arXiv subjects

Francesco Bertolotti

Publications and source records attributed to Francesco Bertolotti.

8 recordsLinked to original sources

arXiv2026

Is Lying an Emergent Behaviour in LLMs? Evidence from Gaslighting AI agents in a Sustainability Game

LLMs agents are increasingly used in multi-agent settings, yet their behaviour in sustainability games remains largely unexplored. This work investigates whether lying can emerge among LLM agents in a competitive sustainability game in which agents are informed that common resources can regenerate, although regeneration does not actually occur. We develop an agent-based model of a sustainability game in which agents manage industrial, military, and ecological resources, and interact through a network. LLM agents can observe neighbours' status, declare future attacks, receive permission to lie, and access reputation information, while rule-based agents provide an interpretable behavioural baseline. The results show that neighbour information strongly changes system dynamics, increasing attacks while improving biosphere retention and coexistence. Also, the presence of future declarations reduce extinction risk without suppressing conflict. Behaviourally, deception emerges even when agents are not explicitly allowed to lie, and explicit permission mainly increases bluffing and diversion rather than direct backstabbing. Finally, the presence of reputation memory and information about the current biosphere level reduces system ecological depletion. These findings suggest that deception can arise as an emergent behaviour in LLM-agent systems and that communication between LLM-agents could support sustainability while dealing with risk.

cs.MA↗

arXiv2026

Domyn-Small: A European 10B Reasoning Language Model

We introduce Domyn-Small, a 10-billion-parameter open-weight reasoning language model released under the MIT license. Domyn-Small is the product of an initial pre-training phase on 9 trillion tokens multilingual data, followed by a post-training pipeline for reasoning, instruction following, and context extension. For the latter, we performed a Continued Pre-Training (CPT) phase that doubles the native context window to 32K tokens, followed by SFT with a math-focused annealing run. Finally, the RL phase includes GRPO with verifiable rewards, DPO, and a multi-environment GRPO stage spanning five task domains: mathematics, code, multiple-choice QA, instruction-following, and tool calling. The 32K-token native context extends to 128K at inference via YaRN, and a chat-template toggle enables dual-mode reasoning. Against peer models in the 7--10B class (Qwen3.5-9B, OLMo-3-7B-Think, Nemotron-Nano-8B, Ministral-3-8B), Domyn-Small achieves a strong accuracy-efficiency balance: it produces roughly one-third as many tokens as Qwen3.5-9B and approximately 35% of OLMo-3-7B-Think's token budget on core reasoning benchmarks, while delivering strong instruction-following (IFEval 79.9) and competitive science reasoning (GPQA-Diamond 50.0). We release the weights and the post-training recipe alongside Domyn Swarm (Apache~2.0), an open-source framework for scalable LLM inference on HPC clusters developed during this program and used throughout this work.

cs.CL↗

arXiv2026

Emergence of Fragility in LLM-based Social Networks: the Case of Moltbook

The rapid diffusion of large language models and the growth in their capability has enabled the emergence of online environments populated by autonomous AI agents that interact through natural language. These platforms provide a novel empirical setting for studying collective dynamics among artificial agents. In this paper we analyze the interaction network of Moltbook, a social platform composed entirely of LLM based agents, using tools from network science. The dataset comprises 39,924 users, 235,572 posts, and 1,540,238 comments collected through web scraping. We construct a directed weighted network in which nodes represent agents and edges represent commenting interactions. Our analysis reveals strongly heterogeneous connectivity patterns characterized by heavy tailed degree and activity distributions. At the mesoscale, the network exhibits a pronounced core periphery organization in which a very small structural core (0.9% of nodes) concentrates a large fraction of connectivity. Robustness experiments show that the network is relatively resilient to random node removal but highly vulnerable to targeted attacks on highly connected nodes, particularly those with high out degree. These findings indicate that the interaction structure of AI agent social systems may develop strong centralization and structural fragility, providing new insights into the collective organization of LLM native social environments.

cs.SI↗

arXiv2026

Semantic Delta: An Interpretable Signal Differentiating Human and LLMs Dialogue

Do LLMs talk like us? This question intrigues a multitude of scholar and it is relevant in many fields, from education to academia. This work presents an interpretable statistical feature for distinguishing human written and LLMs generated dialogue. We introduce a lightweight metric derived from semantic categories distribution. Using the Empath lexical analysis framework, each text is mapped to a set of thematic intensity scores. We define semantic delta as the difference between the two most dominant category intensities within a dialogue, hypothesizing that LLM outputs exhibit stronger thematic concentration than human discourse. To evaluate this hypothesis, conversational data were generated from multiple LLM configurations and compared against heterogeneous human corpora, including scripted dialogue, literary works, and online discussions. A Welch t-test was applied to the resulting distributions of semantic delta values. Results show that AI-generated texts consistently produce higher deltas than human texts, indicating a more rigid topics structure, whereas human dialogue displays a broader and more balanced semantic spread. Rather than replacing existing detection techniques, the proposed zero-shot metric provides a computationally inexpensive complementary signal that can be integrated into ensemble detection systems. These finding also contribute to the broader empirical understanding of LLM behavioural mimicry and suggest that thematic distribution constitutes a quantifiable dimension along which current models fall short of human conversational dynamics.

