<div dir="ltr"><div class="gmail_default" style="font-family:arial,helvetica,sans-serif"><b style="font-size:large"><span style="color:rgb(26,26,26);font-family:Merriweather,Georgia,serif">"If fewer computational resources are dedicated to safety than to capability, then safety issues such as jailbreaks will always exist.</span><span style="color:rgb(26,26,26);font-family:Merriweather,Georgia,serif"> Can we align a language model externally without understanding how they work inside? The answer to this question is a resounding NO."</span></b></div><div class="gmail_default" style="font-family:arial,helvetica,sans-serif"><span style="color:rgb(26,26,26);font-family:Merriweather,Georgia,serif"><font size="4"><br></font></span></div><div class="gmail_default" style=""><span style="color:rgb(26,26,26)"><a href="https://www.quantamagazine.org/cryptographers-show-that-ai-protections-will-always-have-holes-20251210/?mc_cid=db3cb01235&mc_eid=1b0caa9e8c" style=""><font size="4" style="" face="tahoma, sans-serif"><b>Mathematicians Show That AI Protections Will Always be incomplete </b></font></a><br></span></div><div class="gmail_default" style=""><br></div><div class="gmail_default" style=""><div style="color:rgb(80,0,80)"><b><font face="tahoma, sans-serif"><font size="4">John K Clark    See what's on my new list at  </font><font size="6"><a href="https://groups.google.com/g/extropolis" rel="nofollow" target="_blank">Extropolis</a></font></font></b></div><div style="color:rgb(80,0,80)"><b><font face="tahoma, sans-serif"><br></font></b></div><font size="1" color="#ffffff">j]c</font><br class="gmail-Apple-interchange-newline"></div></div>