BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Andrew Parker (Sandbox Quantum)
DTSTART:20250403T163000Z
DTEND:20250403T180000Z
DTSTAMP:20260421T115327Z
UID:BC-Math/27
DESCRIPTION:Title: <a href="https://researchseminars.org/talk/BC-Math/27/"
 >From Calculus to Conversation: Inside LLMs</a>\nby Andrew Parker (Sandbox
  Quantum) as part of Brooklyn College Math Department Events\n\nLecture he
 ld in 330 Ingersoll Extension.\n\nAbstract\nThis talk explores the mathema
 tical foundations and evolutionary trajectory of Large Language Models (LL
 Ms). We begin by examining the pre-training mechanics—how models learn t
 o predict tokens through probability distributions and gradient descent op
 timization across vast corpora of human language. We'll investigate why th
 ese statistical approaches to language processing present unique challenge
 s for mathematical reasoning\, where symbolic manipulation differs fundame
 ntally from natural language processing.\n\nWe will further explore fine-t
 uning’s effects on model behavior as well as reinforcement learning’s 
 efforts to redirect models toward human-aligned responses. We'll analyze t
 he architecture of model conversations\, examining system prompts and mode
 l responses\, while considering the technical challenges in maintaining al
 ignment.\n\nWe'll conclude by exploring the formal reasoning capabilities 
 of modern LLMs and the emerging paradigm of AI agents\, offering practical
  strategies for effectively communicating with these systems. Throughout\,
  we'll provide a balanced perspective on both the capabilities and limitat
 ions of LLMs\, offering insights valuable to anyone interested in the inte
 rsection of mathematics\, computer science\, and artificial intelligence.\
 n\nFood and refreshments will be served.\n
LOCATION:https://researchseminars.org/talk/BC-Math/27/
END:VEVENT
END:VCALENDAR