Large modern office building
What if you could run in the supply room, and Xerox a whole firm? How would this look like?
Well, it can be expensive. But maybe not as expensive as people.
Dwarkesh Patel gives us an idea in a new collaborative essay on January 31st talking about the potential for Allah companies.
Suggesting that “everyone is sleeping in the collective advantages he will have” and “seriously underestimating how different the world will look”, Patel raises some theories of how this would work.
The power of replication changes everything
“Currently, firms are extremely close in employment and training talent,” Patel writes. “But if your talent is one he, you can copy it a thick number of times. What if Google had a million software engineers he? … this ability to turn the capital into calculation and calculate in equivalent of your talent High is a fundamental transformation. Each field, decades of business case studies, intimate knowledge of any system and codes on which the company is based. “
Continuing this premise, the author continues to promote high functionality forecasts, for example, that human teams can be repeated for projects that he will copy in ways that will transform management and work – and that there will be a lot of cooperation narrow among these non -human players who will surprise us. The prediction of “without bad communication, never again”, Patel focuses, throughout the idea that the transitions smoothly will unlock more benefits than we think.
The transition of knowledge in the post-human era
Here is a central component of Patel essay that speaks to it – the author opposes the process of replicating knowledge or broadcasting in human firms, and that of the theoretical corporations:
“The great advantage of mankind has been social learning – our ability to pass knowledge between generations and build it,” he writes. “But human social teaching has a terrible handicap: the biological brain does not allow information to be copied. So you have to spend years (and in many cases decades) teaching people what they need to know to do their job. Look at how the high -rankings in the field are aging and older, perhaps because it takes more time to reach the limit of accumulated knowledge. Talent in cities and high firms produces such great benefits, simply because it enables a little better flow of knowledge among smart people. “
Assuming that innovation grows with the highest population size, Patel welcomes us to consider how the armies of agents it will work hard in Lockstep.
Under the hood: How it works
So how does he realize this kind of internal mind that will allow cooperation, not just the data, fly at the speed of light all over the world?
Patel mentions the practice of speculative decoding, so I looked at it up.
Chatgpt determines it with skill:
“Speculative decoding is a technique to speed up text generation from large language models (LLM) using a smaller “auxiliary” pattern to propose multiple signs immediately, which the largest model then controls or “verifies”. In simpler terms, it is a way to reduce the number of calls to an expensive (large) pattern, without significantly changing the quality or distribution of the text generated. “
You can also call phrases like ensemble learning, or distillation, but ultimately, it comes down to the same thing: robots working together, or cognitive, physically, or both.
What is worth? And a monolithic view of companies
Elsewhere, the essay depends on a Randian territory, the winner-all.
Patel suggests that, given the changes, only one role will still be valuable for companies. That is right: is CEO.
“So what becomes expensive in this world? Roles that justify the massive amounts of test time calculation, “he writes. Just consider what this buys you: millions of subjective hours of strategic planning, Monte Carlo’s simulations of various five-year trajectories, deep analysis of each Line code and technical system, and the exhaustive planning of the scenario.
Imagine Mega-Sundar thinking: “How would FTC answer if we were to win eBay to challenge Amazon? Let me simulate the next three years of market dynamics … ah, I see the result. I have five minutes Data time- Let me evaluate 1,000 alternative strategies.
Later, Patel quotes Gwern Branwen talking about what companies could do to spread themselves:
“Why don’t we see that extraordinary corporations cloning themselves and take on all market segments? Why do corporations not evolve such that all corporations or businesses are now the hyper-deleted descendants of a single corporation 50 years ago, of all other corporations that have disappeared in bankruptcy or have been gained? Why is it so difficult for corporations to keep their “culture their lean youthful efficiency, or if the avoidance of “aging” is impossible, why [not] Copy yourself or reproduce differently to create new corporations like themselves? Corporations are certainly subject to selection for fitness types, and vary greatly. The problem seems to be that corporations cannot repeat themselves … Corporations are made by people, non -exchangeable, easily copied widget or strands of DNA. “
In the same way, Patel quotes von Neumann: “All the lasting processes we will anticipate. All the unstable processes we will control. “And then you do this set of questions:
So then the question is: if you can create (the agents of it) for any task you need, why would you ever pay any note for another firm, when you can simply repeat them inside instead? Why would there be other firms? Will the first firm understand that he can understand how to automate everything will form a conglomerate that takes over the whole economy? “
Companies, he writes, exists to reduce transaction costs. Accurate and detail -oriented success is based on a process of arguing in a function of external loss. These are thin moons, however, for human importance. The reality is that we all have to take a close look at what will happen to the business.