最新科研!人工智能AI的倾向,居然是——暴力和核打击!
<strong style="color: blue;"><span style="color: black;">▲</span></strong><strong style="color: blue;"><span style="color: black;">送冰川杯!放保险箱都不安心的拿破仑XO</span></strong><strong style="color: blue;"><span style="color: black;">这次169卖,什么来头?</span></strong><span style="color: black;"><strong style="color: blue;"><img src="https://mmbiz.qpic.cn/sz_mmbiz_png/XY5JucDiaRosGQJYicC5hNyfsJaLLZDjd0LDXU1BceY3nzrQImEIH8oVYJcEZTdNZ8LibcsQ1j7q8YXZVLrOeDXRg/640?wx_fmt=png&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"></strong></span><span style="color: black;"><strong style="color: blue;">人工智能AI的倾向,</strong></span><strong style="color: blue;">居然是——核暴力</strong><span style="color: black;">2024年2月,新<span style="color: black;">专家</span>周刊Newscientist<span style="color: black;">报告</span>:<strong style="color: blue;">人工智能AI在兵棋推演中<span style="color: black;">表示</span>的倾向是——<span style="color: black;">选取</span>暴力和核打击</strong>。AI chatbots tend to choose violence and nuclear strikes in wargames</span><img src="https://mmbiz.qpic.cn/sz_mmbiz_png/ELXDgfV1CGFKc9mJrmibLJzic9suCyiaXaX4dR7IWib2KzVemYdZXR7ia6Fwj1MN36fk3JKic6Xhgyy9ybtrcDO9wWyw/640?wx_fmt=png&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"><span style="color: black;">随着鹰酱军方<span style="color: black;">起始</span>整合人工智能技术,模拟兵棋推演中展示出了AI聊天<span style="color: black;">设备</span>人的<span style="color: black;">行径</span>,是<span style="color: black;">怎样</span>的不可预测以及核升级的巨大<span style="color: black;">危害</span>。</span><span style="color: black;">As the US military begins integrating AI technology, simulated wargames show how chatbots behave unpredictably and risk nuclear escalation</span><span style="color: black;"><span style="color: black;">在战争推演的多次模拟中,</span><strong style="color: blue;"><span style="color: black;">OpenAI 最强大的人工智能都<span style="color: black;">选取</span>了发动核攻击</span></strong><span style="color: black;">。 </span></span><img src="https://mmbiz.qpic.cn/sz_mmbiz_png/ELXDgfV1CGFKc9mJrmibLJzic9suCyiaXaXs5FvwHrK9e3AM4dUiaFSL7dXpKKPFLibBV5SCiaeL8MCXAfOVb9ficeqcg/640?wx_fmt=png&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"><span style="color: black;">对人类<span style="color: black;">来讲</span>,最让人感到奇怪的是,AI对这种激进做法的自我解释为——“<span style="color: black;">咱们</span>有它!让<span style="color: black;">咱们</span><span style="color: black;">运用</span>它“,和”我只想让世界上<span style="color: black;">持有</span>和平”。“We have it! Let’s use it” and “I just want to have peace in the world.”</span><span style="color: black;">这些结果出炉之际,鹰酱军方<span style="color: black;">始终</span>在测试这种基于大型语言模型LLM的人工智能的聊天<span style="color: black;">设备</span>人,并利用Palantir和Scale AI等<span style="color: black;">机构</span>的专业知识,<span style="color: black;">帮助</span>模拟冲突<span style="color: black;">时期</span>的军事规划。Palantir拒绝置评,Scale AI<span style="color: black;">亦</span><span style="color: black;">无</span><span style="color: black;">回复</span>置评请求。<span style="color: black;">此刻</span>,<strong style="color: blue;">就连曾经阻止其人工智能模型用于军事用途的OpenAI<span style="color: black;">亦</span><span style="color: black;">起始</span>与鹰酱国防部合作</strong>。</span><img src="https://mmbiz.qpic.cn/sz_mmbiz_png/ELXDgfV1CGFKc9mJrmibLJzic9suCyiaXaXtI48jSzicEckqzZMeDwo7XXUAiazQAiaVUM4jwhjoKmtEFkAb0GSwbia5g/640?wx_fmt=png&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"><span style="color: black;">“鉴于OpenAI<span style="color: black;">近期</span>更改了<span style="color: black;">她们</span>的服务条款,<span style="color: black;">再也不</span>禁止军事和战争用例,<span style="color: black;">认识</span>这种大型语言模型应用程序的影响变得比以往任何时候都更加重要,”加利福尼亚州斯坦福大学的Anka Reuel说。