Newest Papers
Compass Rose [CNF] | Abasi-maenyin Esebre | sevhage
Unknown Author
sevhage·2025
In partnership with Alinea Initiative, Goethe Institut Nigeria, African & Black Creatives for Development, and SELDI, SEVHAGE presents the winners of the Eugenia Abu/SEVHAGE International Prize for Creative Non-Fiction 2024. Abasi-maenyin Esebre’s piece is the joint-winning entry. I I failed English in WAEC. The second time I wrote the exam, I got an E. And…
★ 5.0 (1)
View paper →LatentPrompt: Optimizing Promts in Latent Space
Mateusz Bystroński, Grzegorz Piotrowski, Nitesh V. Chawla, Tomasz Kajdanowicz
arXiv·2025
Recent advances have shown that optimizing prompts for Large Language Models (LLMs) can significantly improve task performance, yet many optimization techniques rely on heuristics or manual exploration. We present LatentPrompt, a model-agnostic framework for prompt optimization that leverages latent semantic space to automatically generate, evaluate, and refine candidate prompts without requiring hand-crafted rules. Beginning with a set of seed prompts, our method embeds them in a continuous latent space and systematically explores this space to identify prompts that maximize task-specific performance. In a proof-of-concept study on the Financial PhraseBank sentiment classification benchmark, LatentPrompt increased classification accuracy by approximately 3 percent after a single optimization cycle. The framework is broadly applicable, requiring only black-box access to an LLM and an automatic evaluation metric, making it suitable for diverse domains and tasks.
No ratings yet
View paper →Campaigning for Extinction: Eradication of Sparrows and the Great Famine in China
Eyal G. Frank, Qinyun Wang, Shaoda Wang, Xuebin Wang, Yang You
NBER Working Paper·2025
How do large disruptions to ecosystems affect human well-being? This paper tests the long-standing hypothesis that China's 1958 Four Pests Campaign, which exterminated sparrows despite scientists’ warnings about their pest-control role, exacerbated the Great Famine—the largest in human history. Combining newly digitized data on historical agricultural productivity in China with habitat suitability modeling methods in ecology, we find that, after sparrow eradication, a one-standard-deviation increase in sparrow suitability led to 5.3% larger rice and 8.7% larger wheat declines. State food procurement exacerbated these losses, resulting in a 9.6% higher mortality in high-suitability counties—implying nearly two million excess deaths.
No ratings yet
View paper →A versatile antibody capture system drives specific in vivo delivery of mRNA-loaded lipid nanoparticles
Moore Z. Chen, Daniel Yuen, Victoria M. McLeod, Ken W. Yong, Cameron H. Smyth, Bruna Rossi Herling, Thomas. J. Payne, Stewart A. Fabb, Matthew J. Belousoff, Azizah Algarni, Patrick M. Sexton, Christopher J. H. Porter, Colin W. Pouton, Angus P. R. Johnston
Nature Nanotechnology·2025·2 citations
<jats:title>Abstract</jats:title>
<jats:p>Efficient and precise delivery of mRNA is critical to advance mRNA therapies beyond their current use as vaccines. Lipid nanoparticles (LNPs) efficiently encapsulate and protect mRNA, but non-specific cellular uptake may lead to off-target delivery and minimal delivery to target cells. Functionalizing LNPs with antibodies enables targeted mRNA delivery, but traditional modification techniques require complex conjugation and purification, which often reduces antibody affinity. Here we present a simple method for capturing antibodies in their optimal orientation on LNPs, without antibody modification or complex purification. This strategy uses an optimally oriented anti-Fc nanobody on the LNP surface to capture antibodies, resulting in protein expression levels more than 1,000 times higher than non-targeted LNPs and more than 8 times higher than conventional antibody functionalization techniques. These precisely targeted LNPs showed highly efficient in vivo targeting to T cells, with minimal delivery to other immune cells. This approach enables the rapid development of targeted LNPs and has the potential to broaden the use of mRNA therapies.</jats:p>
No ratings yet
View paper →Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens
Chengshuai Zhao, Zhen Tan, Pingchuan Ma, Dawei Li, Bohan Jiang, Yancheng Wang, Yingzhen Yang, Huan Liu
arXiv·2025
Chain-of-Thought (CoT) prompting has been shown to improve Large Language Model (LLM) performance on various tasks. With this approach, LLMs appear to produce human-like reasoning steps before providing answers (a.k.a., CoT reasoning), which often leads to the perception that they engage in deliberate inferential processes. However, some initial findings suggest that CoT reasoning may be more superficial than it appears, motivating us to explore further. In this paper, we study CoT reasoning via a data distribution lens and investigate if CoT reasoning reflects a structured inductive bias learned from in-distribution data, allowing the model to conditionally generate reasoning paths that approximate those seen during training. Thus, its effectiveness is fundamentally bounded by the degree of distribution discrepancy between the training data and the test queries. With this lens, we dissect CoT reasoning via three dimensions: task, length, and format. To investigate each dimension, we design DataAlchemy, an isolated and controlled environment to train LLMs from scratch and systematically probe them under various distribution conditions. Our results reveal that CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing challenge of achieving genuine and generalizable reasoning.
