arXiv subjects

Nikos I. Bosse

Publications and source records attributed to Nikos I. Bosse.

8 recordsLinked to original sources

arXiv2026

Evaluating Strategic Reasoning in Forecasting Agents

Forecasting benchmarks produce accuracy leaderboards but little insight into why some forecasters are more accurate than others. We introduce Bench to the Future 2 (BTF-2), 1,417 pastcasting questions with a frozen 15M-document research corpus in which agents reproducibly research and forecast offline, producing full reasoning traces. BTF-2 detects accuracy differences of 0.004 Brier score, and can distinguish differential agent strengths in research vs. judgment. We build a forecaster 0.011 Brier more accurate than any single frontier agent, and use it to evaluate agent strategic reasoning without hindsight bias. We find the better forecaster differs primarily in its pre-mortem analysis of its blind spots and consideration of black swans. Expert human forecasters found the dominant strategic reasoning failures of frontier agents are in assessing political and business leaders' incentives, judging their likelihood to follow through on stated plans, and modeling institutional processes.

cs.AI↗

arXiv2026

Automating Forecasting Question Generation and Resolution for AI Evaluation

Forecasting future events is highly valuable in decision-making and is a robust measure of general intelligence. As forecasting is probabilistic, developing and evaluating AI forecasters requires generating large numbers of diverse and difficult questions, and accurately resolving them. Previous efforts to automate this laborious work relied on recurring data sources (e.g., weather, stocks), limiting diversity and utility. In this work, we present a system for generating and resolving high-quality forecasting questions automatically and at scale using LLM-powered web research agents. We use this system to generate 1499 diverse, real-world forecasting questions, and to resolve them several months later. We estimate that our system produces verifiable, unambiguous questions approximately 96% of the time, exceeding the rate of Metaculus, a leading human-curated forecasting platform. We also find that our system resolves questions at approximately 95% accuracy. We verify that forecasting agents powered by more intelligent LLMs perform better on these questions (Brier score of 0.134 for Gemini 3 Pro, 0.149 for GPT-5, and 0.179 for Gemini 2.5 Flash). Finally, we demonstrate how our system can be leveraged to directly improve forecasting, by evaluating a question decomposition strategy on a generated question set, yielding a significant improvement in Brier scores (0.132 vs. 0.141).

cs.LG↗

arXiv2025

Bench to the Future: A Pastcasting Benchmark for Forecasting Agents

Forecasting is a challenging task that offers a clearly measurable way to study AI systems. Forecasting requires a large amount of research on the internet, and evaluations require time for events to happen, making the development of forecasting benchmarks challenging. To date, no forecasting benchmark provides a realistic, hermetic, and repeatable environment for LLM forecasters. We introduce Bench To the Future (BTF), a "pastcasting" benchmark with hundreds of high-quality questions for which the resolution is already known. Each question is accompanied by a large offline corpus of tens of thousands of relevant web pages, enabling a way to elicit realistic "forecasts" on past events from LLMs. Results suggest that our pastcasting environment can produce results comparable to those based on forecasts using the internet on at-the-time unresolved questions. We show results benchmarking agent and chain-of-thought forecasting approaches using several LLMs, including the recently-released Claude 4 models, and demonstrate BTF's ability to track steady forecasting capability progress over time. We intend this to be a living benchmark, with new questions added continually to account for increasing training data cutoff dates. We invite researchers to contact us at hello@futuresearch.ai to utilize our benchmark or tooling for their own research.

cs.CL↗

arXiv2025

Deep Research Bench: Evaluating AI Web Research Agents

Amongst the most common use cases of modern AI is LLM chat with web search enabled. However, no direct evaluations of the quality of web research agents exist that control for the continually-changing web. We introduce Deep Research Bench, consisting of 89 multi-step web research task instances of varying difficulty across 8 diverse task categories, with the answers carefully worked out by skilled humans. We provide a "RetroSearch" environment with a large frozen set of scraped web pages, and demonstrate that offline "RetroSearch" agents perform comparably to "live web" agents, enabling reliable evaluations of models over time. We provide robust agent tooling and scaffolding to benchmark major LLMs as they are released, including "thinking" models like o3 and Gemini 2.5 Pro. We include automated evaluations of the lengthy agent traces to report progress over time in hallucinations, tool use, and forgetting. Finally, we evaluate the major web research products branded as "Deep Research", "Deep Search", "Search", or "Research." Results are available on a public leaderboard at https://drb.futuresearch.ai/.

