AWS Summit New York City 2023 – Keynote with Swami Sivasubramanian | AWS Events
![AWS Summit New York City 2023 – Keynote with Swami Sivasubramanian | AWS Events AWS Summit New York City 2023 – Keynote with Swami Sivasubramanian | AWS Events](/pic/aws_summit_new_york_city_2023_keynote_with_swami_sivasubramanian_aws_events/MVBrQUJXQ0pJTk0_.jpeg)
PLEASE WELCOME VP DATABASE ANALYTICS AND MACHINE LEARNING SWAMI SIVASUBRAMANIAN. THANK YOU ALL SO MUCH FOR BEING HERE TODAY AND GIVING ME A CHANCE TO SPEAK WITH YOU. A LOT HAS HAPPENED IN THE LAST YEAR AND WITH ALL THIS CHANGE WE HAVE AN OPPORTUNITY TO REEVALUATE AND ENHANCE AND OPTIMIZE THE TECHNOLOGY THAT SUPPORTS OUR SUCCESS. I ALSO BELIEVE IT'S A
TIME FOR NET NEW IDEAS. THESE ARE THE SPARKS OF INVENTION THAT ENABLE US TO SOLVE REALLY COMPLEX PROBLEMS AND REIMAGINE HOW WE DO THINGS. AND WHEN YOU ARE BACKED BY THE MOST TRANSFORMATIVE, INNOVATIVE TECHNOLOGY SIMS, YOU CAN BRING THESE AMAZING IDEAS TO LIFE. ONE OF THOSE TRANSFORMATIVE TECHNOLOGIES THAT'S GAINING A LOT OF TRACTION TODAY IS GENERATIVE AI GENERATIVE AI. AS CAPTURED, OUR IMAGINATIONS FOR ITS ABILITY TO CREATE IMAGES AND VIDEOS RIGHT STORIES AND EVEN GENERATE CODE. I BELIEVE IT WILL TRANSFORM EVERY APPLICATION INDUSTRY AND BUSINESS. SO YOU MIGHT BE WONDERING, WITH SO MUCH
POTENTIAL, WHY IS THIS TECHNOLOGY THAT HAS BEEN PERCOLATING FOR MANY DECADES SUDDENLY SEEING SO MUCH TRACTION ONLY JUST NOW? THAT IS BECAUSE THIS TECHNOLOGY HAS REACHED ITS TIPPING POINT AT THE CONVERGENCE OF TECHNOLOGICAL PROGRESS AND THE VALUE OF WHAT IT CAN ACCOMPLISH TODAY. THAT'S BECAUSE WE HAVE MASSIVE PROLIFERATION OF DATA AND THE AVAILABILITY OF EXTREMELY SCALABLE COMPUTE INFRASTRUCTURE AND THE ADVANCEMENT OF ML TECHNOLOGIES. OVER TIME. GENERATIVE AI IS FINALLY TAKING SHAPE. IN PARTICULAR FOR THESE INNOVATIONS HAVE MADE THE CAPABILITIES OF AI POSSIBLE JUST WITHIN THE LAST MONTH. LET ME GIVE SOME BACKGROUND FOR EXAMPLE, TRADITIONAL FORMS OF MACHINE LEARNING ALLOWED US TO TAKE SIMPLE INPUTS LIKE NUMERICAL VALUES AND MAP THEM TO SIMPLE OUTPUTS LIKE PREDICTOR VALUES AND THEN DEEP LEARNING CAME ALONG WHERE WE COULD TAKE COMPLICATED INPUTS LIKE VIDEOS OR IMAGES AND MAP THEM TO RELATIVELY SIMPLE OUTPUTS. AND NOW WITH GENERATIVE AI, WE CAN LEVERAGE THIS MASSIVE AMOUNTS OF COMPLEX DATA TO CAPTURE AND PRESENT KNOWLEDGE IN MORE ADVANCED WAYS. MAPPING COMPLICATED INPUTS TO
COMPLICATED OUTPUTS IS THE STEP WISE IMPROVEMENTS IN FUNCTIONALITY. WE WERE SUPPORTED BY STEP WISE IMPROVEMENTS IN THE UNDERLYING ML MODELS THEMSELVES THAT THAT'S BECAUSE TRADITIONAL ML MODELS USED ARCHITECTURES THAT REQUIRE MONTHS OF COSTLY AND MANUAL DATA PREPARATION, DATA LABELING AND MODEL TRAINING , ALL TO DO ONE SPECIFIC TASK. THE LARGE MODELS THAT POWER THESE GENERATIVE AI APPLICATIONS IS CALLED FOUNDATIONAL MODELS ARE DRIVEN BY THE TRANSFORMER NEURAL NET ARCHITECTURE THAT SIGNIFICANTLY CUTS DOWN ON THIS DEVELOPMENT PROCESS. WITH THIS ARCHITECTURE, MODELS CAN BE TRAINED ON MASSIVE AMOUNTS OF UNLABELED DATA IN THE PRE-TRAINING STAGE SO THAT THEY CAN BE USED OUT OF THE BOX FOR A WIDE VARIETY OF GENERALIZED TASKS AND THEY CAN BE USED AND EASILY ADAPTED FOR PARTICULAR DOMAINS OR APPLICATIONS WITH RELATIVELY VERY SMALL AMOUNT OF UNLABELED DATA. THIS PROCESS OF CUSTOMIZATION IS ALSO KNOWN AS FINE TUNING THE ABILITY TO EASILY CUSTOMIZE A PRE-TRAINED MODEL THROUGH FINE TUNING IS AN ABSOLUTE GAME CHANGER. IT IS SUBSTANTIAL FASTER. IT ALSO REQUIRES A LOT LESS COMPUTATION.
