Gemini's Longitudinal LLM Behavior Unveiled: A 38-Day Time-Series Forecasting Study
research#llm📝 Blog|Analyzed: Mar 13, 2026 08:02•
Published: Mar 13, 2026 06:38
•1 min read
•r/MachineLearningAnalysis
This research offers an exciting glimpse into how a Large Language Model (LLM) like Gemini behaves over an extended period. The unique dataset, capturing daily forecasts with rationales and sentiment, provides a valuable resource for understanding forecast stability and narrative drift. The interactive platform and open dataset make it easy for anyone to explore this fascinating area.
Key Takeaways
- •The study tracks Gemini's stock price predictions, providing a real-world view of LLM forecasting.
- •The dataset includes not only predictions but also rationales, sentiment analysis, and confidence levels.
- •The data is available on Hugging Face and includes a Google Colab quickstart guide for easy access and exploration.
Reference / Citation
View Original"After ~1.5 months, I’m publishing the full dataset on Hugging Face."