This paper introduces PhyloLM, a method adapting phylogenetic algorithms to Large Language Models (LLMs) to explore whether and how they relate to each other and to predict their performance characteristics. Our method calculates a phylogenetic distance metric based on the similarity of LLMs' output. The resulting metric is then used to construct dendrograms, which satisfactorily capture known relationships across a set of 111 open-source and 45 closed models. Furthermore, our phylogenetic distance predicts performance in standard benchmarks, thus demonstrating its functional validity and paving the way for a time and cost-effective estimation of LLM capabilities. To sum up, by translating population genetic concepts to machine learning, we propose and validate a tool to evaluate LLM development, relationships and capabilities, even in the absence of transparent training information.

The ability to correctly estimate the probability of one's choices being correct is fundamental to optimally re-evaluate previous choices or to arbitrate between different decision strategies. Experimental evidence nonetheless suggests that this metacognitive process-confidence judgment- is susceptible to numerous biases. Here, we investigate the effect of outcome valence (gains or losses) on confidence while participants learned stimulus-outcome associations by trial-and-error. In two experiments, participants were more confident in their choices when learning to seek gains compared to avoiding losses, despite equal difficulty and performance between those two contexts. Computational modelling revealed that this bias is driven by the context-value, a dynamically updated estimate of the average expected-value of choice options, necessary to explain equal performance in the gain and loss domain. The biasing effect of context-value on confidence, revealed here for the first time in a reinforcement-learning context, is therefore domain-general, with likely important functional consequences. We show that one such consequence emerges in volatile environments, where the (in)flexibility of individuals' learning strategies differs when outcomes are framed as gains or losses. Despite apparent similar behavior- profound asymmetries might therefore exist between learning to avoid losses and learning to seek gains.

Depression is characterized by a marked decrease in social interactions and blunted sensitivity to rewards. Surprisingly, despite the importance of social deficits in depression, non-social aspects have been disproportionally investigated. As a consequence, the cognitive mechanisms underlying atypical decision-making in social contexts in depression are poorly understood. In the present study, we investigate whether deficits in reward processing interact with the social context and how this interaction is affected by self-reported depression and anxiety symptoms in the general population. Two cohorts of subjects (discovery and replication sample: N = 50 each) took part in an experiment involving reward learning in contexts with different levels of social information (absent, partial and complete). Behavioral analyses revealed a specific detrimental effect of depressive symptoms-but not anxiety-on behavioral performance in the presence of social information, i.e. when participants were informed about the choices of another player. Model-based analyses further characterized the computational nature of this deficit as a negative audience effect, rather than a deficit in the way others' choices and rewards are integrated in decision making. To conclude, our results shed light on the cognitive and computational mechanisms underlying the interaction between social cognition, reward learning and decision-making in depressive disorders.

Value-based decision-making involves trading off the cost associated with an action against its expected reward. Research has shown that both physical and mental effort constitute such subjective costs, biasing choices away from effortful actions, and discounting the value of obtained rewards. Facing conflicts between competing action alternatives is considered aversive, as recruiting cognitive control to overcome conflict is effortful. Moreover, engaging control to proactively suppress irrelevant information that could conflict with task-relevant information would presumably also be cognitively costly. Yet, it remains unclear whether the cognitive control demands involved in preventing and resolving conflict also constitute costs in value-based decisions. The present study investigated this question by embedding irrelevant distractors (flanker arrows) within a reversal-learning task, with intermixed free and instructed trials. Results showed that participants learned to adapt their free choices to maximize rewards, but were nevertheless biased to follow the suggestions of irrelevant distractors. Thus, the perceived cost of investing cognitive control to suppress an external suggestion could sometimes trump internal value representations. By adapting computational models of reinforcement learning, we assessed the influence of conflict at both the decision and learning stages. Modelling the decision showed that free choices were more biased when participants were less sure about which action was more rewarding. This supports the hypothesis that the costs linked to conflict management were traded off against expected rewards. During the learning phase, we found that learning rates were reduced in instructed, relative to free, choices. Learning rates were further reduced by conflict between an instruction and subjective action values, whereas learning was not robustly influenced by conflict between one's actions and external distractors. Our results show that the subjective cognitive control costs linked to conflict factor into value-based decision-making, and highlight that different types of conflict may have different effects on learning about action outcomes.

Previous studies suggest that factual learning, that is, learning from obtained outcomes, is biased, such that participants preferentially take into account positive, as compared to negative, prediction errors. However, whether or not the prediction error valence also affects counterfactual learning, that is, learning from forgone outcomes, is unknown. To address this question, we analysed the performance of two groups of participants on reinforcement learning tasks using a computational model that was adapted to test if prediction error valence influences learning. We carried out two experiments: in the factual learning experiment, participants learned from partial feedback (i.e., the outcome of the chosen option only); in the counterfactual learning experiment, participants learned from complete feedback information (i.e., the outcomes of both the chosen and unchosen option were displayed). In the factual learning experiment, we replicated previous findings of a valence-induced bias, whereby participants learned preferentially from positive, relative to negative, prediction errors. In contrast, for counterfactual learning, we found the opposite valence-induced bias: negative prediction errors were preferentially taken into account, relative to positive ones. When considering valence-induced bias in the context of both factual and counterfactual learning, it appears that people tend to preferentially take into account information that confirms their current choice.

Adolescence is a period of life characterised by changes in learning and decision-making. Learning and decision-making do not rely on a unitary system, but instead require the coordination of different cognitive processes that can be mathematically formalised as dissociable computational modules. Here, we aimed to trace the developmental time-course of the computational modules responsible for learning from reward or punishment, and learning from counterfactual feedback. Adolescents and adults carried out a novel reinforcement learning paradigm in which participants learned the association between cues and probabilistic outcomes, where the outcomes differed in valence (reward versus punishment) and feedback was either partial or complete (either the outcome of the chosen option only, or the outcomes of both the chosen and unchosen option, were displayed). Computational strategies changed during development: whereas adolescents' behaviour was better explained by a basic reinforcement learning algorithm, adults' behaviour integrated increasingly complex computational features, namely a counterfactual learning module (enabling enhanced performance in the presence of complete feedback) and a value contextualisation module (enabling symmetrical reward and punishment learning). Unlike adults, adolescent performance did not benefit from counterfactual (complete) feedback. In addition, while adults learned symmetrically from both reward and punishment, adolescents learned from reward but were less likely to learn from punishment. This tendency to rely on rewards and not to consider alternative consequences of actions might contribute to our understanding of decision-making in adolescence.