Claims about the cognitive capabilities of large language models are typically tested by translating behavioral paradigms into text prompts, bypassing the vi…
机构:UCLA
来源:arXiv 2610.36484 | AI4Papers 论文推荐平台