Base optimizer and mesa-optimizer comparison

-

 


Here I will be documenting my understanding of what a base optimizer and a mesa-optimizer are, and putting them in comparison. 

The Github link is here: https://github.com/evhub/mesa-optimization/blob/master/post1.md

- The base optimizer is a gradient descent process to create a model, and the model is designed to accomplish some specific task
- The mesa optimizer produces a base optimizer that is itself good at optimizing systems 
- The mesa-optimizer is different from a subsystem because it is an optimization process, not an agent. 
- Mesa-optimization happens when the base optimizer can find a model that exists for the purpose of optimizing another 
- Within every optimizer, there are objectives
- Unlike the base objective, the mesa-objective is not specified directly by the programmers
- Mesa-optimization sometimes leads to mismatch of base and mesa-objectives. This is called misalignment. 
- We can call a model generated by the base optimizer as a learned algorithm

Now we will clarify with an example: 

Base optimizer is Evolution
Base objective is caring about certain favorable alleles' frequency in the population
Mesa-optimizers are human brains 
Mesa-objective is organisms’ behavior—behavior that is not necessarily aligned with evolution. For example, choosing to not have children and thereby decreasing the exchange of favorable alleles' frequency. 

Comments

Post a Comment

Popular posts from this blog

A year, introspectively summarized

-
Image
               A year, introspectively summarized      A primal and gravitational feeling is the feeling of belonging--- though the whole sense of "purpose" is artificial, defined deliberately by people who want others to emulate them. Emulation stunts creativity, just as the proclivity to belong stunts agency. I am delighted by the platform that I am given to practice my agentic self, now and continuously onward.       There will be a time when t he basis of your understanding, you learn, was fabricated---even some of your strongest beliefs need reinterpretation, the very framework under which you operate proves inflexible and restrictive. Perhaps the best remediation strategy then is desensitization of the action-oriented part of your mind. You must compromise scaling for precision.       In January, I wrangled with partial differential equations and AI Safety fundamentals course with BlueDo...
Read >>

Fiat Lux!

-
Image
True scholars prefer the clandestine accommodations of secluded chateaux to the ostentatious finance and interlocking corporate consortium They indulge in ambrosia-glazed idols in a Bayreuth opera house And vials with oil of brazilwood, distilled Amazon dew, creme de Java  And I could fill the days with flame-kissed roses, powdered saltwater pearls, essence of lilac---  half-heartedly allude to commentaries that blend in with volumes of Penrose's crowds Revert disillusioned nebulous eyes mine into an exhilarating rhinestone gleam:  "to let the warm light in!"
Read >>

Manhattan

-
Image
Impinging upon a full year in New York is a whimsical sentiment, I feel these foreign faces have imprinted a familiar shade on me.  Let me tell you, Reader, of a story when I was 11. Cindy, my only close acquaintance in elementary school in Toronto, had haughtily told her mom in English that Lisa hasn't ever read Nancy Drew nor Geronimo Stilton books. That, in comparison, she has finished both series and enjoyed them.  She knew that I couldn't understand English well enough to answer her articulately in English, much less to read those books. It was my 4th month in the North American continent.  Perhaps it was her effort to impress her mom, or a childish tease at an immigrant's poor English--- she succeeded. Her mom said that her Cindy is the smartest, and some children can't read yet. Cindy's mom wanted to see me hurt, so she said it loud and clear, in Chinese.   But I was the Head Girl in Beijing! My fragile ego felt tremors. (Interestingly, they never warn yo...
Read >>