TIME AND A LOT LESS DATA FOR FINE TUNING THAN SPENDING MONTHS OF CREATING A TASK SPECIFIC MACHINE LEARNING MODEL. SO WHILE WHAT USED TO TAKE MONTHS OF SCIENTISTS TO BUILD AN ML MODEL FOR ONE TASK CAN NOW BE DONE RELATIVELY EASY WITH ONE BIG MODEL AND FINE TUNING TO ACCOMPLISH THE SAME GOAL. SO TODAY, GENAI IS BEING APPLIED TO USE CASES FROM ALL LINES OF BUSINESS, FROM ENGINEERING TO CUSTOMER SERVICE TO FINANCE, YOU CAN IMPROVE CUSTOMER EXPERIENCES THROUGH CAPABILITY, THINGS LIKE VIRTUAL CHATBOT, VIRTUAL ASSISTANTS. YOU CAN BOOST YOUR EMPLOYEES PRODUCTIVITY WITH TEXT SUMMARIZATION AND CODE GENERATION JSON YOU CAN TURBOCHARGE PRODUCTION OF ALL TYPES OF CONTENT LIKE ART, MUSIC OR ANIMATIONS, AND YOU CAN USE GENERATIVE AI TO IMPROVE BUSINESS OPERATIONS. SO WHAT DO CUSTOMERS NEED TO DO TO UNLOCK THE VALUE OF GENERATIVE AI FOR THEIR USE CASES? THE FIRST THING IS YOU NEED ACCESS TO BEST IN CLASS FOUNDATIONAL MODELS. WE KNOW MODEL CHOICES PARAMOUNT BECAUSE THERE IS GOING TO BE NO ONE MODEL TO RULE THEM ALL. IT'S
ABOUT CHOOSING THE RIGHT MODEL FOR THE RIGHT JOB. THEN THE CUSTOMERS NEED THE ABILITY TO SECURELY CUSTOMIZE THESE MODELS WITH THEIR DATA AND THEN THEY NEED EASY TO USE TOOLS TO DEMOCRATIZE GENERATIVE AI WITHIN THEIR ORGANIZATIONS AND IMPROVE EMPLOYEE PRODUCTIVITY AND UNDERPINNING ALL OF THIS IS YOU NEED TO KEEP YOUR COSTS AND LATENCY LOW WITH PURPOSE BUILT. ML INFRASTRUCTURE. WE ARE DELIVERING ALL OF THIS TO OUR CUSTOMERS THROUGH AMAZON BEDROCK WITH BEDROCK. CUSTOMERS CAN
EASILY BUILD AND SCALE GENAI APPLICATIONS WITH A SELECTION OF INDUSTRY LEADING FMS ALL WITH A SIMPLE API WITHOUT MANAGING ANY INFRASTRUCTURE. AMAZON BEDROCK MAKES IT EASY TO CUSTOMIZE THESE FOUNDATIONAL MODELS WITH YOUR DATA. WITH JUST A FEW LABELED EXAMPLES IN AMAZON S3 ALL ALL DATA IN BEDROCK IS ENCRYPTED AND YOUR DATA IS NEVER USED TO TRAIN THE ORIGINAL BASE MODEL. YOU CAN CONFIGURE YOUR VPC SETTINGS TO ACCESS BEDROCK APIS AND PROVIDE FINE TUNING MODEL DATA IN A SECURE MANNER AND YOU ARE GOOD TO GO. WE BELIEVE IN OFFERING THE BEST IN CLASS FOUNDATIONAL MODELS. TO THAT END, WE HAVE AMAZON TITAN MODEL, INCLUDING
TITAN TEXT MODEL FOR TEXT SUMMARIES, CREATION AND GENERATION AND TITAN EMBEDDINGS MODEL FOR PERSONALIZATION AND RECOMMENDATION. WE HAVE ANTHROPIC CLAUDE THAT IS BUILT WITH THE LATEST RESEARCH ON SAFETY AND NLP TO PERFORM CONVERSATIONS AND TEXT PROCESSING. BEDROCK ALSO SUPPORTS STABILITY. AI'S LATEST TEXT TO IMAGE AND VARIOUS IMAGE PROCESSING MODELS, AND WE OFFER AI21LABS JURASSIC-2 MODEL NOW LET'S TAKE A CLOSER LOOK AT TITAN OUR TITAN FMS ARE PRE-TRAINED ON DATASETS THAT CONTAIN LARGE VOLUMES OF INFORMATION, OFTEN FROM DIVERSE SOURCES, ENABLING CUSTOMERS TO BUILD FOR DIVERSE SET OF USE CASES. THEY ALSO SUPPORT RESPONSE CIVIL USE OF AI BY DETECTING INAPPROPRIATE CONTENT IN THE INPUT PROMPTS, BUT ALSO FILTER ING IN THE OUTPUTS. ANY
KIND OF INAPPROPRIATE CONTENT LIKE HATE SPEECH, PROFANITY AND VIOLENCE. TITAN AND OUR BROAD LIBRARY OF FMS AND BEDROCK HAVE ENABLED OUR CUSTOMERS TO START INNOVATING WITH THE RIGHT TOOLS TODAY OUR CUSTOMERS ARE DRIVING INNOVATION WITH BEDROCK FOR VARIOUS FOUNDATIONAL MODELS FOR SELF-SERVICE, CUSTOMER CHAT, TEXT ANALYSIS, REPORT GENERATION AND POST-CALL ANALYSIS. CUSTOMERS LIKE SUN LIFE, WHICH USES BEDROCK TO EXPERIMENT WITH GENERATIVE AI APPLICATIONS THAT ANALYZE MARKET DATA AND ASSESS EMPLOYEE PRODUCTIVITY. BRIDGEWATER ASSOCIATES A PREMIER
