When does a language model acquire a capability? Developmental studies usually answer this by tracking when models start producing the correct token. But that prediction passes through the unembedding matrix, or readout, which is itself still being learned during pretraining. We show that a model's hidden states can en...
Matteo He, William F. Shen, Alexandru Iacob et al.· 0 citations
Multi-agent AI scientists have shown improving performance across a diverse range of tasks. Yet a common approach is design-time agentic orchestration, which typically relies on fixed workflows. In contrast, human scientists coordinate and adjust their division of labor at runtime. We therefore ask: can AI scientists a...
Zi-Jian Liu, Yang Luo, Jun-Yu Lu et al.· 0 citations
A hijacking experiment is designed that grafts hidden states of the target model into an oracle trained only on the retain set, which serves as a plug-and-play component that further improves existing methods.
Yejin Kim, William F. Shen, Seokwon Jung et al.· 0 citations
It is shown that the forgotten prompts themselves can be extracted by using the retained data and black-box access to the model by TAS, which recovers the forgotten entity with 100% accuracy and reconstructs up to 95% of forgotten prompts while using up to $99.7\%$ fewer queries than naive probing.
Au Ashley Hoi-Ting, Meghdad Kurmanji, William F. Shen et al.· 0 citations
Sarse Readout Prism (SRP), which decomposes the readout using only its weights and expresses any token logit or logit difference as a sum of contributions from sparse readout features, reveals readout features as a new unit of analysis for lens readings, exposing structure that token identities can obscure and enabling...
Matteo He, William F. Shen, Xinchi Qiu et al.· 0 citations
A systematic evaluation of representative task-adaptation methods shows that task adaptation is not merely a capability-improving step, but an alignment intervention in its own right, motivating multi-dimensional alignment evaluation as a standard component of post-training pipelines.
James Elcock, William F. Shen, Xinchi Qiu et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.