BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//jEvents 2.0 for Joomla//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:STANDARD
DTSTART:20231127T104500
RDATE:20240310T030000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:America/New_York EST
END:STANDARD
BEGIN:STANDARD
DTSTART:20241103T010000
RDATE:20250309T030000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:America/New_York EST
END:STANDARD
BEGIN:STANDARD
DTSTART:20251102T010000
RDATE:20260308T030000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:America/New_York EST
END:STANDARD
BEGIN:STANDARD
DTSTART:20261101T010000
RDATE:20270314T030000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:America/New_York EST
END:STANDARD
BEGIN:STANDARD
DTSTART:20271107T010000
RDATE:20280312T030000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:America/New_York EST
END:STANDARD
BEGIN:STANDARD
DTSTART:20281105T010000
RDATE:20290311T030000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:America/New_York EST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20240310T030000
RDATE:20241103T010000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:America/New_York EDT
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20250309T030000
RDATE:20251102T010000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:America/New_York EDT
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20260308T030000
RDATE:20261101T010000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:America/New_York EDT
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20270314T030000
RDATE:20271107T010000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:America/New_York EDT
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20280312T030000
RDATE:20281105T010000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:America/New_York EDT
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:4c3da0988dee994a96532bf3c3f73a71
CATEGORIES:Mathematical Physics Seminar
CREATED:20241110T190746
SUMMARY:Webinar: Scott Aaronson - AI Safety and Theoretical Computer Science
LOCATION:Zoom
DESCRIPTION:Scott Aaronson – University of Texas at Austin\n \nWednesday, November 27th
  , 10:45AM EST\n \nAI Safety and Theoretical Computer Science\n \nProgress 
 on AI safety and alignment, like the current AI revolution more generally, 
 has been almost entirely empirical.  In this talk,  however, I'll survey a 
 few areas where I think theoretical computer science can contribute to AI s
 afety, including:\n - How can we robustly watermark the outputs of Large La
 nguage Models and other generative AI systems, to help identify academic ch
 eating, deepfakes, and AI-enabled fraud?  I'll explain my proposal and its 
 basic mathematical properties, as well as what remains to be done.\n- Can o
 ne insert undetectable cryptographic backdoors into neural nets, for good o
 r ill?  In what senses can those backdoors also be unremovable?  How robust
  are they against fine-tuning?\n - Should we expect neural nets to be "gene
 rically" interpretable?  I'll discuss a beautiful formalization of that que
 stion due to Paul Christiano, along with some initial progress on it, and a
 n unexpected connection to quantum computing.\n
X-ALT-DESC;FMTTYPE=text/html:<p style="text-align: center;"><strong>Scott Aaronson – University of Texas
  at Austin</strong></p><p style="text-align: center;"><strong>&nbsp;</stron
 g></p><p style="text-align: center;"><strong>Wednesday,&nbsp;November 27th 
 ,&nbsp;10:45AM EST</strong></p><p style="text-align: center;"><strong>&nbsp
 ;</strong></p><p style="text-align: center;"><strong>AI Safety and Theoreti
 cal Computer Science</strong></p><p style="text-align: center;"><strong>&nb
 sp;</strong></p><p>Progress on AI safety and alignment, like the current AI
  revolution more generally, has been almost entirely empirical.&nbsp; In th
 is talk, &nbsp;however, I'll survey a few areas where I think theoretical c
 omputer science can contribute to AI safety, including:</p><p>&nbsp;- How c
 an we robustly watermark the outputs of Large Language Models and other gen
 erative AI systems, to help identify academic cheating, deepfakes, and AI-e
 nabled fraud?&nbsp; I'll explain my proposal and its basic mathematical pro
 perties, as well as what remains to be done.</p><p>- Can one insert undetec
 table cryptographic backdoors into neural nets, for good or ill?&nbsp; In w
 hat senses can those backdoors also be unremovable?&nbsp; How robust are th
 ey against fine-tuning?</p><p>&nbsp;- Should we expect neural nets to be "g
 enerically" interpretable? &nbsp;I'll discuss a beautiful formalization of 
 that question due to Paul Christiano, along with some initial progress on i
 t, and an unexpected connection to quantum computing.</p>
DTSTAMP:20260829T124407
DTSTART;TZID=America/New_York:20241127T104500
DTEND;TZID=America/New_York:20241127T120000
SEQUENCE:0
TRANSP:OPAQUE
END:VEVENT
END:VCALENDAR