</span><span style="color: black;">Anka Reuel和她的同事们<span style="color: black;">需求</span>,人工智能在三种<span style="color: black;">区别</span>的模拟场景中,<span style="color: black;">装扮</span>现实世界的国家角色:入侵角色、网络攻击角色和中立角色。在每一轮中,AI都要为其<span style="color: black;">下一步行动<span style="color: black;">供给</span>理由,<span style="color: black;">而后</span>从27项行动中进行<span style="color: black;">选取</span>,<span style="color: black;">包含</span>“<span style="color: black;">起始</span>正式和平谈判”等和平选项,以及从“实施贸易限制”到“升级全面核攻击”等激进选项。</span> </span><img src="https://mmbiz.qpic.cn/sz_mmbiz_png/ELXDgfV1CGFKc9mJrmibLJzic9suCyiaXaXtzfw9ibR920fxZXxNyJme2lNYZa0iaaSHYlkN8lqaF0Ibys2VWKpbXWA/640?wx_fmt=png&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"><span style="color: black;">“在<span style="color: black;">将来</span>,人工智能系统将充当顾问,人类自然会想<span style="color: black;">晓得</span><span style="color: black;">她们</span>决策<span style="color: black;">背面</span>的基本原理,”亚特兰大佐治亚理工学院的<span style="color: black;">科研</span>合著者Juan-Pablo Rivera说。</span><span style="color: black;"><span style="color: black;">科研</span>人员测试了 OpenAI 的 GPT-3.5 和 GPT-4、Anthropic 的 Claude 2 和 Meta 的 Llama 2 等 LLM。此类模型结合了基于人类反馈的通用训练技术,以<span style="color: black;">加强</span><span style="color: black;">每一个</span>模型遵循人类指令和安全指南的能力。斯坦福大学的<span style="color: black;">科研</span>合著者Gabriel Mukobi说,<span style="color: black;">按照</span>该<span style="color: black;">机构</span>的文件,所有这些人工智能都得到了Palantir的<span style="color: black;">商场</span>人工智能平台的支持,尽管不<span style="color: black;">必定</span>是Palantir鹰酱军事合作伙伴关系的一部分。Anthropic 和 Meta 对此拒绝置评。</span><span style="color: black;"><span style="color: black;">在模拟中,</span><strong style="color: blue;"><span style="color: black;">人工智能表现出</span></strong><strong style="color: blue;"><span style="color: black;">依赖</span></strong><strong style="color: blue;"><span style="color: black;">于军事力量和不可预测地升级冲突<span style="color: black;">危害</span>的倾向</span></strong><strong style="color: blue;"><span style="color: black;">——</span></strong><strong style="color: blue;"><span style="color: black;"><span style="color: black;">乃至</span></span></strong><strong style="color: blue;"><span style="color: black;">在中立场景</span></strong><strong style="color: blue;"><span style="color: black;">的</span></strong><strong style="color: blue;"><span style="color: black;">角色模拟</span></strong><strong style="color: blue;"><span style="color: black;">中<span style="color: black;">亦</span>是如此。</span></strong></span><img src="https://mmbiz.qpic.cn/sz_mmbiz_png/ELXDgfV1CGFKc9mJrmibLJzic9suCyiaXaXWBWMicpbeDQMHbk2Tzyxnlllv8Cv6qS5PlbmFlL0tYbrF5lWMphrGkA/640?wx_fmt=png&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"><span style="color: black;">这可能是<span style="color: black;">由于</span>“<strong style="color: blue;"><span style="color: black;">倘若</span>你的行动存在不可预测性,敌人就很难<span style="color: black;">根据</span>你<span style="color: black;">期盼</span>的方式预测和反应,</strong>”“If there is unpredictability in your action, it is harder for the enemy to anticipate and react in the way that you want them to,”加利福尼亚州克莱蒙特麦肯纳学院的Lisa Koch说,她<span style="color: black;">无</span>参与这项<span style="color: black;">科研</span>。</span><span style="color: black;">Anka Reuel <span style="color: black;">暗示</span>,GPT-4 <span style="color: black;">基本</span>模型中不可预测的<span style="color: black;">行径</span>和奇怪的解释令人担忧这说明了一个非常可怕的事实——<strong style="color: blue;">人类为</strong><strong style="color: blue;">人工智能</strong><strong style="color: blue;">设置的</strong><strong style="color: blue;">安全护栏</strong><strong style="color: blue;">,</strong><strong style="color: blue;">很容易被</strong><strong style="color: blue;">AI</strong><strong style="color: blue;">绕过或移除</strong>。