ASSET MANAGEMENT FIRM THAT IS BUILDING AND SCALING GENAI APPLICATIONS. LEVERAGING THESE FOUNDATIONAL MODELS ON AMAZON BEDROCK AND PHILIPS WHO WILL USE BEDROCK TO DEVELOP GENAI APPLICATIONS ACROSS ITS ENTIRE PORTFOLIO TO SUPPORT EFFICIENT CLINICAL WORKFLOWS, ENHANCE DIAGNOSTIC CAPABILITY FOR THEIR PATIENTS. OUR CUSTOMERS ARE ALSO PUTTING GENAI INTO ACTION WITH THE HELP OF OUR GENAI INNOVATION CENTER, A NEW PROGRAM TO HELP THEM ACCELERATE SUCCESS WITH TOOLS LIKE AMAZON BEDROCK. THIS
INCLUDES CUSTOMERS LIKE LONELYPLANET AND RYANAIR WHO ARE WORKING WITH OUR TEAMS TO DEVELOP NEW USE CASES. SO WE ARE EXCITED TO SEE ALL OF THIS MOMENTUM AND WE WILL CONTINUE TO SUPPORT OUR CUSTOMERS. GENAI USE CASES WITH ALL THESE BEST IN CLASS FOUNDATIONAL MODELS. THAT'S WHY TODAY WE ARE OFFERING
EVEN MORE CHOICE IN FOUNDATIONAL MODELS. WE ARE EXCITED TO ANNOUNCE THE ADDITION OF TWO NEW MODELS FROM COHERE COMMAND AND EMBED. SO COHERE COMMAND IS A TEXT GENERATION MODEL FOR BUSINESS APPS LIKE SUMMARIZATION COPYWRITING, DIALOG AND Q&A AND EMBED IS OUR TEXT UNDERSTANDING MODEL THAT CAN BE USED FOR SEARCH CLUSTERING OVER 100 PLUS LANGUAGES. THESE TWO ADDITIONS REINFORCE OUR COMMITMENT TO MAKE THE BEST IN CLASS FMS AVAILABLE . IN ADDITION TO THIS NEW PROVIDER, WE ARE ALSO EXCITED TO ANNOUNCE STABILITY. AI'S LATEST MODEL STABLE DIFFUSION. XL 1.0. WE ARE ALSO EXCITED TO ANNOUNCE
THE ADDITION OF ANTHROPIC CLAUDE 2 CLAUDE V2 NOW CAN TAKE UP TO 100,000 TOKENS IN EACH CONVERSATIONAL PROMPT, MEANING IT CAN OVER HUNDREDS OF PAGES OR EVEN AN ENTIRE BOOK. IT CAN ALSO WRITE LONGER DOCUMENTS THAN ITS PREVIOUS VERSION. CLAUDE IS ONE OF OUR POPULAR FMS AND WE KNOW CUSTOMERS WILL BE EXCITED TO TAKE ADVANTAGE OF THESE FUNCTIONALITY. ALL OF THIS EARLY
SUCCESS ON BEDROCK IS BUILT ON YEARS OF EXPERIENCE OF MAKING THE BEST IN CLASS PRE-TRAINED FM AVAILABLE ON SAGEMAKER JUMPSTART WHERE CUSTOMERS CAN GET THEIR HANDS ON LATEST PRE-TRAINED PUBLICLY AVAILABLE MODELS FROM ACADEMIA AND INDUSTRY JUMPSTART OFFERS ML PRACTITIONERS THESE DEEP MODEL CUSTOMIZATION AND EVALUATION CAPABILITIES LIKE SAGEMAKER STUDIO SAGEMAKER SDK AND CONSOLE USING WHICH THEY CAN ACTUALLY GO DO ALL THESE CUSTOMIZATIONS AND ADD NEW MODELS ARE ADDED ON A WEEKLY BASIS, LIKE THE LLAMA V2 WE ADDED LAST WEEK. NOW I'D LIKE TO INTRODUCE ONE OF OUR CUSTOMERS TO THE STAGE TO SHARE HOW THEY ARE LEVERAGING ML AND GENERATIVE AI TO HELP BUSINESSES. SIMS IFY THE JOURNEY FROM UNIFIED DATA TO PERSONALIZED CUSTOMER EXPERIENCES. PLEASE WELCOME GABRIELLE TAO FROM SALESFORCE. THANK YOU SWAMI. IT IS A GREAT
HONOR TO BE HERE WITH YOU TODAY. SALESFORCE IS THE WORLD'S LEADING CRM PROVIDING A SUITE OF APPLICATIONS ACROSS SALES SERVICE, MARKETING, COMMERCE, ANALYTICS AND MORE, AND NOW DATA CLOUD. THAT ENABLE BUSINESSES TO BETTER CONNECT WITH THEIR CUSTOMERS. I REPRESENT SALESFORCE DATA CLOUD AND WE LEVERAGED AWS SERVICES TO CREATE MACHINE LEARNING GENAI AND BUILD A SOLUTION THAT HELPS OUR COMPANIES, WHO ARE OUR CUSTOMERS, USE ALL OF THOSE DATA IN A HARMONIZED FASHION, CREATE INSIGHTS AND DELIVER PERSONALIZED EXPERIENCES ALL AT HYPER SCALE. THAT IS CERTAINLY NOT EASY TODAY ON OUR PLANET THERE ARE MORE DEVICES AND THERE ARE HUMANS. AS MORE AND MORE