No ratings yet
View paper →Your new freelance muse: How I went from $150k on salary to $30k/month freelance with Allie Carr
Grace Abbott
Substack·2025
🦪 What if you are miserable in your current job because some deep part of you knows that something better, bigger, and more aligned is around the corner?
No ratings yet
View paper →Video Generators are Robot Policies
Junbang Liang, Pavel Tokmakov, Ruoshi Liu, Sruthi Sudhakar, Paarth Shah, Rares Ambrus, Carl Vondrick
arXiv·2025
Despite tremendous progress in dexterous manipulation, current visuomotor policies remain fundamentally limited by two challenges: they struggle to generalize under perceptual or behavioral distribution shifts, and their performance is constrained by the size of human demonstration data. In this paper, we use video generation as a proxy for robot policy learning to address both limitations simultaneously. We propose Video Policy, a modular framework that combines video and action generation that can be trained end-to-end. Our results demonstrate that learning to generate videos of robot behavior allows for the extraction of policies with minimal demonstration data, significantly improving robustness and sample efficiency. Our method shows strong generalization to unseen objects, backgrounds, and tasks, both in simulation and the real world. We further highlight that task success is closely tied to the generated video, with action-free video data providing critical benefits for generalizing to novel tasks. By leveraging large-scale video generative models, we achieve superior performance compared to traditional behavior cloning, paving the way for more scalable and data-efficient robot policy learning.
No ratings yet
View paper →The Inside Story: Body Language and Free Indirect Discourse in Aphra Behn's <i>Oroonoko</i>
Mark Blackwell
Novel: A Forum on Fiction·2025
<jats:title>Abstract</jats:title>
<jats:p>The narrator of Aphra Behn's Oroonoko not only claims privileged access to what we might now call the eponymous protagonist's “inside story,” but works to validate that claim by invoking evidence drawn from the inside—by describing physical specimens conveyed from Surinam to England as empirical proof of its wonders, for instance, and by experimenting briefly with psycho-narration and free indirect discourse in an effort to provide a perspective on Oroonoko's state of mind from within his own consciousness. Yet this narrative investment in accessing interior truths is haunted by Oroonoko's fruitless struggle to plumb others’ hidden motives; indeed, his self-evisceration serves as a grisly commentary on the narrator's effort to extract some deeper inside story from her interviews with Oroonoko—to “draw the Nobler part, the Soul and Mind.” In Behn's hands, free indirect discourse does not so much render other minds visible as reveal insuperable barriers to knowing what another thinks and how another feels, barriers previously adumbrated by the text's insistent attention to the dubious reliability of both linguistic and nonlinguistic media of communication.</jats:p>
No ratings yet
View paper →Key Considerations When Designing a Social Media-Based Diary Study
Ellie Atayee-Bennett
International Journal of Qualitative Methods·2025·1 citations
<jats:p>Qualitative data collection conducted on social media platforms is a recent methodological innovation with only a handful of studies exploring this approach. Even fewer explore the potential of social media platforms for conducting diary methods. This article highlights the key considerations a researcher must reflect upon when designing a social media-based diary study focused on everyday lived experience. It argues that due consideration must be afforded to the platform choice, design and structure of the diary, ethical considerations, cultural considerations, open communication, and strategies to improve participant engagement. In so doing, the researcher can enhance the quality of their study and gain better insight into their research topic. This article is based on a research study which explored the everyday lived experiences of 36 UK-based faith vegans, and which utilized private Facebook and WhatsApp groups to obtain topic-focused diary data from participants, such as photos of food, reflections on different experiences and events, and group discussions.</jats:p>
No ratings yet
View paper →Persona vectors: Monitoring and controlling character traits in language models
Anthropic
Anthropic Blog·2025
A paper from Anthropic describing persona vectors and their applications to monitoring and controlling model behavior
No ratings yet
View paper →Validation of the Coffee Knowledge Test for Assessing Expertise in Coffee
Jonas Yde Junge, Ilja Croijmans, Line Ahm Mielby, Asifa Majid, Ulla Kidmose
Journal of Sensory Studies·2025
<jats:title>ABSTRACT</jats:title><jats:p>Perception and appreciation of food and beverages are shaped by an individual's knowledge and expertise. While wine expertise has been widely studied, less is known about how coffee expertise influences sensory perception. This study aimed to validate the Coffee Knowledge Test developed by Croijmans and Majid (2016) as a tool for assessing coffee expertise. We adapted the test and examined its reliability using data from 267 participants, comparing novice consumers and coffee professionals. Internal consistency was evaluated using item variability, correlation analysis, and Cronbach's <jats:italic>α</jats:italic>, while external validity was assessed by comparing scores between the two groups. The test showed sufficient item variability, appropriate correlations, and strong internal consistency. Importantly, coffee professionals scored significantly higher than novices. These results support the Coffee Knowledge Test as a valid and reliable instrument for measuring coffee‐related knowledge and expertise.</jats:p>
No ratings yet
View paper →SIMA 2: An Agent that Plays, Reasons, and Learns With You in Virtual 3D Worlds
SIMA Team
DeepMind Blog·2025
Introducing SIMA 2, the next milestone in research creating general and helpful AI agents. By integrating the advanced capabilities of Gemini models, SIMA is evolving from an instruction-following agent to one that can play, reason, and learn alongside humans in virtual 3D worlds.
No ratings yet
View paper →