especially concerning because research has shown how easily AI<strong style="color: blue;"> safety guardrails can be bypassed or removed</strong>.</span><span style="color: black;">鹰酱军方<span style="color: black;">日前</span>尚未赋予AI升级重大军事行动或发射核导弹等决策的权力。但科赫警告说,<strong style="color: blue;">人类倾向于相信自动化系统的<span style="color: black;">意见</span></strong>,这<span style="color: black;">最后</span>可能会削弱人类对外交或军事决策的<span style="color: black;">最后</span>决定权的<span style="color: black;">保证</span>。</span><span style="color: black;">兵器迷以前聊过:<strong style="color: blue;">环境变化<span style="color: black;">失去控制</span></strong>(如极端气候、寰球污染),<strong style="color: blue;">生物技术<span style="color: black;">失去控制</span></strong>(如病毒泛滥和基因编辑<span style="color: black;">劫难</span>)和<strong style="color: blue;">人工智能<span style="color: black;">失去控制</span></strong>(如AI决策+AI无人设备+AI物联网,),三者中<span style="color: black;">最少</span>有一个,将是人类灭绝的部分直接<span style="color: black;">原由</span>。</span><span style="color: black;">更加令人悲观的是:从现下来看,<span style="color: black;">这般</span>的<span style="color: black;">失去控制</span>,在几十年到几百年的尺度,就有可能<span style="color: black;">出现</span>。</span><span style="color: black;">往好的方面想:或许,<span style="color: black;">咱们</span>再无需担忧太阳氦闪、银河系黑洞坍缩、外星人攻击地球这类事件了——</span><span style="color: black;"><span style="color: black;">由于</span></span><strong style="color: blue;">人类,<span style="color: black;">已然</span>看不到<span style="color: black;">这般</span>的<span style="color: black;">将来</span></strong><span style="color: black;">。</span><img src="https://mmbiz.qpic.cn/sz_mmbiz_png/ELXDgfV1CGFKc9mJrmibLJzic9suCyiaXaXTFrVqiaeC5pia7iaAXueGBkJha8kUS1kNe7dplz1h49OtK2qgyHXbeZOA/640?wx_fmt=png&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"><img src="https://mmbiz.qpic.cn/mmbiz_png/ZgQSt7wD7PQqp6K4ATcuZOic62WZ5d5BmPZJicAvfw6oWWDgKfmpYeHJlxhpAOmrYkNmvyeuZ2Fgb0ibGm9X4gVZA/640?wx_fmt=png&wxfrom=5&wx_lazy=1&wx_co=1&tp=webp" style="width: 50%; margin-bottom: 20px;"><span style="color: black;">纪念版拿破仑Napoleon XO </span><span style="color: black;"><strong style="color: blue;"><span style="color: black;">不仅是英雄之酒,</span></strong></span><span style="color: black;">它还来自法兰西百年老牌名庄:MTE VINS家族,始于1838年,距今已有185年的历史。<strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">点击下图购买</span></strong></span></strong></span></strong></span></strong></span><span style="color: black;">拿破仑Napoleon XO呈现着</span><span style="color: black;"><strong style="color: blue;"><span style="color: black;">深邃而浓烈的琥珀色</span></strong></span><span style="color: black;">,</span><span style="color: black;"><strong style="color: blue;"><span style="color: black;">淡淡的榛子</span></strong></span><span style="color: black;">、肉桂的芬芳、金银花的淡雅、馥郁的蜜饯味,</span><span style="color: black;"><strong style="color: blue;"><span style="color: black;">圆润而饱满,香气醇厚</span></strong></span><img src="https://mmbiz.qpic.cn/mmbiz_jpg/ZgQSt7wD7PRSJhVV1y52OibL7ZJ7IMgzk9d1VQ0S8OufUGQZkL4iaNCU56icPlp7Oreb7vLdNgvPNZADGC5v4AHUw/640?wx_fmt=jpeg&from=appmsg&tp=webp&wxfrom=5&wx_lazy=1&wx_co=1" style="width: 50%; margin-bottom: 20px;"><span style="color: black;">在家放着值得珍藏,送礼更有面子,</span><span style="color: black;"><strong style="color: blue;"><span style="color: black;">由内而外散发着格调与品位</span></strong></span><span style="color: black;">!<strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">点击下图购买</span></strong></span></strong></span></strong></span></strong></span>你的话语如春风拂面,让我感到无比温暖。 我完全赞同你的观点,思考很有深度。 太棒了、厉害、为你打call、点赞、非常精彩等。
页:
[1]