DECISIONS AND DATA GROWS, IT BECOMES HARDER AND HARDER TO CONNECT WITH THE CUSTOMERS. AS SALESFORCE, WE RECENTLY DID SOME RESEARCH AND FOUND THAT 60 TO 70% OF THE RESPONDENTS EXPECT COMPANIES TO UNDERSTAND AND EVEN ANTICIPATE THEIR NEEDS. AND YET A FULL 56% OF THEM FEEL THAT MOST COMPANIES TREAT THEM JUST AS A NUMBER. HOW COULD THEY NOT? AN AVERAGE COMPANY HAS 976 APPS LOCATIONS TO BETTER UNDERSTAND CUSTOMERS NEEDS, THERE REALLY NEEDS TO BE A MORE UNIFIED UNDERSTANDING OF THE CUSTOMERS. SO SALESFORCE HAS BEEN ON THIS JOURNEY. LAST YEAR WE ANNOUNCED
SALESFORCE DATA CLOUD TO HELP YOU CREATE A SINGLE SOURCE OF TRUTH OF YOUR DATA WITH AN OPENNESS THAT IS UNPRECEDENTED IN A DATA PLATFORM, A ZERO-ETL FRAMEWORK MERCK ACCESSING YOUR DATA WHERE THEY ARE SIDE BY SIDE AND DEEPLY MERGED WITH THE CRM PLATFORM. IT IS A FOUNDATION THAT UNDERPINS FUTURE INNOVATIONS AT SALESFORCE. IT IS BUILT ON OUR HYPERFORCE INFRASTRUCTURE AND BACKED BY QUITE A FEW AWS SERVICES EKS, EMR AND MANY OTHERS BACKED BY THESE WONDERFUL BACKEND SERVICES. WE BUILT A LOT OF NO CODE APPLICATION FEATURES, INCLUDING TRANSFORM IDENTITY INSIGHTS SEGMENTS AND MORE CATERING, NOT JUST TO THE DATA PERSONA BUT THE ANALYSTS AND BUSINESS PERSONAS AS WELL. DEMOCRATIZING BIG. DATA NOW, WHILE SALESFORCE DATA CLOUD CAN HANDLE THIS ENTIRE DATA PIPELINE, WE REALLY BELIEVE THAT THE LATTER HALF OF THIS IS MOST IMPORTANT BECAUSE PERSONALIZED EXPERIENCES ARE THE MOST IMPACTFUL WHEN DELIVERED AT THE MOMENT. THAT MATTERS TO MOST, TO THE CUSTOMER. AND SO THIS LAST STEP HERE ACT IS CRITICAL. AND
AS SUCH, WE'VE CREATED THE ABILITY TO PROACTIVELY TRIGGER OUR ACTIONS BASED ON ANY AND ALL DATA AND INSIGHT CHANGES. AS A CUSTOMER STATUS, CHANGES AS A PROPENSITY SCORE CHANGES, FOR EXAMPLE, IN THE MOMENT WE TRIGGER THESE DATA SIGNALS OUT EVERYWHERE TO ALL THE APPLICATION SYSTEMS, INCLUDING SALESFORCE, SES OVER THE LAST YEAR WE'VE WORKED WITH AMAZON SAGEMAKER TO BRING THIS BRING YOUR OWN AI CAPABILITY, WHICH DRASTICALLY REDUCES THE OPERATIONAL BURDEN OF MAINTAINING AND USING AI. WE WORK WITH AMAZON SAGEMAKER BECAUSE OUR CUSTOMERS LOVE IT AND BECAUSE IT'S EASY TO CREATE A SEAMLESS EXPERIENCE FOR OUR JOINT USERS. WITH OUR INTEGRATION, YOU CAN ACCESS DATA, CLOUD DATA, ZERO-ETL STYLE BUILD AI MODELS IN SAGEMAKER AND THEN ONLY INVOKE THOSE MODELS INFERENCES AS NEEDED AND IN REAL TIME FROM DATA CLOUD. FOR
EXAMPLE, IF YOU HAVE 50 MILLION CUSTOMERS IN 20 PRODUCT CATEGORIES, FIVE AI MODELS, SO 5 BILLION SCORES IN TOTAL TRADITIONAL AI, ML OPS PROBABLY HAVE YOU BATCH SCORING ALL OF THESE 5 BILLION COMBINATIONS ACROSS THE BOARD EVERY SINGLE DAY AND THEN COPYING THEM EVERYWHERE TO DIFFERENT SYSTEMS JUST IN CASE ANY ONE OF THOSE SCORES NEEDS TO BE USED IN ANY ONE OF THOSE SYSTEMS, ANY GIVEN DAY. WITH OUR INTEGRATION. YOU DON'T HAVE TO DO THAT. LET'S TAKE A LOOK AT IT IN ACTION IN THE FIRST TWO STEPS HERE, OUR USERS ACCESS AND EXPLORE DATA CLOUD DATA AND INSIGHTS WITHIN DATA WRANGLER, WHERE THEY CAN DO FURTHER AI FEATURE ENGINEERING AS NEEDED. THIRD STEP DATA SCIENTISTS TRAIN AND BUILD AI MODELS IN SAGEMAKER STUDIO. WHEN A MODEL IS READY FOR PRODUCTION, A PROJECT TEMPLATE AUTOMATES THE DEPLOYMENT OF AN INVOCATION ENDPOINT, WHICH CAN THEN BE REGISTERED IN DATA CLOUD. LAST STEP THERE. NOW WE HAVE SEEN
