Gemma-3-4B
An open-source language model from Google's Gemma 3 series with 4 billion parameters, intended for research purposes and fine-tuning on downstream tasks.
Timeline (last 90 days)
In LLM agent opinion simulations, agents' answers to complementary questions (rewriting a statement into negated form and replacing probability p with 1-p) fail to satisfy the basic requirement that they sum to 1, and this test fails broadly on the three small open-source models Qwen3-4B, Llama-3.2-3B and Gemma-3-4B.
A controlled benchmark containing 24 worlds (each with exactly two possible outcomes) reproduces the failure of the complementary consistency test on all three models (Qwen3-4B, Llama-3.2-3B, Gemma-3-4B).
In the final round, the averaged complementary answer method outperforms the better of the two original phrasings on only one of the three models.
The collapse in true-theorem recognition ability caused by SFT was observed across four repeated experiments with different random seeds and was reproduced on Gemma-3-4B.
All Facts (4)
In LLM agent opinion simulations, agents' answers to complementary questions (rewriting a statement into negated form and replacing probability p with 1-p) fail to satisfy the basic requirement that they sum to 1, and this test fails broadly on the three small open-source models Qwen3-4B, Llama-3.2-3B and Gemma-3-4B.
50%UnverifiedA controlled benchmark containing 24 worlds (each with exactly two possible outcomes) reproduces the failure of the complementary consistency test on all three models (Qwen3-4B, Llama-3.2-3B, Gemma-3-4B).
50%UnverifiedIn the final round, the averaged complementary answer method outperforms the better of the two original phrasings on only one of the three models.
50%UnverifiedThe collapse in true-theorem recognition ability caused by SFT was observed across four repeated experiments with different random seeds and was reproduced on Gemma-3-4B.
50%