Search papers, labs, and topics across Lattice.
3
0
6
4
Even the largest open-weight LLMs fail to generate accurate cybersecurity CLI commands over 58% of the time, but fine-tuning with verifiable rewards on KaliBench closes the gap with massive MoE models.
Leading open-source MLLMs achieve an alarming 61.9% average accuracy on fine-grained fire and smoke reasoning, but adapting vision encoders on just 7% targeted domain data more than triples classification performance.
Language-dependent biases in text-to-image generation reveal significant gaps in model performance across different cultures and languages.