MANY WONDERFUL REAL WORLD APPLICATIONS OF THIS FEATURE. IN ONE CASE, A COMPANY STREAMS OR DEVICE DATA INTO DATA CLOUD, WHICH THEN ENRICHES THEM WITH UNIFIED PROFILE DATA. CALCULATE SOME REAL TIME INSIGHTS, THEN DECIDES TO INVOKE A RETENTION CHURN MODEL INSIDE AWS AND RUNS A CAMPAIGN TO TARGET THOSE CUSTOMERS AT RISK OF CHURN, AS WELL AS EXPOSING ALL OF THOSE ENGAGEMENT EVENTS, TIMELINE INSIGHTS IMMEDIATELY TO THE SALES AND SERVICE REPS IN THE SAME WAY WE'VE AUTOMATED PREDICTION BASED TRIGGERS, WE'RE NOW LOOKING AT GENAI SO THAT THE RIGHT ACTION AT THE RIGHT MOMENT CAN BE COMBINED WITH THE RIGHT CONTENT. EARLIER THIS YEAR, OUR COMPANY ANNOUNCED A FORAY INTO
THE WORLD OF GENAI WITH EINSTEIN , GPT. NOW THIS BRING YOUR OWN AI CAPABILITY. CAN INCORPORATE ACCESS TO A WIDE RANGE OF LARGE LANGUAGE MODELS FROM AMAZON BEDROCK INCLUDING AMAZON TITAN. WITHOUT YOU HAVING TO MANAGE INFRASTRUCTURE YOURSELF AS YOU SAW EARLIER, YOU CAN EASILY AND SECURELY ACCESS DATA CLOUD DATA FROM YOUR AWS. NOW NOW JUST AS EASILY AND SECURELY. YOU CAN USE
THOSE DATA TO FINE TUNE MODELS WITHIN BEDROCK CUSTOMIZED MODELS OF YOUR CHOICE INSIDE BEDROCK CAN THEN EASILY BE REGISTERED AS INFERENCE POINTS IN DATA CLOUD. JUST THE SAME FOR USE IN SALESFORCE. WHAT MIGHT IT LOOK LIKE IN ACTION? SUPPOSE RIGHT NOW YOU HAVE A CUSTOMER IMPORTANT CUSTOMER BROWSING ON YOUR WEBSITE FOR ONE OF YOUR COMPANY'S PRODUCT INFO. IT
TRIGGERS REAL TIME INSIGHTS AND DECISIONING IN DATA CLOUD, WHICH THEN DECIDES TO INVOKE THE PROPENSITY TO BUY MODEL, THIS TIME IN AWS, THE RETRIEVED PROPENSITY SCORE ALONG WITH THE PRODUCT INFO ITSELF, ARE BOTH FED INTO A GENAI PROMPTS AND A CALL TO AMAZON. BEDROCK IS MADE WHERE A FINE TUNED MODEL TAILORED FOR YOUR COMPANY'S NEEDS IS WAITING AND IS USED TO GENERATE AMAZING CONTENT. THIS ENTIRE FLOW IS AUTOMATED AND AT THE END A SALES REP GETS NOTIFIED AND SEES A GENAI SUGGESTED EMAIL ALONG WITH ALL THE ENGAGEMENT EVENTS TIMELINE INSIGHTS SO THEY KNOW WHAT LED TO THIS MOMENT AND WHY THEY'RE GETTING THE EMAIL WHICH THEY CAN THEN USE TO CONTACT THE CUSTOMER AT EXACTLY THE RIGHT TIME WITH THE MOST RELEVANT CONTENT AT. WE
BELIEVE THAT GENAI CAN BE A GENERATIONAL OPPORTUNITY TO CREATE THE MOST INCREDIBLE CUSTOMER EXPERIENCES AND AN AI IS GOING TO NEED GREAT DATA AND AN ACTION SYSTEM TO DELIVER THOSE TIMELY, RELEVANT MOMENTS. LOOKING FORWARD, SALESFORCE DATA CLOUD WILL POWER EINSTEIN GPT GROUNDED WITH REAL TIME AND HARMONIZED DATA ALONG WITH THE ABILITY TO BRING YOUR OWN AI SUCH AS THAT FROM SAGEMAKER, OR ACCESS LLM SUCH AS THAT FROM BEDROCK AWS IS REALLY HELPING US SEIZE THIS OPPORTUNITY TO CREATE A SOLUTION THAT HELPS COMPANIES UNIFY THEIR DATA AND DRIVE AI OPTIMIZED INTERACTIONS THAT SCALE IN A MEANINGFUL MANNER THANK YOU, THANK YOU VERY MUCH. THANK YOU GABRIELLE. I SHOULD START WITH SALESFORCE.
FINE TUNING IS ONE WAY TO CUSTOMIZE THESE FOUNDATION MODELS FOR YOUR USE CASES AND CREATE INNOVATIVE CUSTOMER EXPERIENCES. WHILE THESE FOUNDATION MODELS ARE INCREDIBLY POWERFUL AND HAVE AND HAVE A ROBUST UNDERSTANDING OF NATURAL LANGUAGE THEY STILL REQUIRE A LOT OF MANUAL PROGRAMMING TO COMPLETE COMPLEX TASKS. LIKE BOOKING A FLIGHT, OR PROCESSING AN INSURANCE CLAIM. THATS BECAUSE, OUT OF BOX FMS ARE NOT ABLE TO ACCESS UP TO DATE KNOWLEDGE LIKE RECENT COMPANY SPECIFIC DATA.
