Reward Modelling

"Reward modelling" is the term used to describe a process that guides a Large Language Model toward human preferences in generative artificial intelligence. The Deep Seek development team and others at Tsinghua University in Beijing, China are preparing to release version 2 of their economical  GAI in the near future. 

Human preferences? Human preferences are dependent on the combination of particular cultural factors and individual idiosyncrasies, the number of which could easily be infinite. Like every aspect of GAI in whatever form, its development is structured by those doing the developing. No large language model will be able to incorporate something as yet undiscovered. 

An authoritarian regime could easily use directed GAI to explain and justify its policies to its population, just as genuine humans have in the past. It's unlikely that if the human preference is found to be greater personal freedom policies will reflect that. 

The Deep Seek phenomenon is already having repercussions in the US data center growth picture. If smaller scale or more efficient GAI installations become part of the US data center complex, electrical demand will be a fraction of what is currently projected. More Deep Seek publicity will put the brakes on funding needed for what could be called "conventional" data center development. It will also be necessary for reliable customers to buy in to the concept and make GAI profitable. Nobody knows yet if that will be the case but some are willing to accept the risk rather than being left on the sidelines.    

An MIT Analysis Of The Tariff Fiasco

First of all, in a society that values freedom, there are no tariffs. A free citizen is free to purchase whatever legal item he or she wishes from anyone who cares to sell it to them. Both sides agree on the terms of the transaction without government interference. This was the ultimate message of Adam Smith's seminal book from 1776, The Wealth of Nations, free trade, accepted as gospel even today.

According to an MIT Technology Review article: "  Trump administration cuts to the Department of Energy and other federal programs could also take away money from demonstration projects that help cleantech companies test and scale up their technologies. And if Congress does eliminate certain subsidies in the Inflation Reduction Act, it could halt billion-dollar projects that are being planned or perhaps even some that are already under construction."

Perfect logic. Keep spending taxpayer funds that will be used for planned or commenced projects in spite of their lack of viability.

“The biggest challenge for companies that are making hundred-million- or billion-dollar capital investments is dealing with the uncertainty,” Turner says. “Uncertainty is a real deterrent to making big bets.”

 Yes, as Ben Franklin is credited with saying, "There's nothing certain but death and taxes." Other wise it wouldn't be a bet at all. The Neo-Whigs that wish to get in on the clean tech gravy train want to do it with taxpayer funding. They want a sure thing, not a good bet.

Former Biden Treasury Secretary Janet Yellen said that tariffs on Chinese solar panels and other "clean" products are needed if the move to non-hydrocarbon energy is to proceed. Ergo the existential threat to life on earth is economically dependent on making some products more expensive. If an exploring space ship from another galaxy appeared and gave away solar panels for free it would be a disaster, according to the academic economist.