9607 research outputs found
Sort by
Prompt refinement or fine-tuning? best practices for using LLMs in computational social science tasks
Get PDFLarge Language Models are expressive tools that enable complex tasks of text understanding within Computational Social Science. Their versatility, while beneficial, poses a barrier for establishing standardized best practices within the field. To bring clarity on the values of different strategies, we present an overview of the performance of modern LLM-based classification methods on a benchmark of 23 social knowledge tasks. Our results point to three best practices: prioritize models with larger vocabulary and pre-training corpora; avoid simple zero-shot in favor of AI-enhanced prompting; fine-tune on task-specific data, and consider more complex forms instruction-tuning on multiple datasets only when only training data is more abundant
Nicer Than Humans: How Do Large Language Models Behave in the Prisoner's Dilemma?
Get PDFThe behavior of Large Language Models (LLMs) as artificial social agents is largely unexplored, and we still lack extensive evidence of how these agents react to simple social stimuli. Testing the behavior of AI agents in classic Game Theory experiments provides a promising theoretical framework for evaluating the norms and values of these agents in archetypal social situations. In this work, we investigate the cooperative behavior of three LLMs (Llama2, Llama3, and GPT3. 5) when playing the Iterated Prisoner's Dilemma against random adversaries displaying various levels of hostility. We introduce a systematic methodology to evaluate an LLM's comprehension of the game rules and its capability to parse historical gameplay logs for decision-making. We conducted simulations of games lasting for 100 rounds and analyzed the LLMs' decisions in terms of dimensions defined in the behavioral economics literature. We find that all models tend not to initiate defection but act cautiously, favoring cooperation over defection only when the opponent's defection rate is low. Overall, LLMs behave at least as cooperatively as the typical human player, although our results indicate some substantial differences among models. In particular, Llama2 and GPT3. 5 are more cooperative than humans, and especially forgiving and non-retaliatory for opponent defection rates below 30%. More similar to humans, Llama3 exhibits consistently uncooperative and exploitative behavior unless the opponent always cooperates. Our systematic approach to the study of LLMs in game theoretical scenarios is a step towards using these simulations to inform practices of LLM auditing and alignment
Estrangement through Silence
No full textHow can we cultivate deeper attunement to one another, ourselves, and the environment that can, in turn, inform and enrich design? Over the course of four workshops conducted across 1.5 years - primarily outdoors - the authors engaged in prolonged periods of shared silence. This collective silence functioned as an estrangement method, revealing the porous and interdependent boundaries between people and things, mutually constituting one another. We unpack some of the experiential qualities emerging from these experiments and mobilize them for future design processes, including: cultivating multifaceted sensibilities, dynamic modes of noticing and interacting, such as coming together and dispersing, being alone together, and acting or playing in unison; the malleability of silence to specific, orchestrated design activities, such as cooking or designing; and reframing silence, not as an absence, but as a presence - rich with sounds, interactions, and possibilities for engagement. We discuss how to set up temporal and spatial boundaries, alongside boundaries within and between ourselves
Which Graph Motif Parameters Count?
No full textFor a fixed graph H, the function #Ind(H → ⋆) maps graphs G to the count of induced H-copies in G; this function obviously "counts something" in that it has a combinatorial interpretation. Linear combinations of such functions are called graph motif parameters and have recently received significant attention in counting complexity after a seminal paper by Curticapean, Dell and Marx (STOC'17). We show that, among linear combinations of functions #Ind(H → ⋆) involving only graphs H without isolated vertices, precisely those with positive integer coefficients maintain a combinatorial interpretation. It is important to note that graph motif parameters can be nonnegative for all inputs G, even when some coefficients are negative.Formally, we show that evaluating any graph motif parameter with a negative coefficient is impossible in an oracle variant of #P, where an implicit graph is accessed by oracle queries. Our proof follows the classification of the relativizing closure properties of #P by Hertrampf, Vollmer, and Wagner (SCT'95) and the framework developed by Ikenmeyer and Pak (STOC'22), but our application of the required Ramsey theorem turns out to be more subtle, as graphs do not have the required Ramsey property.Our techniques generalize from graphs to relational structures, including colored graphs. Vastly generalizing this, we introduce motif parameters over categories that count occurrences of sub-objects in the category. We then prove a general dichotomy theorem that characterizes which such parameters have a combinatorial interpretation. Using known results in Ramsey theory for categories, we obtain a dichotomy for motif parameters of finite vector spaces as well as parameter sets
Extending perceived stereo baseline with vector-base amplitude panning and polarity inversion
No full textWe propose an extension to Vector-Base Amplitude Panning (VBAP) that simulates audio positioning beyond the physical baseline of a stereo speaker pair by introducing polarity inversion. The extension to VBAP, where audio sources are positioned on the line segment between loudspeaker vectors, refers to enabling panning outside the speaker positions, and thereby flipping the polarity of the signal from the opposite side speaker. The implementation allows for real-time processing. Listening experiments were conducted in a room with low reverberation on a pair of Bloomline Omnidrive Pro Mk II speakers, with eight participants who localized audio samples inside and beyond the stereo field. Results show that participants were able to perceive sound source positions up to 60% wider than the physical speaker span. The findings imply that by the polarity inversion VBAP can be generalised to, simulate audio positioning outside the speaker bounds and gives rise to a tool for creating virtual sound sources beyond the physical boundaries of the room in future immersive 3D applications
Tangles: Unpacking Extended Collision Experiences with Soma Trajectories
Get PDFWe reappraise the idea of colliding with robots, moving from a position that tries to avoid or mitigate collisions to one that considers them an important facet of human interaction. We report on a soma design workshop that explored how our bodies could collide with telepresence robots, mobility aids and a quadruped robot. Based on our findings, we employed soma trajectories to analyse collisions as extended experiences that negotiate key transitions of consent, preparation, launch, contact, ripple, sting, untangle, debris and reflect. We then employed these ideas to analyse two collision experiences, an accidental collision between a person and a drone and the deliberate design of a robot to play with cats, revealing how real-world collisions involve the complex and ongoing entanglement of soma trajectories. We discuss how viewing collisions as entangled trajectories, or ‘tangles’, can be used analytically, as a design approach, and as a lens to broach ethical complexity
The complementary and substitutional effects of forced and emergent mechanisms in multisourcing
No full textThis paper examines the effect of forced and emergent competition- and cooperation-enhancing mechanisms on joint multisourcing performance. We draw on research on coopetition in IS multisourcing and the literature on the crowding-out effect to theorise the interplay between these mechanisms. We argue that the key to understanding whether these mechanisms complement or substitute each other lies in the distinction between forced and emergent mechanisms, as these respectively invoke either an economic or a social logic among vendors. We test these ideas through a survey study of 108 multisourcing arrangements. Our results show that while a forced competition and an emergent cooperation mechanism can individually improve joint performance in multisourcing, the co-existence of economic and social logics results in a substitutional effect. A complementary effect is achieved when competition and cooperation mechanisms are of the same logic. Our study extends the existing IS outsourcing literature by shedding light on the role of forced and emergent mechanisms, either as competition or cooperation-enhancing, in enhancing multisourcing performance