Bugs in Deep Learning (DL) libraries may affect almost all downstream DL applications, and it is crucial to ensure the quality of such systems. It is challenging to generate valid input programs for fuzzing DL libraries, since the input programs need to satisfy both the syntax/semantics of the supported languages (e.g., Python) and the tensor/operator constraints for constructing valid computational graphs. Recently, the TitanFuzz work demonstrates that modern Large Language Models (LLMs) can be directly leveraged to implicitly learn all the language and DL computation constraints to generate valid programs for fuzzing DL libraries. However, LLMs tend to generate ordinary programs following similar patterns/tokens with typical programs seen in their massive training corpora (e.g., GitHub), while fuzzing favors unusual inputs that cover edge cases or are unlikely to be manually produced. To fill this gap, this paper proposes AtlasFuzz, the first technique to prime LLMs to synthesize unusual programs for fuzzing. AtlasFuzz is built on the well-known hypothesis that historical bug-triggering programs may include rare/valuable code ingredients important for bug finding. Meanwhile, while traditional techniques leveraging such historical information require intensive human efforts to both design dedicated generators and ensure the syntactic/semantic validity of generated programs, AtlasFuzz demonstrates that this process can be fully automated via the intrinsic capabilities of LLMs (including fine-tuning and in-context learning), while being generalizable and applicable to challenging domains. Moreover, AtlasFuzz also shows the potential of directly leveraging the instruct-following capability of the recent ChatGPT for effective fuzzing. The experimental study on two popular DL libraries (PyTorch and TensorFlow) shows that AtlasFuzz can substantially outperform TitanFuzz, detecting 76 bugs, including 48 already confirmed as previously unknown bugs.