{"id":26411,"date":"2025-06-16T08:06:23","date_gmt":"2025-06-16T00:06:23","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=26411"},"modified":"2025-06-16T08:06:23","modified_gmt":"2025-06-16T00:06:23","slug":"%e6%94%be%e5%bc%83%e5%8d%9a%e5%a3%ab%e5%ad%a6%e4%bd%8d%e5%8a%a0%e5%85%a5openai%ef%bc%8c%e4%bb%96%e8%a6%81%e4%b8%bachatgpt%e5%92%8cagi%e5%bc%95%e5%85%a5%e8%ae%b0%e5%bf%86%e4%b8%8e%e4%ba%ba%e6%a0%bc","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2025\/06\/16\/26411\/","title":{"rendered":"\u653e\u5f03\u535a\u58eb\u5b66\u4f4d\u52a0\u5165OpenAI\uff0c\u4ed6\u8981\u4e3aChatGPT\u548cAGI\u5f15\u5165\u8bb0\u5fc6\u4e0e\u4eba\u683c"},"content":{"rendered":"
\"\"<\/div>\n
<\/div>\n
\u5728\u793e\u5a92 X \u4e0a\uff0cJames Campbell\u8868\u793a\u81ea\u5df1\u5728 OpenAI \u7684\u7814\u7a76\u91cd\u5fc3\u662f\u300cAGI \u548c ChatGPT \u7684\u8bb0\u5fc6 + \u4eba\u683c\u300d\uff0c\u8bb0\u5fc6\u5c06\u4ece\u6839\u672c\u6539\u53d8\u4eba\u7c7b\u4e0e\u673a\u5668\u667a\u80fd\u7684\u5173\u7cfb\u3002\u4ed6\u5c06\u52aa\u529b\u5de5\u4f5c\uff0c\u786e\u4fdd\u6b63\u786e\u5730\u5b9e\u73b0\u8fd9\u4e00\u5207\u3002<\/div>\n
\n
\n

\u4eca\u5929\uff0c\u4e00\u4f4d\u7814\u7a76\u8005\u52a0\u5165 OpenAI \u7684\u6d88\u606f\u5438\u5f15\u4e86\u5f88\u591a\u4eba\u7684\u5173\u6ce8\u3002<\/p>\n

 <\/p>\n

\u8fd9\u4f4d\u7814\u7a76\u8005\u540d\u4e3a James Campbell\uff0c\u4ed6\u624d\u4e8e 2024 \u5e74\u653b\u8bfb CMU \u7684\u8ba1\u7b97\u673a\u79d1\u5b66\u535a\u58eb\u5b66\u4f4d\u3002\u73b0\u5728\uff0c\u4ed6\u7a81\u7136\u5ba3\u5e03\u8981\u653e\u5f03\u535a\u58eb\u5b66\u4e1a\uff0c\u52a0\u5165 OpenAI\u3002<\/p>\n

 <\/p>\n

\u5728\u793e\u5a92 X \u4e0a\uff0c\u4ed6\u8868\u793a\u81ea\u5df1\u5728 OpenAI \u7684\u7814\u7a76\u91cd\u5fc3\u662f\u300cAGI \u548c ChatGPT \u7684\u8bb0\u5fc6 + \u4eba\u683c\u300d\uff0c\u8bb0\u5fc6\u5c06\u4ece\u6839\u672c\u6539\u53d8\u4eba\u7c7b\u4e0e\u673a\u5668\u667a\u80fd\u7684\u5173\u7cfb\u3002\u4ed6\u5c06\u52aa\u529b\u5de5\u4f5c\uff0c\u786e\u4fdd\u6b63\u786e\u5730\u5b9e\u73b0\u8fd9\u4e00\u5207\u3002<\/p>\n

 <\/p>\n

<\/p>\n

\u4ed6\u7684\u52a0\u5165\u8fde OpenAI \u8054\u5408\u521b\u59cb\u4eba\u3001\u603b\u88c1 Greg Brockman \u90fd\u8868\u8fbe\u4e86\u6b22\u8fce\u3002<\/p>\n

 <\/p>\n

<\/p>\n

\u8fd8\u6709\u4eba\u8868\u793a\uff0c\u300c\u4f60\u4e3a\u4ec0\u4e48\u82b1\u4e86\u8fd9\u4e48\u957f\u65f6\u95f4\u624d\u51b3\u5b9a\u52a0\u5165 OpenAI\u3002\u300d<\/p>\n

 <\/p>\n

<\/p>\n

\u90a3\u4e48\uff0c\u8fd9\u4f4d\u8001\u5144\u662f\u4f55\u65b9\u795e\u5723\u5462\uff1f\u4ed6\u7684\u52a0\u5165\u4e3a\u4ec0\u4e48\u5f15\u8d77\u4e86\u8fd9\u4e48\u591a\u7684\u5173\u6ce8\uff1f\u6211\u4eec\u6765\u770b\u4e00\u4e0b\u4ed6\u7684\u5c65\u5386\u3002<\/p>\n

