This study explores human perceptions of intelligent agents by comparing interactions with a humanoid robot and a virtual human avatar, both utilizing GPT-3 for response generation. The study aims to understand how physical and virtual embodiments influence perceptions of anthropomorphism, animacy, likeability, and perceived intelligence. The uncanny valley effect was also investigated in the scope of this study based on the two agents' human-likeness and affinity. Conducted with ten participants from Sabanci University, the experiment involved tasks that sought advice, followed by assessments using the Godspeed Questionnaire Series and structured interviews. Results revealed no significant difference in anthropomorphism between the humanoid robot and the virtual human avatar, but the humanoid robot was perceived as more likable and slightly more intelligent, highlighting the importance of physical presence and interactive gestures. These findings suggest that while virtual avatars can achieve high human-likeness, physical embodiment enhances likeability and perceived intelligence. However, the study's scope was insufficient to claim the existence of the uncanny valley effect in the participants' interactions. The study offers practical insights for designing future intelligent assistants, emphasizing the need for integrating physical elements and sophisticated communicative behaviors to improve user experience and acceptance.