大语言模型LLM和智能体：AGI的路线图和未来愿景.pdf

1、LLMLLMReasoning Reasoning and AI and AI AgentsAgents:Jun Wang,UCLHinton created Gatsby Unit,where Hassabis was a postHinton created Gatsby Unit,where Hassabis was a post-doc researcherdoc researcher“To the Boltzmann machine work cited by the Nobel Prize committee led on to his development of the Res

2、tricted Boltzmann machine while at the UCL Gatsby Unit,and this in turn helped to feed the late 2000s resurgence of deep learning on which current AI systems depend.Professor Rees said:“Demis has always been a brilliant and highly creative researcher and innovator,and we are immensely proud to have

3、him as part of the UCL community.It is tremendously exciting to hear he has won the Nobel Prize,and at the same time entirely unsurprising.What is a learning?What is a learning?In biology,learning means a change of behaviour as a result of experienceIn classical conditioning1,animals can learn to id

4、entify a useful pattern in the environment by associating onestimulus with another:repeated given ring-a-bell O,food X,a dog will start to salivate(anticipate the upcoming of the food)when bell rings again X O X Learned behaviours are adaptive and thus are essential for animals to survive in the cha

5、nging environment e.g.,they may learn not to eat certain foods if they have ever become ill after eating them more learned behaviours more intelligentIvan Petrovitch Pavlov and William Gantt.“Lectures on conditioned reflexes:Twenty-five years of objective study of the higher nervous activity(behavio

6、ur)of animals.”.In:(1928).AI agent as a systemAI agent as a systemAgent(|)Perception:(|)Actuator:WorldMutual information I(X,O)Blackbox optimisation problemBlackbox optimisation problemBO:wide applicationsBO:wide applications hyper-parameter tuning/autoML:f is validation performanceand a is a set of