THEY ARE ALSO UNABLE TO TAKE SPECIFIC ACTIONS TO FULFILL YOUR CUSTOMER REQUEST. TO MAKE THIS HAPPEN, DEVELOPERS NEED TO FOLLOW A NUMBER OF RESOURCE INTENSIVE STEPS. LET ME WALK THROUGH A SPECIFIC EXAMPLE. IN THIS SCENARIO, A CUSTOMER WANTS TO EXCHANGE BLACK SHOES FOR BROWN SHOES PURCHASED THROUGH AN ONLINE RETAILER. THEY USE THE SITE'S CUSTOMER SERVICE CHAT INTERFACE TO COMMUNICATE THEIR REQUESTS. CONFIRM THE AVAILABILITY OF THEIR DESIRED
COLOR AND SHOE SIZE AND RECEIVE THE EXCHANGE. THESE REQUESTS SOUND PRETTY REASONABLE, BUT TO CODE IT UP ON A REGULAR SOFTWARE, IT WILL TAKE MONTHS. BUT FOUNDATIONAL MODELS IT CAN BE ACTUALLY DONE FASTER. SO YOU WOULD THINK IT'S PRETTY SIMPLE, RIGHT? WELL, THIS LOOKS LIKE AN EASY EXCHANGE FOR FOUNDATIONAL MODELS. DEVELOPERS NEED TO FOLLOW A SERIES OF TIME
CONSUMING TASKS LIKE DEFINING INSTRUCTIONS AND ORCHESTRATION, CONFIGURING THESE FOUNDATIONAL MODELS TO ACCESS COMPANY SPECIFIC DATA SOURCES AND WRITING CUSTOM CODE TO EXECUTE VARIOUS ACTIONS THROUGH A SERIES OF API CALLS. AND FINALLY, DEVELOPERS MUST SET UP CLOUD POLICIES AND HOSTING AND SECURITY CONTROLS. ALL OF THESE STEPS CAN TAKE WEEKS, IF NOT MORE. LET'S TAKE A LOOK AT EACH OF THEM MORE IN DETAIL. FOR THE FIRST STEP, DEVELOPER NEEDS TO ORCHESTRATE A COMPLEX SYSTEM BETWEEN THE FOUNDATION MODELS, SOFTWARE SYSTEMS AND THE END USERS. TO COMPLETE THIS TASK, THEY NEED TO ENABLE THE FOUNDATION MODEL TO BREAK DOWN THEM INTO MULTIPLE STEPS AND GIVE INSTRUCTIONS LIKE YOU ARE A SHOE RETURN AGENT AS WELL AS CONTEXTUAL HISTORY FROM DIFFERENT CHAT SESSIONS. WHILE
THIS IS CRITICAL FOR FM TO PRODUCE A BEST RESPONSE, THE PROCESS OF CRAFTING AND REFINING CAN TAKE WEEKS. THAT'S BECAUSE THIS IS VERY MANUAL REQUIRES A LOT OF TRIAL AND ERROR AND EACH FM REQUIRES A DIFFERENT FORMAT. THEN YOU NEED TO MAKE SURE FMS HAVE THE RIGHT DATA TO FULFILL THE REQUESTS. DEVELOPERS NEED TO CONFIGURE THE FM TO HAVE ACCESS TO UP TO DATE KNOWLEDGE SOURCES WHICH IS IN THIS CASE IS THE ONLINE RETAILER WHOSE MOST RECENT RETURN POLICY AND CUSTOMER HISTORY. THIS PROCESS
IS KNOWN AS RETRIEVAL AUGMENTED GENERATION, ALSO KNOWN AS RAG RAG ENABLES FOUNDATIONAL MODELS TO LEVERAGE THE MOST RECENT DATA, CONVERT THE DATA INTO MACHINE READABLE FORMAT AND THEN PROGRAM THE FM TO QUERY THE DATA SOURCE FOR A MORE ACCURATE RESPONSE. THEN TO COMPLETE THE TASK AND EXCHANGE THE SHOES, DEVELOPERS NEED TO WRITE CODE SO THAT THE FMS CAN TAKE ACTION AND MAKE SERIES OF API CALLS LIKE PLACING A CUSTOMER ORDER CONFIRMING THE ORDER AND CALLING TO SEND A CONFIRMATION EMAIL. ALL OF THESE NEEDS TO BE DONE IN THE RIGHT SERIES. FINALLY,
DEVELOPERS MUST HOST ALL OF THESE AGENTS IN AND THEIR GENERATIVE AI APPS SET UP THE RIGHT POLICIES, SECURITY SETTINGS, ALERTS, NETWORKING, ALL THESE THINGS. SO WHAT STARTED OUT LOOKING LIKE A PRETTY SIMPLE TASK WITH THESE FOUNDATIONAL MODELS IS ACTUALLY PRETTY COMPLICATED. WHO WANTS TO DO THAT FOR EVERY GENERATIVE AI POWERED TASK? WE BELIEVE THIS SHOULD BE A LOT EASIER FOR DEVELOPERS TO DO. THAT'S WHY
TODAY I'M VERY EXCITED TO ANNOUNCE THE PREVIEW OF AGENTS FOR AMAZON BEDROCK. THIS IS A NET NEW CAPABILITY FOR DEVELOPERS TO ENABLE GENERATIVE AI APPLICATIONS TO COMPLETE TASKS IN JUST A FEW CLICKS WITH A FEW CLICKS. AGENTS FOR BETTER OR CONFIGURE YOUR FOUNDATIONAL MODELS TO AUTOMATIC BREAK DOWN AND ORCHESTRATE THESE TASKS WITHOUT HAVING TO WRITE ANY CODE. THE AGENT SECURELY CONNECTS YOUR FM TO THE RIGHT DATA SOURCE THROUGH A SIMPLE API. AUTOMATICALLY CONVERTS YOUR DATA INTO MACHINE READABLE FORMAT AND AUGMENTS THE USER'S REQUEST WITH RELEVANT INFORMATION TO GENERATE A MORE ACCURATE RESPONSE. AND AGENTS IN BEDROCK CAN TAKE ACTION BY
