Category intelligence

Research Briefing — March 1, 2026

9 current items analyzed and ranked.

Executive synthesis

Research Summary

An unusually thin day for research output. The sole standout is Andrew Critch's (BERI/CHAI) conceptual framework of Schelling goodness — proposing shared morality as a coordination equilibrium among diverse intelligent agents, including AI systems, with direct relevance to multi-agent alignment.

A conceptual model frames LLM behavior as navigation through semantic topology with attractor basins, offering intuition for prompt engineering and failure modes
The "AI slop as vegan hamburger" analogy provides a useful model for why AI-generated content fails despite surface-level pattern-matching to human output
Remaining items cover epistemics, rhetorical analysis, and non-AI topics (lithium/Alzheimer's, meditation) with minimal research substance

Overall, today's pool lacks empirical papers, benchmarks, or technical contributions; only the top three items offer frameworks with any bearing on AI research or practice.

Key Themes

AI Safety and Alignment · 1Language Models · 2Epistemics and Rationality · 3Non-AI Topics · 3

Primary evidence

Top Ranked Signals

Research LessWrong Feb 27

Schelling Goodness, and Shared Morality as a Goal

By Andrew_Critch

48 score

AI Analysis

Andrew Critch introduces 'Schelling goodness' — a framework for thinking about moral coordination among diverse intelligent agents who share no history, where agents try to converge on moral verdicts using only common knowledge and civilizational survival pressures. This connects game theory, moral philosophy, and multi-agent coordination.

Also available in markdown at theMultiplicity.ai/blog/schelling-goodness. This post explores a notion I'll call Schelling goodness. Claims of Schelling goodness are not first-order moral verdicts like "X is good" or "X is bad." They are claims about a class of hypothetical coordination games in the sense of Thomas Schelling, where the task being coordinated on is a moral verdict. In each such game, participants aim to give the same response regarding a moral question, by reasoning about what a v

AI SafetyAlignmentMulti-Agent CoordinationMoral Philosophy

Research LessWrong Feb 27

The Topology of LLM Behavior

By Quentin FEUILLADE--MONTIXI

35 score

AI Analysis

Presents an intuitive mental model for understanding LLM behavior as navigation through a semantic space with attractors (helpful responses, language matching) and repellers (refusals, safety boundaries), using topological metaphors to explain prompt engineering phenomena like jailbreaks and mode-switching.

I have this mental image that keeps coming back when I do prompt engineering. It's not a formalism, it's more like... the picture I see in my head when I'm working with these systems. I think it's useful, and maybe some of you will find it useful too.The spaceWhen you're having a conversation with an LLM, there's a state: everything that's been said so far. I think of this as a point in some kind of semantic space.Each time the model generates a token, it moves. It computes a probability distrib

Language ModelsPrompt EngineeringInterpretability

Research LessWrong Feb 28

AI slop is a vegan hamburger

By pku

22 score

AI Analysis

Uses the analogy of vegan meat substitutes to argue that AI-generated content ('slop') pattern-matches to real content on superficial inspection but fails on deeper engagement, potentially relating this to Kolmogorov complexity differences between genuine and generated content.

Excerpt below, but read the (not much longer) full thing for the part involving Kolmogorov complexity. I suspect reading the full thing is better than reading the excerpt first for most people, in expectation. It's not that much longer.As a man who has lived in both Israel and Northern California, I have been a member of more than a few majority-vegan social circles over the years. Among the Vegans, the traditional form of dining is to eat something that’s almost, but not quite, entirely like re

AI Content QualityLanguage Models

Research LessWrong Feb 28

"Fibbers’ forecasts are worthless"

By Random Developer

15 score

AI Analysis

Argues that when evaluating practical proposals (not abstract arguments), the credibility and track record of the proposer matters significantly. Draws on Dan Davies' 'One Minute MBA' framework to suggest that known fibbers' forecasts and proposals should be heavily discounted, even in communities that prize idea-over-identity evaluation.

