與人工智能對話：轉變會議學習與數據參與方式.pdf

《與人工智能對話：轉變會議學習與數據參與方式.pdf》由會員分享，可在線閱讀，更多相關《與人工智能對話：轉變會議學習與數據參與方式.pdf（44頁珍藏版）》請在三個皮匠報告上搜索。

1、Conversations With AI:Transforming Conference Learning&Data EngagementBrian PichmanSolving Problems with Technology I am fascinated by how technology can be used to enhance ourselves and society.The goal for me is to introduce technology in an affordable and scalable way.General House KeepingUsually

2、 When Discussing AI:The Many Uses of AIAI Can Be Used Almost EverywhereThere are tools for nearly ever sector,every industry.Data Generation Estimateshttps:/ Project|Brian Pichman8Generative AILensa AppEvolve Project|Brian Pichman9Generative AIProfessional HeadshotsEvolve Project|Brian Pichman11Befo

3、re and After AI ToolsEvolve Project|Brian Pichman12MeetGeekhttps:/ Project|Brian Pichman18BHuman AIAI Content CreationAI Twin,presented by Twin Protocol,What Problem Do I Want to SolveEvolve Project|Brian Pichman21The Problems Great Conferencebut how can I be in more places than once?I dont ever loo

4、k at my notes again and forget half the things I wrote why did I draw a million doodles?I wish I had someone I could discuss this conference with.Evolve Project|Brian Pichman22Imaginehttps:/ Project|Brian Pichman244 Knowledge GraphsEvolve Project|Brian Pichman25Infranodus Knowledge GraphsEquipmentTa

5、scam-DR-40XAnker M650Used Whisper to convert speech to text)offline https:/ Transcriptions Sound Quality Challenge Using devices that have built-in noise cancelation and can be plugged directly into a sound system.Transcription Accuracy Challenge The approach focuses less on“exact”phrases but on ove

6、rall themes.So,while there may be some transcription issues,the vast majority of the data is transcribed accurately(80-90%)It is still essential to spot-check and scan through transcriptions before they are fed into a model for summarization.Building a RAG RAG Retrieval Augmented Generation Think of

7、 it as a discovery layer https:/ https:/ With LM Studio,you can.-Run LLMs on your laptop,entirely offline -Use models through the in-app Chat UI or an OpenAI compatible local server -Download any compatible model files from HuggingFace repositories-Discover new¬eworthy LLMs in the apps home pageW

8、hat are the minimum hardware/software requirements?Apple Silicon Mac(M1/M2/M3)with macOS 13.6 or newerWindows/Linux PC with a processor that supports AVX2(typically newer PCs)16GB+of RAM is recommended.For PCs,6GB+of VRAM is recommendedNVIDIA/AMD GPUs supportedEvolve Project|Brian Pichman37Understan

9、ding LMStudioModel Sizes(e.g.,7B,13B,30B,65B)These numbers refer to the number of parameters(billions)in the model:7B(7 billion parameters)Smaller,runs faster,less powerful 13B(13 billion parameters)More capable,still relatively efficient 30B(30 billion parameters)Even smarter but requires more RAM

10、65B(65 billion parameters)Very powerful but needs high-end hardwareWhich to choose?If you have limited RAM(8-16GB)7B or 13B If you have more RAM(24GB+)30B or higher for better quality If you have a high-end system(64GB+RAM,strong GPU)Go for 65BEach parameter is essentially a tiny part of how the mod

11、el processes and generates text.More parameters generally mean a smarter and more capable model,but also require more computing power to run.Context Length(e.g.,4K,16K,32K)This determines how much text the model can“remember”in one interaction.4K(4,096 tokens)Shorter memory,but faster 16K or 32K Can

12、 handle longer conversations,uses more RAMWhich to choose?For chatting and casual use?4K-8K For research,summarization,or coding?16K+AI Model Collapse/AI Degradation Overtime,a model can get over saturated with“nonsense”data Best solution use an LLM and a RAG Purge the LLM and refresh based on use R

13、AG Data doesnt change LLM data can as a result of user input Leveraging The Bot Ask about transcripts but it wont quote Ask to compare sessions together Recommend other topics to explore.Will ask questions as a follow up to make an engaging conversation Your Conference Assistanthttps:/links.evolveproject.org/ds25Other Tools to Check OutQuestions Brian Pichman bpichmanevolveproject.orghttps:/links.evolveproject.org/ds25