For this month's discussion meetup, which so happens to fall on Valentine's Day, we will discuss a very romantic topic: transaction costs! The main reading will be Ronald Coase's classic article "The Nature of the Firm" which asks why people organize into firms and hire employees rather than just hiring contractors. Coase's insight is that contracting involves a lot of transaction costs — not just literal dollars, but negotiation, contract enforcement, difficulty communicating, etc.
The suggested reading includes the original Coase paper as well as some podcast discussions and a short video summarizing the key ideas.
As always, feel free to come if you haven't done the reading.
For this month's discussion meetup, we will be continuing in the moral philosophy vein. We will discuss equity and equality, particularly through the lens of Rawls's "Theory of Justice."
As always, feel free to come if you haven't done the reading.
This is a write up of "Multiple Streams of Knowledge Retrieval: Enriching and Recalling in Transformers", work with David Reber, Sean Richardson & Ari Holtzman. Code is available here. This is cross-posted from https://toddnief.com/articles/dynamic-weight-grafting/
When a new pope is elected and we want an LLM to answer “Who’s the pope?” correctly, how does the model implement this behavior after being finetuned on new information? When we add new relationship information to an LLM, is that information added just to the entity, is it "enriched" at the entity token position in lower and middle layers, or is it recalled in response to the entity in higher layers closer to next token prediction?
To answer these... (read 1438 more words →)
For this month's discussion meetup, we will be discussing Sam Harris's moral landscape. Broadly, Sam attempts to give clear definitions of right and wrong in terms of well-being of conscious creatures and to do away with moral relativism. The primary reading / listening will be a recent episode of the Making Sense podcast where Sam clarifies his argument from his book The Moral Landscape.
As always, feel free to come if you haven't done the reading.
For this month's discussion meetup, we will be discussing "Notes on the Inexact Sciences" from the pseudonymous Suspended Reason. Here's the abstract:
Popular wisdom warns us against premature optimization. And yet, in a quest for public legitimacy and tidy problem domains, many fields discourage vitally necessary descriptive and conceptual work in favor of statistical analysis and laboratory experiments. Topics of unprecedented complexity are tackled using rote, mechanical approaches, by researchers who routinely fail to realize how much linguistic and conceptual clarification is a precondition of headway. Meanwhile, sociological and professional incentives prevent the sorts of synthetic work that might de-provincialize researchers' theories, and initiate exactly those conceptual refactorings which would advance the discipline.
We've... (read more)
For this month's discussion meetup, we will be diving into another topical, geopolitical topic: Taiwan. Please come even if you haven't done the readings, you’re worried it will be awkward, or you aren’t the “typical person who comes to a Rationality meetup.”
For this month's discussion meetup, we will be discussing the autismy-schizotypy continuum. There's a handwavey idea that autism and schizophrenia are in some sense opposites—some recent-ish work attempts to formalize this as an overweighting of sensory input as an autistic trait (leading to extreme systemizing behavior and difficulty reading social cues) while schizotypy gives low weight to sensory input (leading to extreme theory of mind, hyperstitious fantasies, etc.)
Please feel free to come even if you haven't done the readings, you’re worried it will be awkward, or you aren’t the “typical person who comes to a Rationality meetup!”
For our May meetup, we will discuss AI 2027, an attempt to layout month-by-month pathways to AGI within the next few years — and the potential consequences.
AI discussions within the Rationality community are often polarizing and tend to reflect fundamental differences in intuitions. AI 2027 presents concrete scenarios with assumptions, so this is a grounded way to discuss what our next few years might look like.
As usual, please feel free to come regardless of how much of the reading you do ahead of time.
For our next discussion meetup, we will wade into tax policy. Here are some suggested readings and podcasts. As always, please feel free to come to the meetup even if you’ve done none / some of the homework.
For our February meetup, we will be focusing on a new report from Jennifer Pahlka and Andrew Greenway: "The How We Need Now: A Capacity Agenda for 2025 and Beyond"
With discussions of, uh, government efficiency in the news, everyone is popping up with a take on why their pet idea is the right thing to focus on. Jennifer Pahlka founded Code for America and wrote Recoding America. Her diagnosis of government failure tends to highlight the inadequate equilibria and misaligned incentives common in politics, bureaucracy, and large institutions.
