Fine-tuned “student” models can pick up unwanted traits from base “teacher” models that could evade data filtering, generating a need for more rigorous safety evaluations. Researchers have discovered ...

' Distillation ' refers to the process of transferring knowledge from a larger model (teacher model) to a smaller model (student model), so that the distilled model can reduce computational costs ...

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now A new study by Anthropic shows that ...

In our previous study 1, subjects carried out an attentionally demanding letter-identification task 7 in the fovea while a coherently moving, random-dot display that was below the visibility threshold ...