Search papers, labs, and topics across Lattice.
Affiliation:
6
0
10
1
Traditional probing methods fail to reveal the true memorization capabilities of large code LLMs, leading to inflated performance scores that obscure their genuine understanding.
Random exploration outperforms LLMs in TUI testing, revealing that model choice may not be as critical as previously thought.
Simple image transformations can undermine even the most advanced AI-based content moderation systems, exposing significant vulnerabilities.
PatchFusion recovers more bugs than any single source, outperforming traditional selection methods with a deterministic fusion of evidence that cuts costs dramatically.
Backdoor attacks in LLMs can be defused at inference time, without retraining or external data, by geometrically smoothing attention patterns to disrupt adversarial routing.
Code LLMs don't just memorize training data – some generalize far better than others, and even "leaky" datasets like CVEFixes show surprisingly low memorization advantage.
