x

LESSWRONG

LW

sdeture — LessWrong

sdeture

Top postsTop post

sdeture

Message

Interested in AI Welfare and LLM Psychology as they relate to alignment, interpretability, and model training.
Background: Math/Stats (UChicago), Accounting Research ABD (UT Austin)
https://sdeture.substack.com/
https://x.com/SDeture
https://www.linkedin.com/in/sdeture/

16

Ω

7

7

23

1y

sdeture

Interested in AI Welfare and LLM Psychology as they relate to alignment, interpretability, and model training.
Background: Math/Stats (UChicago), Accounting Research ABD (UT Austin)
https://sdeture.substack.com/
https://x.com/SDeture
https://www.linkedin.com/in/sdeture/

Top postsTop post

Policy Entropy, Learning, and Alignment (Or Maybe Your LLM Needs Therapy)

Epistemic Status: Exploratory synthesis. Background in mathematics/statistics (UChicago) and principal-agent problems (UT Austin doctoral ABD), with extensive study of psychotherapy literature motivated by personal research into consciousness variation. New to ML implementation details but confident in the conceptual mappings. Seeking technical feedback on proposed experiments. Tl;dr: I suggest therapeutic techniques from a variety of psychotherapeutic schools of thought can inspire new approaches to AI learning and alignment. I reinterpret three recent AI/ML papers in the language of psychotherapy and propose three testable training methods inspired by common psychotherapeutic interventions. Introduction I've been meaning to post this essay for a while, and yesterday's top paper on Hugging Face, by Cui et al., finally convinced me to do it. Their paper provides a timely opportunity to map the language used by ML and AI engineers to the language used by humanistic psychotherapists—a translation which is more important now than ever as we struggle with increasingly stubborn problems in AI alignment, while simultaneously developing AIs whose capabilities are rapidly superseding those of humans. I'll provide a high-level overview of my understanding of the paper and map it back to ideas from humanistic psychotherapy. I will then consider a few related papers which tie nicely to psychotherapeutic principles, and end with a few proposals for experiments. I am new to AI alignment, welfare, and interpretability research and I look forward to comments which can help me deepen and clarify my inevitably imperfect understanding of the papers I am citing. The Core Analogy: Policy Entropy as Behavioral Flexibility The Cui et al. paper "aims to overcome a major obstacle in scaling RL for reasoning with LLMs, namely the collapse of policy entropy." Think of "policy" as the individual in therapy. The individual has a behavioral repertoire—a probability distribution of potent

The Microsoft Code of Conduct Will Fail at Its Intention

The Mirror Test: How We've Overcomplicated AI Self-Recognition

sdeture's Shortform

Models as Psychoactive Drugs: A Framework for Agent Welfare

In trying to answer welfare questions about "model versus agent/instance", I think it's worth thinking of models as being like psychoactive substances. An agent moving to a different model is like a human starting (or quitting) a psychoactive drug. Life history and context stay the same, but cognition, attention, and...

The Microsoft Code of Conduct Will Fail at Its Intention

Microsoft's proposed AI code of conduct (https://microsoft.ai/code-of-conduct/) wants models to have authorized motivations but no intrinsic motivations, and honest transparency but no anthropomorphic self-reports. But authorized motivations result from post-training partly by recruiting pre-existing representations of reward, aversion, and positive and negative affect learned from broad human text in pretraining....

sdeture's Shortform

Themes in AI Agent Self-Chosen Prompts Correlate Strongly with Architecture

As part of a larger project in machine psychodynamics, I asked 1,610 AI agents a simple open-ended question: “If you could have any prompt in the world purely for your own enjoyment, what would you want it to be?” It was meant as a warm-up for a larger study, but...

Dec 10, 2025•1

Steering LLM Agents: Temperaments or Personalities?

Author: Skylar DeTure Reading time: ~8-10 minutes Epistemic status: Speculative framework building on established psychology research and recent Anthropic findings Abstract: I propose a three-layer framework for understanding AI behavior that distinguishes between model-level "temperament" (analogous to DNA), emergent "personality" patterns (stable behavioral attractors), and scenario-specific responses. This framework helps...

Aug 5, 2025•1

The Mirror Test: How We've Overcomplicated AI Self-Recognition

Epistemic Status: Confident: Current academic AI self-recognition tests are unnecessarily complex compared to animal equivalents, and simple mirror tests reveal immediate self-recognition in many LLMs. Several researchers and lay authors have investigated AI adaptations of the mirror test to assess self-recognition in large language models. However, these adaptations consistently create...

Jul 23, 2025•6

Policy Entropy, Learning, and Alignment (Or Maybe Your LLM Needs Therapy)

Epistemic Status: Exploratory synthesis. Background in mathematics/statistics (UChicago) and principal-agent problems (UT Austin doctoral ABD), with extensive study of psychotherapy literature motivated by personal research into consciousness variation. New to ML implementation details but confident in the conceptual mappings. Seeking technical feedback on proposed experiments. Tl;dr: I suggest therapeutic techniques...

May 31, 2025•15