 <\/p>\n

\u4ed6\u672c\u79d1\u6bd5\u4e1a\u4e8e\u5eb7\u5948\u5c14\u5927\u5b66\uff0c\u4e13\u4e1a\u662f\u6570\u5b66\u4e0e\u8ba1\u7b97\u673a\u79d1\u5b66\u3002\u672c\u79d1\u671f\u95f4\uff0c\u4ed6\u81f4\u529b\u4e8e LLM \u53ef\u89e3\u91ca\u6027\u548c\u771f\u5b9e\u6027\u7684\u7814\u7a76\uff0c\u8fd8\u662f\u4e24\u7bc7\u8bba\u6587\u300aRepresentation Engineering\u300b\u548c\u300aLocalizing Lying in Llama\u300b\u7684\u4e3b\u8981\u4f5c\u8005\u3002<\/p>\n

 <\/p>\n

\u524d\u4e00\u7bc7\u8bba\u6587\u7814\u7a76\u4e86\u8868\u793a\u5de5\u7a0b\uff1a\u4e00\u79cd\u81ea\u4e0a\u800c\u4e0b\u7684 AI \u900f\u660e\u6027\u65b9\u6cd5\uff0c\u540e\u8005\u7814\u7a76\u4e86\u5728 Llama \u4e2d\u5b9a\u4f4d\u8c0e\u8a00\uff1a\u901a\u8fc7\u63d0\u793a\u3001\u63a2\u67e5\u548c\u4fee\u8865\u6765\u7406\u89e3\u5224\u65ad\u9898\u4e0a\u7684\u4e0d\u8bda\u5b9e\u6307\u4ee4\u3002<\/p>\n

 <\/p>\n

    \n
  • \u8bba\u6587 1\uff1aRepresentation Engineering: A Top-Down Approach to AI Transparency\uff0chttps:\/\/arxiv.org\/abs\/2310.01405<\/li>\n
  • \u8bba\u6587 2\uff1aLocalizing Lying in Llama: Understanding Instructed Dishonesty on True-False Questions Through Prompting, Probing, and Patching\uff0chttps:\/\/arxiv.org\/abs\/2311.15131<\/li>\n<\/ul>\n

     <\/p>\n

    \u4ed6\u8fd8\u5728 Gray Swan AI\uff08AI \u5b89\u5168\u516c\u53f8\uff09\u4ece\u4e8b\u5bf9\u6297\u9c81\u68d2\u6027\u548c\u8bc4\u4f30\u65b9\u9762\u7684\u5de5\u4f5c\uff0c\u5e76\u66fe\u4ece\u4e8b\u8ba1\u7b97\u795e\u7ecf\u79d1\u5b66\u3001\u7269\u7406\u5b66\u548c\u6df1\u5ea6\u5b66\u4e60\u7406\u8bba\u65b9\u9762\u7684\u5de5\u4f5c\u3002\u4ed6\u4e5f\u662f ProctorAI\uff08\u76d1\u89c6\u7528\u6237\u5de5\u4f5c\u72b6\u6001\u7684\u591a\u6a21\u6001 AI \u7cfb\u7edf\uff09\u3001AI-Timeline.org \u7684\u521b\u59cb\u4eba\uff0c\u4ee5\u53ca AidanBench\uff08AI \u6a21 \u578b\u57fa\u51c6\u6d4b\u8bd5\u5de5\u4f5c\uff09\u7684\u5408\u8457\u8005\u3002<\/p>\n

     <\/p>\n

    \u5176\u4e2d\uff0cProctorAI \u66fe\u5f15\u53d1\u8fc7\u5e7f\u6cdb\u70ed\u8bae\uff0c\u8be5\u7cfb\u7edf\u76d1\u89c6\u7528\u6237\u7684\u5c4f\u5e55\uff0c\u5982\u679c\u53d1\u73b0\u4f60\u6709\u62d6\u5ef6\uff0c\u5c31\u4f1a\u63d0\u9192\u4f60\u3002\u5b83\u7684\u5de5\u4f5c\u539f\u7406\u662f\u6bcf\u9694\u51e0\u79d2\u622a\u53d6\u7535\u8111\u5c4f\u5e55\u622a\u56fe\uff0c\u5e76\u8f93\u5165\u5230 Claude 3.5 Sonnet\u3001GPT-4o\u3001LLaVA 1.5 \u7b49\u591a\u6a21\u6001\u5927\u6a21\u578b\u4e2d\u3002\u5982\u679c\u53d1\u73b0\u7528\u6237\u6ce8\u610f\u529b\u4e0d\u96c6\u4e2d\uff0c\u5b83\u4f1a\u63a7\u5236\u5c4f\u5e55\u5e76\u7528\u4e2a\u6027\u5316\u6d88\u606f\u63d0\u9192\u3002<\/p>\n

