《Your LLM, Your Data.pdf》由会员分享,可在线阅读,更多相关《Your LLM, Your Data.pdf(47页珍藏版)》请在三个皮匠报告上搜索。
1、Your LLM,Your Data,Your InfrastructureSharon Zhou,PhDCofounder&CEO,Lamini realSharonZhouDatabricks20231_DAIS_Title_SlideDemoPlay with this LLM at app.lamini.aiWe couldnt have gotten to this level of LLM use and accuracy without Lamini.VP Eng/Director,BigTech(Fortune 10)Why train your own LLM?You wan
2、t LLMs in your productsYoure not alone:Over 20%of S&P 500 have AI in earnings calls,Q1 2023Add magic to your product,acquire new customers,increase your top line.For example:Customers can answer their own questions in seconds.Access all of your documentation with personalized information.Every new f
3、eature would be 10 x faster to build with a copilot.Reduce your engineering and operational costs.Generative AI is the next platformAI is about using data to craft the right use cases.Like the last platform,e.g.mobile,you need in-house know-how to be at the cutting edge.Our own LLM,trained with Lami
4、ni,was the best and closest to our use case in a blind test,comparing the model to ChatGPT with retrieval.Prompt engineer,Midsize AI startup1_DAIS_Title_SlideLevel up from promptsOff-the-shelf LLMs arent perfectLack personalization to:Your data Your use casesImagine how powerful GPT-4 would be if ad
5、apted to your business!Hundreds of engineering leaders have told us these reasons:Youre worried about:Data leaks from a 3rd partySending all of your IP and source codeGiving up data moatYou dont want an update which happens often to break your critical use cases.Your other option:Hire dozens of top
6、AI researchersLike Github did for CopilotLike OpenAI did for ChatGPTThese LLMs are trained on someone elses dataYou cant just hand over proprietary dataThese LLMs need reliability&maintenanceHow to train your own LLMs1_DAIS_Title_SlideTraining LLMs inside your Databricks clusterYour LLM,Your Data,Your InfrastructureSharon Zhou,PhDCofounder&CEO,Lamini realSharonZhouDatabricks2023