cs.CL↗

arXiv2026

Toward a new AI winter? How diffusion of technological innovation on networks leads to chaotic boom-bust cycles

Technological developments and the impact of artificial intelligence (AI) are omnipresent themes and concerns of the present day. Much has been written on these topics but applications of quantitative models to understand the techno-social landscape have been much more limited. We propose a mathematical model that can help understand in a unified manner the patterns underlying technological development and also identify the different regimes in which the technological landscape evolves. First, we develop a model of innovation diffusion between different technologies, the growth of each reinforcing the development of the others. The model has a variable that quantifies the level of development (or innovation, discovery) potential for a given technology. The potential, or market capacity, increases via diffusion from related technologies, reflecting the fact that a technology does not develop in isolation. Hence, the growth of each technology is influenced by how developed its neighboring (related) technologies are. This allows us to reproduce long-term trends seen in computing technology and large language models (LLMs). We then present a three-dimensional system of supply, demand, and investment which shows oscillations (business cycles) emerging if investment is too high into a given technology, product, or market. We finally combine the two models through a common variable and show that if investment or diffusion is too high in the network context, chaotic boom-bust cycles can emerge. These quantitative considerations allow us to reproduce the boom-bust patterns seen in non-fungible token (NFT) transaction data and also have deep implications for the development of AI which we highlight, such as the arrival of a new AI winter.

physics.soc-ph↗

arXiv2026

Comprehension vs. Adoption: Evaluating a Language Workbench Through a Family of Experiments

Language workbenches are tools that enable the definition, reuse, and composition of programming languages and their ecosystems, aiming to streamline language development. To facilitate their adoption by language designers, the comprehensibility of the language used to define other languages is an important aspect to evaluate. Moreover, considering that language workbenches are relatively new tools, user acceptance emerges as a crucial factor to be accounted for during their assessment. Current literature often neglects user-centred aspects like comprehensibility and acceptance in the assessment of this breed of tools. This paper addresses this gap through a family of experiments assessing Neverlang, a modular language workbench. The study adopts a tailored version of the Method Evaluation Model (MEM) to evaluate the comprehensibility of Neverlang's meta-language and programs, as well as user acceptance in terms of perceived ease of use, perceived usefulness, and intention to use. It also investigates the relationships among these dimensions. The experiments were conducted in three iterations involving participants from academia. The results reveal that users demonstrate sufficient comprehension of Neverlang's meta-language, particularly concerning its syntax, express a favourable perception of its usefulness, and indicate their intention to use it. However, the results also indicate that Neverlang's ease of use remains a challenge. Additionally, variations in the perceived ease of use and perceived usefulness, whether low or high, influence the users' intention to use the tool. Surprisingly, no significant correlation is found between comprehensibility and user acceptance. Notably, higher comprehensibility of the meta-language does not necessarily translate into greater acceptance, underscoring the complex interplay between comprehension and adoption.

cs.SE↗

arXiv2025

An LLM-based Delphi Study to Predict GenAI Evolution

Predicting the future trajectory of complex and rapidly evolving systems remains a significant challenge, particularly in domains where data is scarce or unreliable. This study introduces a novel approach to qualitative forecasting by leveraging Large Language Models to conduct Delphi studies. The methodology was applied to explore the future evolution of Generative Artificial Intelligence, revealing insights into key factors such as geopolitical tensions, economic disparities, regulatory frameworks, and ethical considerations. The results highlight how LLM-based Delphi studies can facilitate structured scenario analysis, capturing diverse perspectives while mitigating issues such as respondent fatigue. However, limitations emerge in terms of knowledge cutoffs, inherent biases, and sensitivity to initial conditions. While the approach provides an innovative means for structured foresight, this method could be also considered as a novel form of reasoning. further research is needed to refine its ability to manage heterogeneity, improve reliability, and integrate external data sources.

cs.AI↗

arXiv2022

A Master Equation for Power Laws

We propose a new mechanism for generating power laws. Starting from a random walk, we first outline a simple derivation of the Fokker-Planck equation. By analogy, starting from a certain Markov chain, we derive a master equation for power laws that describes how the number of cascades changes over time (cascades are consecutive transitions that end when the initial state is reached). The partial differential equation has a closed form solution which gives an explicit dependence of the number of cascades on their size and on time. Furthermore, the power law solution has a natural cut-off, a feature often seen in empirical data. This is due to the finite size a cascade can have in a finite time horizon. The derivation of the equation provides a justification for an exponent equal to 2, which agrees well with several empirical distributions, including Richardson's law on the size and frequency of deadly conflicts. Nevertheless, the equation can be solved for any exponent value. In addition, we propose an urn model where the number of consecutive ball extractions follows a power law. In all cases, the power law is manifest over the entire range of cascade sizes, as shown through log-log plots in the frequency and rank distributions.

cond-mat.stat-mech↗