Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B · HackerPapers discussions

Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B — mentioned by xianshou

Thu, 05 Sep 2024 21:06:04 GMT

Crazy how simple the technique is if this holds up. Just <think> and <reflection> plus synthetic data, used to finetune Llama 3.1 70B. Note that there's a threshold for how smart the model has to be to take advantage of this flow (https://x.com/mattshumer_/status/1831775436420083753) - 8B is too dumb. In which case, what happens if you apply this to a GPT-4o finetune, or to Claude 3.5 Sonnet? What happens if you combine it with variants of tree-based reasoning? With AlphaProof (https://www.nature.com/articles/s41586-023-06747-5#Sec3)? With MCTSr (https://arxiv.org/abs/2406.07394)?

Paper and discussion history. A mention is not an endorsement.

Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B — HN submission

Fri, 02 Aug 2024 03:40:53 GMT

Ask HN: Research Papers on Prompt Engineering?

Since prompting is such an important topic, I'm looking for a list of papers on how to improve prompts (around a year old, but preferably less). So far, I've found two: - "Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models" (https://arxiv.org/pdf/2305.04091v3) - "Accessing GPT-4 Level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B: A Technical Report" (https://arxiv.org/pdf/2406.07394v2) Do you recommend any other papers?

1 point; 2 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B — mentioned by cgearhart

Fri, 14 Jun 2024 21:58:40 GMT

[1] applied AlphaZero style search with LLMs to achieve performance comparable to GPT-4 Turbo with a llama3-8B base model. However, what's missing entirely from the paper (and the subject article in this thread) is that tree search is massively computationally expensive. It works well when the value function enables cutting out large portions of the search space, but the fact that the LLM version was limited to only 8 rollouts (I think it was 800 for AlphaZero) implies to me that the added complexity is not yet optimized or favorable for LLMs. [1] https://arxiv.org/abs/2406.07394

Paper and discussion history. A mention is not an endorsement.

Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B — HN submission

Wed, 12 Jun 2024 15:38:45 GMT

Accessing Math Solutions via Monte Carlo Self-Refine with LLaMa-3 8B

100 points; 9 discussion replies on HN. Paper and source history. A submission is not an endorsement.