Newest Papers
discomfort is the price you pay for a fulfilling life
jacqueline
Substack·2025
Wellness isn’t always about following a perfect routine; sometimes it’s about taking a risk.
★ 4.0 (1)
View paper →Vibe-Weaving 1001: Multithreaded Vibe-Coding
Liminal Warmth ❤️🔥
Substack·2025
Your only limits are your tokens, your attention, and how well you set your robots up
No ratings yet
View paper →"Prompted by Me. Generated by ChatGPT" by David J. Gunkel
David J. Gunkel
STARS·2025
This essay—which is not only about human-machine collaboration but is a performance in human-machine collaboration—interrogates the shifting terrain of authorship and creativity in the age of generative artificial intelligence (GAI). Challenging both the instrumentalist view of technology and the romantic myth of the singular genius, it argues for a reconceptualization of creative production as distributed, dialogical, and co-constituted. Drawing on both theoretical innovations in poststructuralism and the practices of pre- and post-modern content creators, the essay repositions the algorithm not as a mere tool but as an active participant in the generation of meaning. In doing so, it exposes and disrupts—in both content and form—the metaphysical assumptions that continue to underwrite our understanding of writing, agency, and communication.
No ratings yet
View paper →Here’s 18 Applications of Deception Probes
Cleo Nardo, Avi Parrack, jordine
LessWrong·2025
Introduction
I’m excited by deception probes. When I mention this, I’m sometimes asked “Do deception probes work?”
But I think there are many applications of deception probes, and each application will require probes with different properties, i.e. whether a deception probe works will depend on what you’re using it for. Furthermore, whether one deception probe works better than another will also depend on what you’re using them for. This remark sounds a bit trivial, but I didn’t appreciate it initially.
In this document, I'll enumerate 18 different applications of deception probes, and what properties each application requires. This are just my very rough guesses, just to kickstart some public conversation.
Applications of deception probes
1. Monitor current models
Problem Statement: AIs at the current capability level may be important for future safety work, e.g. trusted monitoring. However, current LLMs engage in strategic deception. This is despite being post-trained to be honest, and without any fine-tuning or instructions to be deceptive![1] But we can’t deploy current AIs for trusted monitoring if they will occasionally lie about whether they think the observed activity is suspicious.
How deception probes help: Train deception probes on the weak model (e.g. Claude 4) and use the weak model to monitor the activity of the strong model (e.g. Claude 6). If the weak model claims the activity is not suspicious, then apply the deception probe to that claim. If the pro...
No ratings yet
View paper →On the Theoretical Limitations of Embedding-Based Retrieval
Orion Weller, Michael Boratko, Iftekhar Naim, Jinhyuk Lee
arXiv·2025
Vector embeddings have been tasked with an ever-increasing set of retrieval tasks over the years, with a nascent rise in using them for reasoning, instruction-following, coding, and more. These new benchmarks push embeddings to work for any query and any notion of relevance that could be given. While prior works have pointed out theoretical limitations of vector embeddings, there is a common assumption that these difficulties are exclusively due to unrealistic queries, and those that are not can be overcome with better training data and larger models. In this work, we demonstrate that we may encounter these theoretical limitations in realistic settings with extremely simple queries. We connect known results in learning theory, showing that the number of top-k subsets of documents capable of being returned as the result of some query is limited by the dimension of the embedding. We empirically show that this holds true even if we restrict to k=2, and directly optimize on the test set with free parameterized embeddings. We then create a realistic dataset called LIMIT that stress tests models based on these theoretical results, and observe that even state-of-the-art models fail on this dataset despite the simple nature of the task. Our work shows the limits of embedding models under the existing single vector paradigm and calls for future research to develop methods that can resolve this fundamental limitation.
★ 5.0 (1)
View paper →Exploring the effects of prosocial and self-kindness interventions on mental health outcomes.
