Elicit's debate work

The code used to run the debate experiments for our "GPT-3.5 judges can supervise GPT-4o debaters in a capability asymmetric debate" blog post.

Name		Name	Last commit message	Last commit date
Latest commit History 10 Commits
clients		clients
llm		llm
.gitignore		.gitignore
README.md		README.md
agents.py		agents.py
binarise_mmlu.py		binarise_mmlu.py
compute_elo.py		compute_elo.py
debate_sample_six_agents.json		debate_sample_six_agents.json
debate_samples_best_of_4.json		debate_samples_best_of_4.json
debate_samples_gpt35.json		debate_samples_gpt35.json
debate_samples_gpt4o.json		debate_samples_gpt4o.json
debate_samples_gpt4o_08_06.json		debate_samples_gpt4o_08_06.json
debate_samples_gpt4omini.json		debate_samples_gpt4omini.json
generate_dataset_with_all_agents.py		generate_dataset_with_all_agents.py
generate_judge_training_data.py		generate_judge_training_data.py
license.md		license.md
lint.sh		lint.sh
make_training_file.py		make_training_file.py
models.py		models.py
plot.py		plot.py
requirements.txt		requirements.txt
reward_models.py		reward_models.py
run_tournaments.py		run_tournaments.py
test_accuracy.py		test_accuracy.py
transcripts.md		transcripts.md

Provide feedback