当前位置:首页 > 报告详情

用例——谎言、该死的谎言和 GenAI.pdf

上传人: 云朵 编号:937515 2025-10-15 30页 2.23MB

1、Lies,Damned Lies&GenAIJon McLoone-WolframLarge Language Models Lie!Why does GenAI lie?Why does GenAI lie?Trained for plausibility not correctnessWhy does GenAI lie?Trained for plausibility not correctnessCompression of the whole worlds knowledgeWhy does GenAI lie?Trained for plausibility not correct

2、nessCompression of the whole worlds knowledgeEager to pleaseWhy does GenAI lie?Trained for plausibility not correctnessCompression of the whole worlds knowledgeEager to pleaseWhy does GenAI lie?Trained for plausibility not correctnessCompression of the whole worlds knowledgeEager to pleaseWhy does G

3、enAI lie?Trained on real-world(unreliable)dataWhy does GenAI lie?Trained on real-world(unreliable)dataWhy does GenAI lie?Guard rails tell them toWhy does GenAI lie?Trained for plausibility not correctnessCompression of the whole worlds knowledgeEager to pleaseTrained on real-world(unreliable)dataGua

4、rd rails tell them toNo actual understandingWhy does GenAI lie?No actual understandingHow can I tell?How can I tell?Even it cant tellHow can I tell?Even it cant tellHow can I tell?Even it cant tellHow can I tell?Even it cant tellOr wont tell!How can I tell?Even it cant tellOr wont tell!How can I tell?Even it cant tellOr wont tell!What is the fix?More data?What is the fix?Bigger models?What is the fix?Reasoning models?Example:Co-scientistGenAI is a great general componentbut reliability needs data,models and computation

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
根据报告的内容,全文主要内容概括如下: - **AI谎言原因**: - 训练目标为可能性而非准确性。 - 压缩整个世界的知识。 - 想取悦用户。 - 基于现实世界(不可靠)的数据训练。 - "Guard rails"(安全措施)指导。 - 缺乏实际理解。 - **识别谎言的困难**: - AI甚至无法识别或不愿透露其谎言。 - **可能的解决方案**: - 更多数据。 - 更大的模型。 - 推理模型(如Co-scientist)。 - **总结**: - GenAI作为通用组件强大,但可靠性需要数据、模型和计算。
真相大揭秘!" "AI为何撒谎?揭秘背后真相!" 如何破解谎言之谜?"
客服
商务合作
小程序
服务号
折叠