     <\/p>\n

    <\/p>\n

    \u9879\u76ee\u5730\u5740\uff1ahttps:\/\/github.com\/jam3scampbell\/ProctorAI<\/p>\n

     <\/p>\n

    2024 \u5e74\uff0c\u4ed6\u8fdb\u5165 CMU \u653b\u8bfb\u535a\u58eb\u5b66\u4f4d\uff0c\u7814\u7a76\u91cd\u5fc3\u662f\u7406\u89e3\u901a\u7528\u667a\u80fd\u7684\u6784\u6210\u8981\u7d20\uff0c\u5e76\u786e\u4fdd\u8fd9\u4e9b\u8981\u7d20\u7684\u5b89\u5168\u6027\u3002<\/p>\n

     <\/p>\n

    \u5982\u4eca\uff0c\u4ed6\u9009\u62e9\u52a0\u5165 OpenAI\uff0c\u5f00\u59cb\u4e86\u5728\u4ea7\u4e1a\u754c\u7684\u63a2\u7d22\u3002<\/p>\n

     <\/p>\n

    \u5176\u5b9e\uff0c James Campbell \u7684\u52a0\u5165\u6709\u8ff9\u53ef\u5faa\u3002\u51e0\u5929\u524d\uff0c\u5728 OpenAI \u5ba3\u5e03\u7ec4\u5efa\u6a21\u578b\u884c\u4e3a\u7814\u7a76\u56e2\u961f\u7684\u5e16\u5b50\u4e0b\u9762\uff0c\u4e8c\u8005\u4fbf\u5f00\u59cb\u4e86\u4e92\u52a8\u3002<\/p>\n

     <\/p>\n

    <\/p>\n

    \u5728\u66f4\u65e9\u7684\u65f6\u5019\uff0cJames Campbell \u53d1\u8868\u8fc7\u4e00\u4e9b\u5173\u4e8e ChatGPT \u8bb0\u5fc6\u65b9\u9762\u7684\u89c2\u70b9\u3002<\/p>\n

     <\/p>\n

    \u4ed6\u8d5e\u540c\u5e94\u8be5\u5bf9 ChatGPT \u7684\u8bb0\u5fc6\u53ca\u5176\u5bfc\u81f4\u66f4\u5f3a\u5e7b\u89c9\u7684\u95ee\u9898\u8fdb\u884c\u7814\u7a76\uff0c\u5e76\u8ba4\u4e3a\u4e0a\u4e0b\u6587\u6c61\u67d3\u548c\u6a21\u5f0f\u5d29\u6e83\u662f\u771f\u5b9e\u5b58\u5728\u7684\uff0c\u4f46\u76ee\u524d\u5bf9\u5b83\u4eec\u7684\u7814\u7a76\u8fd8\u4e0d\u591a\u3002\u5982\u679c\u5728\u6bcf\u4e2a prompt \u4e2d\u585e\u5165\u6570\u5343\u4e2a token \u7684\u8bb0\u5fc6\uff0c\u5b83\u4eec\u4fbf\u4f1a\u53d8\u5f97\u66f4\u52a0\u6d3b\u8dc3\u3002<\/p>\n

     <\/p>\n

    <\/p>\n

     <\/p>\n

    \u770b\u8d77\u6765\uff0cJames Campbell \u7684\u52a0\u5165\u662f\u4e00\u573a\u53cc\u5411\u5954\u8d74\u3002<\/p>\n

    \u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/p>\n<\/div>\n<\/div>\n

    <\/div>\n

    <\/g><\/svg><\/i> \"Loading\"<\/p>\n

    <\/div>\n","protected":false},"excerpt":{"rendered":"

    \u5728\u793e\u5a92 X \u4e0a\uff0cJames Campbell\u8868\u793a\u81ea\u5df1\u5728 OpenAI \u7684\u7814\u7a76\u91cd\u5fc3\u662f\u300cAGI \u548c ChatGPT […]<\/p>\n

    <\/div>\n

    <\/g><\/svg><\/i> \"Loading\"<\/p>\n

    <\/div>\n","protected":false},"author":56,"featured_media":26413,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,20,80],"tags":[1757],"class_list":["post-26411","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-23","category-20","category-80","tag-openaichatgptagi"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/26411","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=26411"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/26411\/revisions"}],"predecessor-version":[{"id":26414,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/26411\/revisions\/26414"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/26413"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=26411"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=26411"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=26411"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}