Conference Presentation, Fireside Chat, Interview
Max Hodak: Average Is Not Good Enough
- Science plans to dedicate the next 30 minutes to discussing infrastructure and lessons rather than detailing the retinal prosthesis product further.
- Startups involving hardware or non-pure software are expected to continuously purchase thousands of distinct items, where the cost of delaying work for highly paid employees far exceeds potential savings from waiting to buy through auctions.
- Transitioning from academia to startups presents a challenge where setting up vendor accounts and handling insurance and certification paperwork can extend the order process over two weeks.
- Without proper attribution systems, experiments consuming expensive materials may be perceived as "free," complicating the calculation of true pricing and runway.
- A $20 million raise with roughly half allocated to headcount and real estate and $3 million annually to a research budget is expected to be exhausted in three or four years, often faster than teams anticipate.
- Effective hiring is predicted to originate from within the specific "scene" or network that produced the company before expanding to the general public.
- Hiring processes lacking a defined, strictly followed system are expected to become a significant drag on team time.
- A company-wide voting mechanism for initial candidate reviews is expected to be completed within 24 to 48 hours to prevent bottlenecks.
- Without an initial voting stage, the top of the hiring funnel is predicted to become overwhelming as companies grow beyond a couple of years.
- If the conversion rate from the final interview stage to an offer falls below 25%, the company is expected to waste excessive time conducting on-site interviews for non-converting candidates.
- The "eigenreviews" system, which distributes judgment across the company every four to six weeks, is expected to provide better insights into company dynamics than traditional annual performance reviews.
- A severe difference in iteration speed, such as learning one thing a week versus one thing a month, is predicted to render the faster team irrelevant to the slower competitor.
- Deep tech companies are expected to be more likely to fail due to an inability to organize human resources and manage physical infrastructure than due to technology failure.
- CEOs are predicted to reach a point four or five years into the company where high-stakes decisions must be made alone without advice.
- When a company gets stuck, the expected solution is to inject action to produce information and entropy, often by removing strong individuals who are a poor fit for the current phase.
- In fields where the best talent is no longer in academia, industry is expected to move much faster due to resource marshaling.
- Organizational skills and judgment are expected to be learned through the "oral tradition" of working at a high-performing company rather than through a PhD.
- Exceptional ability is expected to be best evidenced by winning "legible competitive games" such as chess or Formula SAE, which provide clear feedback.
- LeetCode is not believed to be a primary evaluation tool for software hiring, where the field attracts smart people due to rapid feedback loops.
- Neuralink provided a lesson in tuning judgment filters by working with someone who had empirically excellent judgment to provide immediate feedback on high-stakes decisions.
- The end goal of BCI technology is predicted to be the creation of "conscious machines" by solving the practical problem of mapping substrate activity to phenomenal content.
- Proving consciousness experiments in humans is expected to remain a practical problem several years away.
- For interdisciplinary problems, a smaller team capable of holding the whole problem in their heads is expected to be more effective than large interdisciplinary centers.
- AI is expected to act as a multiplier for teams rather than a replacement, particularly in coding and regulatory compliance.
- AI is predicted to transform the regulatory compliance process, which historically takes months, by quickly looking up standards and generating evidence tables.
- Building internal software platforms is expected to be more powerful than buying off-the-shelf ERPs once a company grows large enough to define its unique processes.
- In the near term, BCI adoption is expected to be primarily a "longevity story" and a form of healthcare providing large effect sizes for restoring function, rather than just an AI story.
- The next generation of retinal implants is predicted to require significant advances in material science for "packaging" and conformal coatings to survive the body's hostile environment.
- The company, Science, is expected to integrate two distinct cultures: a small group of "transhumanist" visionaries and a larger group of serious clinicians who view the visionaries as crazy but are committed to building medical devices.
- Starting a biotech company is expected to be a commitment to a decade of life that will never be recovered, though the impact on patients is unlike any other sector.
- Building the "Helix" internal software platform is expected to be a contrarian move that might be questioned by investors if the company does not control the narrative.