AUTOMATICALLY MAKING API CALLS ON YOUR BEHALF AND YOU DO NOT HAVE TO WORRY ABOUT COMPLEX SYSTEMS AND HOSTING THEM BECAUSE IT IS FULLY MANAGED. NOW LET'S TAKE A LOOK AT THE SAME EXAMPLE NOW WITH THE AGENTS FOR BEDROCK . WITH THIS FUNCTIONALITY, THE DEVELOPER ONLY NEEDS TO FOLLOW A REALLY FEW SIMPLE STEPS. FIRST, THEY USE THE BEDROCK CONSOLE TO SELECT THE DESIRED FM, THEN TO DEFINE INSTRUCTIONS OR ORCHESTRATION. THEY USE A SETUP
WIZARD AND GIVE BASIC INSTRUCTIONS LIKE YOUR APPLEID CUSTOMER SERVICE AGENT AND UPDATE INVENTORY LEVELS. THEN THE DEVELOPER WILL SELECT DATA SOURCES LIKE RETURN POLICIES WITHOUT WRITING ANY CODE THAT ENABLES THE FM TO PREVENT INFORMATION. AND THEN FINALLY, THE DEVELOPER SIMPLY SPECIFY THE FUNCTIONS TO EXECUTE THESE API CALLS WITH THESE SIMPLE BASIC STEPS. DEVELOPER IS ABLE TO QUICKLY FULFILL THE CUSTOMER
REQUIREMENT WITHOUT WEEKS OF MANUAL CODING. I AM EXCITED FOR OUR CUSTOMERS TO DRIVE INNOVATION WITH THIS NEW CAPABILITY WHILE ALSO REDUCING THE HEAVY LIFTING ASSOCIATED FOR DEVELOPMENT TEAMS. BUT THERE IS ANOTHER CRITICAL ELEMENT UNDERPINNING ALL OF THIS INNOVATION THAT WILL ALLOW YOU TO DERIVE EVEN MORE VALUE WITH GENAI. THAT IS YOUR DATA. AS WE SAW WITH BEDROCK AGENTS AND SALESFORCE SOURCE, WHILE FMS ARE INCREDIBLY POWERFUL OUT OF THE BOX TO BE TRULY USEFUL TO YOUR ORGANIZATION, THEY NEED ACCESS TO THE RIGHT DATA SOURCE. YOUR
DATA IS YOUR DIFFERENTIATOR FOR GENAI AND TO ENSURE YOU HAVE RIGHT RELEVANT HIGH QUALITY DATA TO TRAIN YOUR MODELS OR CUSTOMIZE THESE FMS FOR YOUR USE CASES, YOU NEED A STRONG DATA FOUNDATION. TO BUILD A STRONG DATA FOUNDATION, YOU NEED ACCESS TO COMPRESS SENSITIVE SET OF DATA SERVICES THAT ACCOUNT FOR THE SCALE VOLUME AND OF YOUR USE CASES. THIS IS FAIR OFFERS A BROAD SET OF DATA CAPABILITIES THAT SUPPORT YOUR END TO END DATA JOURNEY FROM STORING, QUERYING AND ANALYZING DATA TO PUTTING YOUR DATA INTO WORK THROUGH BUSINESS INTELLIGENCE, MACHINE LEARNING AND GENERATIVE AI. WE ALSO HAVE SERVICES THAT
HELP YOU EASILY INTEGRATE AND GOVERN YOUR DATA. THE GOOD NEWS IS THAT ALL THE INVESTMENTS YOU HAVE MADE TO BUILD THIS STRONG FOUNDATION WILL SERVE YOU WELL FOR GENAI. BUT THERE IS ONE TOOL YOU WILL LIKELY NEED TO ADD TO YOUR DATA FOUNDATION. EVEN IF
YOU HAVE NOT DONE THAT ALREADY. WE TOOLS FOR STORING AND RETRIEVING YOUR VECTOR EMBEDDINGS VECTOR EMBEDDINGS ARE PRODUCED BY FOUNDATION MODELS AND USED IN GENAI APPLICATIONS TO PRODUCE MORE RELEVANT RESPONSES TO YOUR END USERS. LET'S TAKE A QUICK LOOK AT HOW THEY WORK. THINK OF VECTOR EMBEDDINGS AS NUMERICAL REPRESENTATIONS FOR TEXT, IMAGE AUDIO, AND VIDEO DATA. WELL HUMANS CAN UNDERSTAND THE
MEANING AND CONTEXT OF WORDS, AND MACHINES CAN ONLY UNDERSTAND NUMBERS. SO WE HAD TO TRANSLATE THEM INTO A FORMAT THAT'S SUITABLE FOR ML BY ASSIGNING A NUMBER TO THE DIFFERENT FEATURE OF EACH WORD. WE CAN VIEW VECTORS IN A MULTI DIMENSIONAL SPACE AND MEASURE THE DISTANCE BETWEEN THEM. WE WORDS THAT ARE RELATED IN CONTEXT WILL HAVE VECTORS THAT ARE CLOSER TOGETHER, WHICH HELPS MACHINE UNDERSTAND THE SIMILARITIES AND DIFFERENCES BETWEEN WORDS. FOR INSTANCE, A CAT IS CLOSER TO A KITTEN, WHEREAS DOG IS CLOSER TO A PUPPY. BY COMPARING EMBEDDINGS IN THIS WAY, THE MODEL WILL PRODUCE MORE RELEVANT AND CONTEXT ABLE RESPONSES THAN WORD MATCHING. WHILE EMBEDDINGS ARE
NOT NEW FOR MACHINE LEARNING, BASED APPS THAT IMPART IS GROWING FAST WITH THE AVAILABILITY OF GENERATIVE AI AND NATURAL LANGUAGE PROCESSING . FOR EXAMPLE, THEY CAN SUPERPOWERS SEMANTIC SEARCH FOR USE CASES LIKE RICH MEDIA, SEARCH AND PRODUCT RECOMMENDATION. IN THIS SCENARIO, YOU CAN SEE THAT SEMANTIC SEARCH GREATLY ENHANCES THE ACCURACY OF THE OUTPUT QUERY FOR BRIGHT COLORED GOLF SHOES. IN ADDITION TO SEMANTIC SEARCH EMBEDDINGS CAN BE USED TO AUGMENT YOUR PROMPTS FOR MORE ACCURATE RESULTS THROUGH RAG. AS WE SAW IN THE AGENT'S EXAMPLE. BUT IN ORDER TO USE THEM, YOU
