Language Models for World Event Forecasting
Cohere · youtube · 2026-07-18
This is a Cohere Labs technical talk addressing the question: can language models predict world events?
Speaker Shashwat Goel presents work from the OpenForecaster project, highlighting:
- The definition of forecasting and why it matters
- Evaluation pitfalls: avoiding overestimating model capabilities in prediction tasks
- Leakage-free retrieval: retrieving news without leaking future information
- Training data pipelines: post-training models on daily news for better-calibrated predictions
- RL training results: improving forecasting performance via reinforcement learning
- FutureSim: testing if frontier agents can continuously learn and update predictions when receiving new information by "replaying historical world events"
- Future directions like multi-agent effects and continuous learning visions
The talk is research- and methodology-heavy rather than a product pitch, focusing on systematically measuring and improving LLM forecasting abilities for real-world events.
More from AGI Musings
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22
- AI Power Demand Exposes US Energy Gap, Urging Shift from Scarcity to Abundance — bradneuberg · 2026-07-22
- ControlAI CEO says an international ban on superintelligence is needed to avert extinction risk — zetalyrae · 2026-07-22
- Gary Marcus says LLMs still cannot really do math on their own — GaryMarcus · 2026-07-22
- Gary Marcus says LLM math skills are like knowing only a car’s engine size — GaryMarcus · 2026-07-22
- AI may make digital work infinitely leveraged while offline life gets more human — illscience · 2026-07-22