications, and services; and they let organizations own the value they create, the self-improving models and accumulated knowledge, rather than lose it to a single provider.

Relying solely on closed models is also not inherently safe, despite the common assumption otherwise. Openness lets a broad community examine behavior, find vulnerabilities, and build safeguards, "just as open-source software demonstrated that transparency can be more secure than obscurity."

A 2018 AQR paper by Cliff Asness and Aaron Brown recently resurfaced that uses a study on pulling the goalie in hockey to show how investors can obsess over the wrong measure of risk and why conventional failure often occurs because people are unwilling to risk unconventional success.

In pro hockey, pulling the goalie for a sixth attacker nearly quadruples the opponent's chance of scoring while less than doubling your own, so in expected goals it is a losing move.

Goals, however, are the wrong criterion. "What matters is expected number of standings points. A team down a goal with short time remaining gains a lot by scoring, and loses little if the other team scores [since a loss by one goal or two is the same]."

Coaches pull far too late. Down one goal, the model says pull at 6:10 remaining, not the usual final minute; down two, pull with more than 13 minutes left. A team that pulls optimally gains about 0.05 points per game, roughly four points over an 82-game season, a difference that would have swung six teams' playoff spots.

The reason coaches wait is that they are rewarded for being perceived as good, not strictly for winning, and sins of commission are far more visible than sins of omission. Essentially, winning ugly is undervalued versus losing elegantly. "Worldly wisdom teaches that it is better for reputation to fail conventionally than to succeed unconventionally."

The investing translation is to focus on 1) the risk an investment adds to the whole portfolio rather than its risk in isolation, and 2) on the sufficient extra excess expected return, without the probability of an unacceptable return over your horizon, rather than the short-term volatility.

E.g., cheap stocks tend to beat expensive ones even though owning the ugly ones looks rash, and a CIO who fails conventionally can keep his reputation for competence. At the same time, one who succeeds unconventionally only gets called lucky.

The Athletic clocked Messi at 62.7% of his playing time walking, 24.7% standing still, and just 0.1% sprinting. "Behind that idleness lies a ruthless efficiency," the stillness giving him the calm to read the chaos and find the vulnerability to attack.

Just as with soccer, in business people often feel the need to be sprinting or confuse others sprinting for progress. For instance, the noise of small-business investing: constant self-promotion, "deals, deals, deals," pressure to raise more capital, and a nagging implication that you are never going fast enough.

The reminder from the pitch is that "movement and progress are not the same thing. The best often walk. They observe. They process. They conserve their energy until they recognize an opening." When the opening comes, they move decisively.

Valiante's central idea is that underperformance is baked into human biology. The brain is programmed to seek comfort and safety to keep us alive, not to overperform, so reaching your potential takes a deliberate choice to override the body's built-in governor. Even ultramarathoners almost never push themselves to real harm.

What separates the great is an "inversion," where the work most people avoid becomes the work they love. His examples: elite swimmers who come to love the flip turns and entries everyone else skips, and Tiger Woods grinding three-foot putts until he'd burned a yellow line into the practice green (in roughly 2,000 putts inside three feet over the next two years, he missed just four).

The practical move is to "find out the comfortable habit that's holding you back," whether your phone, your food, or an excuse you tell yourself, change that one thing, and hold yourself accountable to it on a repeated basis. "We don't think our way into a pattern of living; we live our way into a pattern of thought."

Further, he recommends weekly periods for detaching, setting time for "just-challenging" tasks to trigger flow states, and keeping in mind that environments beat goals. I.e., surroundings shape behavior more than willpower, so humans need to build systems that make better habits automatic: "we shrink to our systems, not our goals."