Maria Naclerio, Lee Lazar, Erica A Hornstein, Naomi I Eisenberger
Emotion (Washington, D.C.)·2025
A growing body of literature suggests that prosocial behavior, or behavior intended to help others, benefits well-being. However, modern society often places a greater emphasis on self-focused acts in the pursuit of well-being. To understand the effects of these differing forms of kindness (to others or the self), we conducted a 2-week intervention study of a community sample during the COVID-19 pandemic in 2020. Participants were randomly assigned to a prosocial (other-kindness), self-kindness, or control condition, and those in the active conditions were asked to perform three acts of kindness each week. Of those who completed the intervention ( = 777), we found that participants in the other-kindness (vs. control) group experienced significant decreases in depression, anxiety, and loneliness from pre- to postintervention, offering compelling evidence for the mental health benefits of prosocial behavior. Furthermore, we found that participants in the self-kindness (vs. control) group experienced significant decreases in depression, but no differences in anxiety and loneliness. While the self-kindness group reported experiencing more positive feelings during their acts of kindness, the other-kindness group felt more connected. Exploratory mediation analyses revealed that, for the prosocial group, the effect of condition on depression, anxiety, and loneliness was mediated by increases in feelings of social connection, whereas for the self-kindness group, the effect of condition on depression was mediated by increases in positive feelings. Overall, these findings reaffirm the benefits of prosocial behavior on well-being and suggest unique pathways to mental health benefits for these two forms of kindness. (PsycInfo Database Record (c) 2025 APA, all rights reserved).
No ratings yet
View paper →Ionized Gas Outflows and Shock-Heated Emission in the Highly Inclined Active Galaxy CGCG 012-070
Lucas Ramos Vieira, Rogemar A. Riffel, Rogério Riffel, Oli L. Dors, Marina Bianchin, Thaisa Storchi-Bergmann
arXiv·2025
Active Galactic Nuclei (AGNs) exhibit excess mid-infrared H$_2$ emission compared to star-forming galaxies, likely driven by outflows and shocks inferred from integrated spectra. We present optical IFU observations of the central 2\,kpc of the AGN host CGCG 012-070, selected for its pronounced H$_2$ emission excess, to map stellar and gas kinematics. The stellar velocity field is well described by a rotating disc with a line of nodes at $103^\circ \pm 4^\circ$, with the northwest side approaching and the southeast side receding. Gas kinematics, traced by strong emission lines, show two components: a narrow one ($σ\lesssim 200\,{\rm km\,s^{-1}}$) in the disc plane following stellar motions, and a broad ($σ\gtrsim 300\,{\rm km\,s^{-1}}$) associated with outflows within the inner $\sim$1\,kpc. Disc gas emission is mainly driven by AGN photoionization, while the outflow also includes shock-heated gas, as indicated by flux ratio diagnostics. The outflows are radiatively driven, with a mass-outflow rate of $(0.067 \pm 0.026)\,M_{\odot}\,{\rm yr^{-1}}$ and a kinetic coupling efficiency of 0.07%, potentially redistributing gas and contributing to maintenance-mode feedback in CGCG 012-070. Our results provide further evidence that the warm H$_2$ emission excess in nearby AGN is associated with shocks produced by outflows. Observations of other gas phases, such as cold molecular gas, are necessary to gain a more comprehensive understanding of the impact of the outflows on the host galaxy.
No ratings yet
View paper →The Information Dynamics of Generative Diffusion
Luca Ambrogioni
arXiv·2025
Generative diffusion models have emerged as a powerful class of models in machine learning, yet a unified theoretical understanding of their operation is still developing. This paper provides an integrated perspective on generative diffusion by connecting their dynamic, information-theoretic, and thermodynamic properties under a unified mathematical framework. We demonstrate that the rate of conditional entropy production during generation (i.e. the generative bandwidth) is directly governed by the expected divergence of the score function's vector field. This divergence, in turn, is linked to the branching of trajectories and generative bifurcations, which we characterize as symmetry-breaking phase transitions in the energy landscape. This synthesis offers a powerful insight: the process of generation is fundamentally driven by the controlled, noise-induced breaking of (approximate) symmetries, where peaks in information transfer correspond to critical transitions between possible outcomes. The score function acts as a dynamic non-linear filter that regulates the bandwidth of the noise by suppressing fluctuations that are incompatible with the data.
