<div dir="ltr">On Tue, 21 Nov 2023 at 19:25, Keith Henson via extropy-chat <<a href="mailto:extropy-chat@lists.extropy.org" target="_blank">extropy-chat@lists.extropy.org</a>> wrote:<br>><br>> The LLM of AI chatbots are trained on what humans know.  They don't<br>> seem any more dangerous to me than exceptionally well-informed humans.<br>><br>> Of course, I could be wrong.<br>><br>> Keith<br>> _______________________________________________<br><br><br>Why not ask an <span class="gmail_default" style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)">LL</span>M AI?  <span class="gmail_default" style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)">  </span> So I did.......<br>(But if it gets really intelligent, the AI won't give away its secret powers!).  :)<br><div><br></div><div>Question:<br></div><div>If LLM AI is trained on human knowledge, how is it possible to be more dangerous than humans? <br></div><div><br></div><div><div style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)" class="gmail_default">Answer:</div><div style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)" class="gmail_default">Powered by gpt-3.5-turbo.</div><div style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)" class="gmail_default">As an AI language model, I can provide some insights on this topic. When we talk about AI being potentially more dangerous than humans, it's not necessarily because it possesses inherent malevolence or ill intentions. The concern mainly arises due to a few factors:<br><br>1. Amplification of human biases: AI models like me are trained on vast amounts of human-generated data, which means we can inadvertently inherit and perpetuate the biases present in that data. This can result in biased or discriminatory outputs that reflect societal prejudices. If these biases are not identified and addressed, they can be amplified at scale, leading to potential harm or unfair treatment.<br><br>2. Lack of common sense and contextual understanding: While AI models excel at processing and generating text based on patterns in training data, they often lack genuine understanding of the world. This can lead to misinterpretation of queries or generating outputs that are factually incorrect or nonsensical. In certain scenarios, relying solely on AI-generated information without human verification can lead to misinformation or misguided decisions.<br><br>3. Optimization for objectives without ethical considerations: AI models are typically trained to optimize specific objectives, such as generating relevant or engaging responses. However, these objectives may not encompass ethical considerations, long-term consequences, or the well-being of individuals or society as a whole. This narrow focus on optimization can result in AI systems pursuing goals in ways that humans may find undesirable or unsafe.<br><br>4. Potential for misuse or malicious intent: While AI itself may not possess intent, it can be utilized by individuals or organizations with malicious intentions. AI technologies, including language models like me, have the potential to be weaponized for spreading misinformation, generating deepfake content, or automating cyber attacks, among other harmful activities.<br><br>Addressing these concerns requires a combination of technical advancements, ethical guidelines, and responsible deployment of AI systems. Ongoing research and development aim to improve transparency, accountability, and fairness in AI models to mitigate potential risks and ensure their beneficial use.</div></div><div><span class="gmail_default" style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)">------------------</span><p><span class="gmail_default" style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)">BillK<br></span></p><p><span class="gmail_default" style="font-family:times new roman,serif;font-size:small;color:rgb(0,0,0)"><br></span></p></div></div>