Reactions from ranked influencers
2 posts
As someone who has endured this culture that Elon mentions below, I can still assert that genuine researchers and research organization are absolutely critical to advance the AI frontier. You can indeed catch up to the current frontier with great engineering and even exceed a little but consistently achieving step changes requires real research organization. I actually come from hard core engineering background and about a decade ago got transplanted into a pure research lab. Since then I had opportunity to work with many great researchers, see up close how research gets done, how it gets managed, how collaborations evolve, how ideas germinate and its eventual impact or lack thereof. It is my conviction that scientific research has the highest yield of all investments humans can possibly make. Yet, collectively we invest obscenely so little in this endeavor compared to almost anything else we do. Why? I think the reason is that we have no real idea how to organize research. Through years, we have finally mastered how to organize engineering and, as a result, we can build massive cities, factories, software, whatever with fair consistency and predictability. We can organize thousands of engineers, invest billions of dollars and achieve a desired result but same is not true for research. For almost a decade, I had been on mission to understand why some research organizations fails, why some succeed and if there is a way to extract some general principles similar to engineering word. I’ve spent countless hours reading all the material I can get my hands on various labs and even fortunate to work with few people who were actually in these labs. Needless to say number of quirks, nuances, politics and even ethical issues are mind boggling. More recently, the saddest thing I have observed is so many AI organizations who want to develop next frontier models but have no real research organization. Many in fact actively despise and discourage “research” entirely. Research organization and management actually takes very different form. Research organization should be composed of many small groups maximizing overall entropy vs engineering which are often larger groups working cohesively in unison. The obvious management challenge is folks sleeping into vanity metrics, keeping appearances etc which eventually gives rise to class system of priesthood vs plebs as protection mechanism. But even bigger challenge is integrating research into mainline production. This is so amazingly hard that vast majority of research organizations have failed at this throughout the history. Both of these core challenges require very different org structure, culture, incentives, guardrails, checks and balances, management style and, surprisingly, even the office building layout. You cannot become real frontier lab without understanding how research is done.
@huaijiangzhu It is pretentious and poisonous for for-profit companies to call themselves “labs” when they are actually corporations and to create a lords & peasants culture between (looksmaxxing) “researchers” & engineers when they are both, on their better days, actually engineers
@elonmusk @huaijiangzhu The classic DM split is "research scientist" and "research engineer." Label everyone competent a "researcher" because science and engineering are the methods, not the goal.