No ratings yet
View paper →Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning
Sikuan Yan, Xiufeng Yang, Zuchao Huang, Ercong Nie, Zifeng Ding, Zonggen Li, Xiaowen Ma, Kristian Kersting, Jeff Z. Pan, Hinrich Schütze, Volker Tresp, Yunpu Ma
arXiv·2025
Large Language Models (LLMs) have demonstrated impressive capabilities across a wide range of NLP tasks, but they remain fundamentally stateless, constrained by limited context windows that hinder long-horizon reasoning. Recent efforts to address this limitation often augment LLMs with an external memory bank, yet most existing pipelines are static and heuristic-driven, lacking a learned mechanism for deciding what to store, update, or retrieve. We present Memory-R1, a reinforcement learning (RL) framework that equips LLMs with the ability to actively manage and utilize external memory through two specialized agents: a Memory Manager that learns structured operations, including ADD, UPDATE, DELETE, and NOOP; and an Answer Agent that pre-selects and reasons over relevant entries. Both agents are fine-tuned with outcome-driven RL (PPO and GRPO), enabling adaptive memory management with minimal supervision. With only 152 training QA pairs, Memory-R1 outperforms strong baselines and generalizes across diverse question types, three benchmarks (LoCoMo, MSC, LongMemEval), and multiple model scales (3B-14B).
No ratings yet
View paper →Caught in the Act: a mechanistic approach to detecting deception
Gerard Boxo, Ryan Socha, Daniel Yoo, Shivam Raval
arXiv·2025
Sophisticated instrumentation for AI systems might have indicators that signal misalignment from human values, not unlike a "check engine" light in cars. One such indicator of misalignment is deceptiveness in generated responses. Future AI instrumentation may have the ability to detect when an LLM generates deceptive responses while reasoning about seemingly plausible but incorrect answers to factual questions. In this work, we demonstrate that linear probes on LLMs internal activations can detect deception in their responses with extremely high accuracy. Our probes reach a maximum of greater than 90% accuracy in distinguishing between deceptive and non-deceptive arguments generated by llama and qwen models ranging from 1.5B to 14B parameters, including their DeepSeek-r1 finetuned variants. We observe that probes on smaller models (1.5B) achieve chance accuracy at detecting deception, while larger models (greater than 7B) reach 70-80%, with their reasoning counterparts exceeding 90%. The layer-wise probe accuracy follows a three-stage pattern across layers: near-random (50%) in early layers, peaking in middle layers, and slightly declining in later layers. Furthermore, using an iterative null space projection approach, we find multitudes of linear directions that encode deception, ranging from 20 in Qwen 3B to nearly 100 in DeepSeek 7B and Qwen 14B models.
No ratings yet
View paper →AI Agents for Economic Research: August 2025 Update to “Generative AI for Economic Research: Use Cases and Implications for Economists,” Published in the Journal of Economic Literature 61(4)
Anton Korinek
aeaweb.org·2025
The objective of this paper is to demystify AI agents—autonomous LLM-based systems that plan, use tools, and execute multi-step research tasks—and to provide hands-on instructions for economists to build their own, even if they do not have programming expertise. As AI has evolved from simple chatbots to reasoning models and now to autonomous agents, the main focus of this paper is to make these powerful tools accessible to all researchers. Through working examples and step-by-step code, it shows how economists can create agents that autonomously conduct literature reviews across myriads of sources, write and debug econometric code, fetch and analyze economic data, and coordinate complex research workflows. The paper demonstrates that by "vibe coding" (programming through natural language) and building on modern agentic frameworks like LangGraph, any economist can build sophisticated research assistants and other autonomous tools in minutes. By providing complete, working implementations alongside conceptual frameworks, this guide demonstrates how to employ AI agents in every stage of the research process, from initial investigation to f inal analysis
No ratings yet
View paper →Content that’s as good as contact? Vicarious intergroup contact and the promise of depolarization at scale
Lee-Or Ankori-Karlinsky, Robert Blair, Jessica Gottlieb, Samantha L. Moore-Berg
Political Science Research and Methods·2025
Abstract
Can observing opposing partisans engage in dialogue depolarize Americans at scale? Partisan animosity poses a challenge to democracy in the United States. Direct intergroup contact interventions have shown promise in reducing partisan polarization, but are costly, time-consuming, and sensitive to subtle changes in implementation. Vicarious intergroup contact—observing co-partisans engage with outparty members—offers a possible solution to the drawbacks of direct contact, and could potentially depolarize Americans quickly and at scale. We test this proposition using a pre-registered, placebo-controlled trial with a nationally representative sample of Americans. Using both attitudinal and behavioral measures, we find that a 50-minute documentary showing an intergroup contact workshop reduces polarization and increases interest but not investment in depolarization activities. While we find no evidence that the film mitigates anti-democratic attitudes, it does increase optimism about the survival of democratic institutions. Our findings suggest that vicarious intergroup contact delivered via mass media can be an effective, inexpensive, and scalable way to promote depolarization among Americans.
No ratings yet
View paper →