🧩 Philosophy 6d ago · Ram Potham

Scaffolding vs Reinforcement Finetuning for AI Forecasting

Less Wrong
View Channel →
Scaffolding vs Reinforcement Finetuning for AI Forecasting
Source ↗ 👁 0 💬 0
Epistemic status: low-medium confidence in results, this is work I did last year and has a low sample size. However I think the takeaways are still accurate.I built a forecasting bot using OpenAI’s Reinforcement Finetuning and a multi-agent architecture, then tested it against simpler baselines in a metaculus tournament. The aggregate scores favored the baseline, but when I broke down results by question type, the finetuned model outperformed on numeric questions (average +14.59 vs +9.25 using M

Comments (0)

Sign in to join the discussion

More Like This

📰
Idealism
Stanford Encyclopedia of Philosophy · 7h ago
Toward a Better Evaluations Ecosystem
LessWrong · 10h ago
Model Spec Midtraining: Improving How Alignment Training Generalizes
LessWrong · 10h ago
📰
Positive Feedback Only
LessWrong · 11h ago
📰
What if LLMs are mostly crystallized intelligence?
LessWrong · 11h ago
Decision theory doesn’t prove that useful strong AIs will doom us all
LessWrong · 11h ago