Victor Storchan

Julien Abadji, Marah Abdin, Connor Adams et al.

We present Laguna M.1 and Laguna XS.2, two Mixture-of-Experts foundation models built for long-horizon, agentic coding: M.1 has $225.8$B total parameters ($23.4$B activated per token) and XS.2 has $33.4$B total ($3$B activated). Both models were trained from scratch end-to-end inside the same internal system that we refer to as our Model Factory: a tightly-integrated stack of versioned data, training, evaluation, and inference components that turn model development into an industrial process. We describe the principles and design choices of the Model Factory and also detail the end-to-end training process of our models, throughout pre-training data and architecture, post-training stages, evaluation, and quantization. On agentic software engineering and terminal benchmarks (SWE-bench Verified, SWE-bench Multilingual, SWE-Bench Pro, and Terminal-Bench 2.0) M.1 and XS.2 are competitive with state-of-the-art open models in their respective weight classes. Laguna XS.2 weights are released under Apache~2.0 at https://huggingface.co/collections/poolside/laguna-xs2.

Pierre Colombo, Victor Pellegrain, Malik Boudiaf et al.

Proprietary and closed APIs are becoming increasingly common to process natural language, and are impacting the practical applications of natural language processing, including few-shot classification. Few-shot classification involves training a model to perform a new classification task with a handful of labeled data. This paper presents three contributions. First, we introduce a scenario where the embedding of a pre-trained model is served through a gated API with compute-cost and data-privacy constraints. Second, we propose a transductive inference, a learning paradigm that has been overlooked by the NLP community. Transductive inference, unlike traditional inductive learning, leverages the statistics of unlabeled data. We also introduce a new parameter-free transductive regularizer based on the Fisher-Rao loss, which can be used on top of the gated API embeddings. This method fully utilizes unlabeled data, does not share any label with the third-party API provider and could serve as a baseline for future research. Third, we propose an improved experimental setting and compile a benchmark of eight datasets involving multiclass classification in four different languages, with up to 151 classes. We evaluate our methods using eight backbone models, along with an episodic evaluation over 1,000 episodes, which demonstrate the superiority of transductive inference over the standard inductive setting.

Adrien Basdevant, Camille François, Victor Storchan et al.

Over the past year, there has been a robust debate about the benefits and risks of open sourcing foundation models. However, this discussion has often taken place at a high level of generality or with a narrow focus on specific technical attributes. In part, this is because defining open source for foundation models has proven tricky, given its significant differences from traditional software development. In order to inform more practical and nuanced decisions about opening AI systems, including foundation models, this paper presents a framework for grappling with openness across the AI stack. It summarizes previous work on this topic, analyzes the various potential reasons to pursue openness, and outlines how openness varies in different parts of the AI stack, both at the model and at the system level. In doing so, its authors hope to provide a common descriptive framework to deepen a nuanced and rigorous understanding of openness in AI and enable further work around definitions of openness and safety in AI.

Camille François, Ludovic Péran, Ayah Bdeir et al.

The rapid rise of open-weight and open-source foundation models is intensifying the obligation and reshaping the opportunity to make AI systems safe. This paper reports outcomes from the Columbia Convening on AI Openness and Safety (San Francisco, 19 Nov 2024) and its six-week preparatory programme involving more than forty-five researchers, engineers, and policy leaders from academia, industry, civil society, and government. Using a participatory, solutions-oriented process, the working groups produced (i) a research agenda at the intersection of safety and open source AI; (ii) a mapping of existing and needed technical interventions and open source tools to safely and responsibly deploy open foundation models across the AI development workflow; and (iii) a mapping of the content safety filter ecosystem with a proposed roadmap for future research and development. We find that openness -- understood as transparent weights, interoperable tooling, and public governance -- can enhance safety by enabling independent scrutiny, decentralized mitigation, and culturally plural oversight. However, significant gaps persist: scarce multimodal and multilingual benchmarks, limited defenses against prompt-injection and compositional attacks in agentic systems, and insufficient participatory mechanisms for communities most affected by AI harms. The paper concludes with a roadmap of five priority research directions, emphasizing participatory inputs, future-proof content filters, ecosystem-wide safety infrastructure, rigorous agentic safeguards, and expanded harm taxonomies. These recommendations informed the February 2025 French AI Action Summit and lay groundwork for an open, plural, and accountable AI safety discipline.