cs.AI↗

arXiv2024

Evaluating Forecasts with scoringutils in R

Evaluating forecasts is essential to understand and improve forecasting and make forecasts useful to decision makers. A variety of R packages provide a broad variety of scoring rules, visualisations and diagnostic tools. One particular challenge, which scoringutils aims to address, is handling the complexity of evaluating and comparing forecasts from several forecasters across multiple dimensions such as time, space, and different types of targets. scoringutils extends the existing landscape by offering a convenient and flexible data.table-based framework for evaluating and comparing probabilistic forecasts (forecasts represented by a full predictive distribution). Notably, scoringutils is the first package to offer extensive support for probabilistic forecasts in the form of predictive quantiles, a format that is currently used by several infectious disease Forecast Hubs. The package is easily extendable, meaning that users can supply their own scoring rules or extend existing classes to handle new types of forecasts. scoringutils provides broad functionality to check the data and diagnose issues, to visualise forecasts and missing data, to transform data before scoring, to handle missing forecasts, to aggregate scores, and to visualise the results of the evaluation. The paper presents the package and its core functionality and illustrates common workflows using example data of forecasts for COVID-19 cases and deaths submitted to the European COVID-19 Forecast Hub.

stat.ME↗

arXiv2024

Towards a Realistic Long-Term Benchmark for Open-Web Research Agents

We present initial results of a forthcoming benchmark for evaluating LLM agents on white-collar tasks of economic value. We evaluate agents on real-world "messy" open-web research tasks of the type that are routine in finance and consulting. In doing so, we lay the groundwork for an LLM agent evaluation suite where good performance directly corresponds to a large economic and societal impact. We built and tested several agent architectures with o1-preview, GPT-4o, Claude-3.5 Sonnet, Llama 3.1 (405b), and GPT-4o-mini. On average, LLM agents powered by Claude-3.5 Sonnet and o1-preview substantially outperformed agents using GPT-4o, with agents based on Llama 3.1 (405b) and GPT-4o-mini lagging noticeably behind. Across LLMs, a ReAct architecture with the ability to delegate subtasks to subagents performed best. In addition to quantitative evaluations, we qualitatively assessed the performance of the LLM agents by inspecting their traces and reflecting on their observations. Our evaluation represents the first in-depth assessment of agents' abilities to conduct challenging, economically valuable analyst-style research on the real open web.

cs.CL↗

arXiv2024

Assessing Human Judgment Forecasts in the Rapid Spread of the Mpox Outbreak: Insights and Challenges for Pandemic Preparedness

In May 2022, mpox (formerly monkeypox) spread to non-endemic countries rapidly. Human judgment is a forecasting approach that has been sparsely evaluated during the beginning of an outbreak. We collected -- between May 19, 2022 and July 31, 2022 -- 1275 forecasts from 442 individuals of six questions about the mpox outbreak where ground truth data are now available. Individual human judgment forecasts and an equally weighted ensemble were evaluated, as well as compared to a random walk, autoregressive, and doubling time model. We found (1) individual human judgment forecasts underestimated outbreak size, (2) the ensemble forecast median moved closer to the ground truth over time but uncertainty around the median did not appreciably decrease, and (3) compared to computational models, for 2-8 week ahead forecasts, the human judgment ensemble outperformed all three models when using median absolute error and weighted interval score; for one week ahead forecasts a random walk outperformed human judgment. We propose two possible explanations: at the time a forecast was submitted, the mode was correlated with the most recent (and smaller) observation that would eventually determine ground truth. Several forecasts were solicited on a logarithmic scale which may have caused humans to generate forecasts with unintended, large uncertainty intervals. To aide in outbreak preparedness, platforms that solicit human judgment forecasts may wish to assess whether specifying a forecast on logarithmic scale matches an individual's intended forecast, support human judgment by finding cues that are typically used to build forecasts, and, to improve performance, tailor their platform to allow forecasters to assign zero probability to events.

q-bio.PE↗

arXiv2022

Comparing trained and untrained probabilistic ensemble forecasts of COVID-19 cases and deaths in the United States

The U.S. COVID-19 Forecast Hub aggregates forecasts of the short-term burden of COVID-19 in the United States from many contributing teams. We study methods for building an ensemble that combines forecasts from these teams. These experiments have informed the ensemble methods used by the Hub. To be most useful to policy makers, ensemble forecasts must have stable performance in the presence of two key characteristics of the component forecasts: (1) occasional misalignment with the reported data, and (2) instability in the relative performance of component forecasters over time. Our results indicate that in the presence of these challenges, an untrained and robust approach to ensembling using an equally weighted median of all component forecasts is a good choice to support public health decision makers. In settings where some contributing forecasters have a stable record of good performance, trained ensembles that give those forecasters higher weight can also be helpful.

stat.ME↗