One of the very admirable things about the LessWrong community is their willingness to take arguments very seriously, regardless of who put that argument forward. In many circumstances, this is an excellent discipline! But if you're acting as a manager (or a voter), you often need to consider not just arguments, but also practical proposals made by specific agents: Should X be allowed to pursue project Y? Should I make decisions based on X claiming Z, when I cannot verify Z myself? One key diffe

EpistemicsDecision Making

Research LessWrong Feb 27

Linkpost: "Lithium Prevents Alzheimer’s—Here’s How to Use It"

By Jackson Wagner

12 score

AI Analysis

Linkpost summarizing evidence that very low-dose lithium supplementation (hundreds to thousands of times below psychiatric doses) may reduce Alzheimer's risk by 20-50%, supported by converging evidence from mouse models, observational studies, psychiatric patient data, and mechanistic research.

In this post from a promising-but-niche substack I came across, author Jon Brudvig lays out a compelling case for the high expected-value of taking low-dose lithium supplements for Alzheimer's' prevention. Recent studies of many different kinds -- mouse models, observational studies of areas with high lithium in drinking water, studies of psychiatric patients, and recent preclinical work identifying plausible mechanisms[1] -- all suggest that dramatic reductions in Alzheimer's rates (

Health ScienceNeuroscience

Research LessWrong Feb 28

Burying a Changeling into Foundation of Tower of Knowledge

By siarshai

10 score

AI Analysis

Describes a rhetorical manipulation technique where a speaker substitutes a secondary aspect of a concept for the whole, builds layers of theory on the altered definition ('buffer overflow'), then uses the conclusions to steer audiences while retaining the emotional weight of the original concept.

Rhetorical Attack by Substitution and Buffer OverflowRecently, I've seen following rhetorical technique used in several places:The speaker takes a secondary aspect of some concept and presents it as if the entire concept boils down to that aspect. Or simply provides their own definition of the concept - one that seems acceptable if people don't think too deeply. (substitution)They quickly build several layers of theory on top of that definition. (the buffer overflow)After that, they use conclusi

EpistemicsRhetoric

Research LessWrong Feb 27

Mindscapes and Mind Palaces

By Moon Lesbian

8 score

AI Analysis

Explores the concept of 'mindscapes' — how people visualize the organization of ideas, memories, and emotions in their minds — through informal surveys. Finds most people's mindscapes are disorganized, physical, and primarily visual, and raises concerns about compartmentalization.

Mindscapes are a concept that I have found really interesting lately. My own definition of a Mindscape is "The visualization of the way in which ideas/concepts, memories, and emotions are stored within the mind, and the visualization of their retrieval and interactions". It's similar to the idea of a Mind Palace, but those are more associated with memories than they are emotions, and when prompted to describe their Mind Palace, people seem to be more likely to talk about a library or some actual

Cognitive ScienceIntrospection

Research LessWrong Feb 27

Jhana 0

By 142857

5 score

AI Analysis

Personal reflection on spending a year practicing jhana meditation without achieving any of the eight jhana states. Discusses the technique, the challenges of generating positive feelings as prerequisites, and reflections on the practice.

Happiness is a prerequisite to the jhanas. -- Rob BurbeaThe jhanas are a series of eight, discrete states of experience that are described as extremely happy, pleasurable, and calm. They are accessible through specific meditation practices and are non-addictive. You may have heard of them from Buddhism or Scott Alexander.I have spent a year practicing jhana meditation and have experienced zero of them. Here are some of my reflections.Generate a positive feeling"Good luck at school, Harry. Do you

MeditationPersonal Development

Research LessWrong Feb 28

How can rationalists perform exceptionally well?

By pantalaimon

3 score

AI Analysis

A very brief question post asking what mental tools rationalists could use to perform exceptionally well at novel tasks, using beating a difficult video game on a first try as an example.

For example, beating Baldi's Basics on the first try, without any prior knowledge of the game. It's a game widely known to be very difficult. What sort of mental tools could they use to do that?

Rationality