Opening the AI black box: program synthesis via mechanistic interpretability

EJ Michaud, I Liao, V Lad, Z Liu, A Mudide… - arXiv preprint arXiv …, 2024 - arxiv.org
We present MIPS, a novel method for program synthesis based on automated mechanistic
interpretability of neural networks trained to perform the desired task, auto-distilling the …
被引用次数：11 相关文章

Opening the AI black box: program synthesis via mechanistic interpretability

EJ Michaud, I Liao, V Lad, Z Liu, A Mudide… - CoRR, 2024 - openreview.net
We present MIPS, a novel method for program synthesis based on automated mechanistic
interpretability of neural networks trained to perform the desired task, auto-distilling the …

Opening the AI black box: program synthesis via mechanistic interpretability

EJ Michaud, I Liao, V Lad, Z Liu, A Mudide… - arXiv e …, 2024 - ui.adsabs.harvard.edu
We present MIPS, a novel method for program synthesis based on automated mechanistic
interpretability of neural networks trained to perform the desired task, auto-distilling the …