作者认为，期望生成式AI能够准确地代表世界是错误的，因为这些模型是在人类生成的文本上训练的，而这些文本本质上是不一致和选择性的。这种情况被比作柏拉图的洞穴寓言，暗示了与现实的脱节。 AI
影响 强调了当前AI模型因其训练数据而存在的固有局限性，并暗示需要更强大的方法来代表现实。
排序理由 该条目是一篇表达对AI局限性观点的评论文章。
在 Mastodon — sigmoid.social 阅读 →
AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →
AI的世界观因不一致的人类训练数据而存在缺陷
报道来源 [1]
-
Expecting Generative AI to be able to talk reliably about the world, when it is only trained on what humans have inconsistently elected to say about the world;
Expecting Generative AI to be able to talk reliably about the world, when it is only trained on what humans have inconsistently elected to say about the world; is madness. All of modernity is kind of becoming a sad digital retelling of Plato's allegory of the cave I guess. # GenA…