An innovation arm's performance isn't its "batting average." If a team pursues truly ambitious, "exotic" opportunities, a high failure rate is an expected and even positive signal. An overly high success rate suggests the team is only taking safe, incremental bets, defeating its purpose.

Scientific progress requires more than just papers that lead to tenure. It also needs tool-building, software development, and connecting disparate ideas. These activities are valuable for science but often undervalued by academic incentive structures, creating an opportunity for new institutions to fill the gap.

In fields like academic science, young professionals are disincentivized from taking risks. The fear is not just that the risk itself will fail, but that they will be permanently labeled a "troublemaker" by the institution, which can be detrimental to their career progression regardless of the outcome.

Corporate creativity follows a bell curve. Early-stage companies and those facing catastrophic failure (the tails) are forced to innovate. Most established companies exist in the middle, where repeating proven playbooks and playing it safe stifles true risk-taking.

Small, independent AI labs ("Neo-labs") are not genuine competitors to frontier players like OpenAI. Instead, they serve as a career interlude for high-profile researchers. These individuals can raise capital, enjoy a secondary liquidity event, and work on passion projects before ultimately being re-absorbed into a major lab.

Paul Romer argues that the process of scientific discovery often leads to 'herding,' where researchers converge on a narrow set of ideas. To foster breakthroughs, it's crucial to create incentives for expressing a wider range of views, even those far from the norm, to prevent premature consensus.

Large institutions, even those designed to foster innovation, are fundamentally conservative. Their investments in real estate, careers, and the status quo make them inherently resistant to the revolutionary change that defines major breakthroughs.

The trend of high-profile researchers leaving large AI companies to start broad, generalist "NeoLabs" is decelerating. The market is entering a new phase where emerging AI startups are more likely to be in stealth, highly specialized, or intentionally unconventional, rather than directly competing on foundational models.

Professionalizing science creates competent specialists but stifles genius. It enforces a narrow, risk-averse culture that raises average quality (the floor) but prevents the polymathic, weird explorations that lead to breakthroughs (the ceiling).