Sayash Kapoor, Rishi Bommasani, Kevin Klyman et al.

Foundation models are powerful technologies: how they are released publicly directly shapes their societal impact. In this position paper, we focus on open foundation models, defined here as those with broadly available model weights (e.g. Llama 2, Stable Diffusion XL). We identify five distinctive properties (e.g. greater customizability, poor monitoring) of open foundation models that lead to both their benefits and risks. Open foundation models present significant benefits, with some caveats, that span innovation, competition, the distribution of decision-making power, and transparency. To understand their risks of misuse, we design a risk assessment framework for analyzing their marginal risk. Across several misuse vectors (e.g. cyberattacks, bioweapons), we find that current research is insufficient to effectively characterize the marginal risk of open foundation models relative to pre-existing technologies. The framework helps explain why the marginal risk is low in some cases, clarifies disagreements about misuse risks by revealing that past work has focused on different subsets of the framework with different assumptions, and articulates a way forward for more constructive debate. Overall, our work helps support a more grounded assessment of the societal impact of open foundation models by outlining what research is needed to empirically validate their theoretical benefits and risks.

Eren Kurshan, Jiahao Chen, Victor Storchan et al.

Artificial intelligence (AI) continues to find more numerous and more critical applications in the financial services industry, giving rise to fair and ethical AI as an industry-wide objective. While many ethical principles and guidelines have been published in recent years, they fall short of addressing the serious challenges that model developers face when building ethical AI solutions. We survey the practical and overarching issues surrounding model development, from design and implementation complexities, to the shortage of tools, and the lack of organizational constructs. We show how practical considerations reveal the gaps between high-level principles and concrete, deployed AI applications, with the aim of starting industry-wide conversations toward solution approaches.

Jiahao Chen, Victor Storchan, Eren Kurshan

We review practical challenges in building and deploying ethical AI at the scale of contemporary industrial and societal uses. Apart from the purely technical concerns that are the usual focus of academic research, the operational challenges of inconsistent regulatory pressures, conflicting business goals, data quality issues, development processes, systems integration practices, and the scale of deployment all conspire to create new ethical risks. Such ethical concerns arising from these practical considerations are not adequately addressed by existing research results. We argue that a holistic consideration of ethics in the development and deployment of AI systems is necessary for building ethical AI in practice, and exhort researchers to consider the full operational contexts of AI systems when assessing ethical risks.

Jiahao Chen, Victor Storchan

Regulators have signalled an interest in adopting explainable AI(XAI) techniques to handle the diverse needs for model governance, operational servicing, and compliance in the financial services industry. In this short overview, we review the recent technical literature in XAI and argue that based on our current understanding of the field, the use of XAI techniques in practice necessitate a highly contextualized approach considering the specific needs of stakeholders for particular business applications.

Victor Storchan, Svitlana Vyetrenko, Tucker Balch

In electronic trading markets often only the price or volume time series, that result from interaction of multiple market participants, are directly observable. In order to test trading strategies before deploying them to real-time trading, multi-agent market environments calibrated so that the time series that result from interaction of simulated agents resemble historical are often used. To ensure adequate testing, one must test trading strategies in a variety of market scenarios -- which includes both scenarios that represent ordinary market days as well as stressed markets (most recently observed due to the beginning of COVID pandemic). In this paper, we address the problem of multi-agent simulator parameter calibration to allow simulator capture characteristics of different market regimes. We propose a novel two-step method to train a discriminator that is able to distinguish between "real" and "fake" price and volume time series as a part of GAN with self-attention, and then utilize it within an optimization framework to tune parameters of a simulator model with known agent archetypes to represent a market scenario. We conclude with experimental results that demonstrate effectiveness of our method.