One and Only at SemEval-2026 Task 2: Evaluating Zero-Shot Autonomous LLM Agents and Heuristic Proxies in Ecological Affect Forecasting

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing, Data Science & Analytics · Depth: Expert, quick

Summary

Team One and Only's system for SemEval-2026 Task 2 investigated replacing fine-tuning with zero-shot LLM reasoning for ecological affect forecasting. Their approach combined deterministic statistical priors with frozen LLMs, specifically Gemini 3 Pro, Claude Opus 4.6, and GPT-5.2. For short-term state changes (Subtask 2A), an OLS mean-reversion anchor was paired with LLM-generated impulses. For long-term disposition changes (Subtask 2B), a Chain-of-Thought (CoT) prompt drove direct numeric prediction. The system underperformed fine-tuned methods. However, post-submission ablation revealed that CoT reasoning substantially improved disposition forecasting (rV: -0.185 to +0.129; MAEV: 0.899 to 0.422). Conversely, uncalibrated LLM impulses degraded state-change prediction due to variance collapse (σpred = 0.41 vs. σgold = 1.73).

Key takeaway

For Machine Learning Engineers exploring zero-shot LLM applications in affect forecasting, you should carefully consider the task's temporal scope. While Chain-of-Thought prompting can significantly enhance long-term disposition predictions, uncalibrated LLM impulses may degrade short-term state-change forecasts due to variance collapse. Prioritize robust calibration for impulse-based models and evaluate CoT for complex, multi-step reasoning tasks to avoid performance pitfalls.

Key insights

Zero-shot LLM reasoning combined with statistical priors can be applied to ecological affect forecasting, with task-dependent performance.

Principles

• Chain-of-Thought reasoning significantly improves disposition forecasting.
• Uncalibrated LLM impulses can degrade state-change prediction by causing variance collapse.

Method

Combine OLS mean-reversion with LLM-generated impulses for short-term state changes, and use Chain-of-Thought prompts for direct numeric prediction of long-term disposition changes.

In practice

• Evaluate Chain-of-Thought for improving long-term disposition predictions.
• Be cautious of uncalibrated LLM impulses causing variance collapse in short-term state predictions.

Topics

• SemEval-2026
• Zero-Shot LLMs
• Affect Forecasting
• Chain-of-Thought
• Emotional Valence
• Arousal Prediction
• Large Language Models

Best for: Research Scientist, AI Engineer, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Analysis of the Neglect-Zero Effect in Large Language Models — LLMs may not share the human "neglect-zero effect" cognitive bias regarding empty sets.
• Where Do CoT Training Gains Land in LLM based Agents? — CoT training improves LLM agents' direct action prediction from prompts, with less impact on CoT's relative reasoning advantage.
• Nous: An Attempt to Extract and Inject the Cognition Behind Prediction-Market Behavior — LLM agent ensembles face cognitive monoculture; prompt-level injection of human behavioral profiles fails to induce diversity.
• Zero-Shot Goal Recognition with Large Language Models — LLM goal recognition competence varies significantly, with some models integrating evidence effectively while others rely on world-knowledge priors.
• Tyler: Typed Latent Reasoning for Language Models -- When to Think, What to Compute, and How Much to Allocate — Tyler dynamically manages latent computation in LLMs, deciding when, what type, and how much to allocate for improved reasoning.
• Modeling Pathology-Like Behavioral Patterns in Language Models Through Behavioral Fine-Tuning — Consistent behavioral fine-tuning can induce stable, pathology-like biases in LLM generative distributions and action selection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.