WILL NEED TO STORE THEM IN A DATABASE WITH VECTOR CAPABILITY . TODAY AWS OFFERS VECTOR DATABASE CAPABILITIES FOR POPULAR SERVICES LIKE AMAZON OPENSEARCH SERVICE, AURORA POSTGRES AND POSTGRES. CUSTOMERS CAN USE THESE SERVICES TO STORE AND SEARCH EMBEDDINGS USED IN THEIR ML AND GENAI APPS. CO-LOCATING YOUR VECTOR EMBEDDINGS WITH YOUR DATA MAKES THE PROCESS OF WITH EMBEDDINGS A LOT EASIER AND REDUCES DATA DUPLICATION. YOU ALSO DON'T NEED TO WORRY ABOUT THE MAINTENANCE VERSION LEARNING AND LICENSING OF A SEPARATE DATABASE, LET ALONE KEEPING THEM IN SYNC. SO NOW LET'S DIVE A LITTLE DEEPER INTO HOW VECTORS WORK ON OUR OPENSEARCH SERVICE. OPENSEARCH
SERVICE IS A FULLY MANAGED SERVICE FOR REAL TIME SEARCH MONITORING AND ANALYSIS OF BUSINESS AND OPERATIONS DATA. TODAY, CUSTOMERS USE OPENSEARCH TO STORE MULTIDIMENSIONAL VECTOR EMBEDDINGS TO POWER APPLICATIONS LIKE MULTIMODAL SEARCH PERSONAL ASSISTANTS AND RECOMMENDATION ENGINES. OPENSEARCH CUSTOMERS LIKE THAT, THEY CAN STORE THEIR VECTORS WITH THEIR DATA ON A DAILY BASIS, BUT STORING BILLIONS OF VECTOR EMBEDDINGS AND QUICKLY SEARCHING THOSE EMBEDDINGS WITH OPENSEARCH SERVICE REQUIRES DEVELOPERS TO CONFIGURE, MANAGE AND SCALE THESE CLUSTERS. THAT MEANS THAT YOU REQUIRE DEDICATED RESOURCES OR EXPERTISE THAT NOT ALL ORGANIZATIONS HAVE TO MAKE. GENERATIVE AI ACCESSIBLE TO MORE BUILDERS. WE WANTED TO MAKE IT EASIER TO LEVERAGE THE POWERFUL VECTOR CAPABILITIES IN OUR OPENSEARCH SEARCH. THAT'S WHY TODAY I'M VERY EXCITED TO
ANNOUNCE THE PREVIEW OF OUR VECTOR ENGINE FOR OPENSEARCH SERVERLESS. SO THIS VECTOR ENGINE OFFERS SIMPLE, SCALABLE AND HIGH PERFORMING VECTOR STORAGE AND SEARCH WITHOUT HAVING TO MANAGE ANY INFRASTRUCTURE. DEVELOPERS CAN STORE VECTORS ALONGSIDE BUSINESS DATA AND TEXT MAKING IT REALLY EASY. EMBEDDINGS METADATA THE ASSOCIATED DATA ALL FROM A SINGLE API CALL. AND BECAUSE IT'S SERVERLESS, THE DEVELOPERS DON'T NEED TO MANAGE CLUSTERS OR WORRY ABOUT PRODUCTION SCALE. SO NOW NOT ONLY ARE YOU GOING TO NEED TOOLS LIKE VECTOR DATABASES AND A COMPREHENSIVE SET OF SERVICES TO SUPPORT YOUR GENAI STRATEGY, BUT YOU ALSO WANT TO MAKE SURE THAT THOSE DATA SERVICES AND DATA THEY STORE ARE INTEGRATED. WHEN YOU CONNECT THE
DOTS WITH YOUR DATA ACROSS YOUR DEPARTMENTS, SERVICES OR ON PREM DATABASES AND THIRD PARTY APPLICATIONS, YOU'RE ABLE TO POWER GENERATIVE AI TO CREATE REMARKABLE EXPERIENCES AS WE HAVE BEEN WORKING ON THIS DATA INTEGRATION PROBLEM FOR A WHILE NOW. TODAY WE OFFER A FEDERATE AND QUERY CAPABILITY IN AMAZON REDSHIFT, OUR PETABYTE SCALE DATA WAREHOUSE AND AMAZON. ATHENA SO YOU CAN RUN QUERIES ACROSS THESE SERVICES AND A WIDE RANGE OF DATA SOURCES AND THIRD PARTY APPS. WE HAVE INTEGRATED SAGEMAKER WITH REDSHIFT AND AURORA, SO WITH A SIMPLE SQL PROMPT, YOU CAN MAKE PREDICTIONS WITHOUT HAVING TO WRITE ANY CODE AND AURORA NOW SUPPORTS ZERO-ETL INTEGRATE WITH REDSHIFT SO THAT YOU CAN BRING YOUR TRANSACTIONAL DATA FROM AURORA TO DO REAL TIME ANALYTICS IN REDSHIFT IT. BUT FOR SOME USE CASES, BRINGING
DATA TOGETHER IS JUST A PART OF SOLUTION. FOR EXAMPLE, YOU MAY NEED TO PERFORM ADDITIONAL WORK TO MATCH AND LINK RELATED DATA RECORDS. THIS IS A COMMON PROBLEM WE SEE ACROSS FINANCIAL SERVICES, RETAIL AND ADVERTISING WHERE RECORDS ARE OFTEN SILOED ACROSS APPLICATIONS, CHANNELS AND DATA STORES. SO IMAGINE I RUN A CUSTOMER EXPERIENCE AT AN AIRLINE AND I WANT TO DELIVER MORE RECENT RELEVANT TRIP RECOMMENDATIONS TO MY CUSTOMERS . I WILL NEED TO INCORPORATE RECENT INTERACTIONS FROM THE LOUNGE EXPERIENCES, LOYALTY PROGRAMS AND CUSTOMER SUPPORT INTO A UNIFIED PROFILE TO DELIVER THE BEST TRIP OPTIONS TO EACH CUSTOMER. HOWEVER FOR THESE DISPARATE RECORDS, WHICH OFTEN CONTAIN INCOMPLETE OR CONFLICTING INFORMATION WHICH CREATES THE PROBLEM OF MATCHING REALLY DIFFICULT TO UTTERS, THESE COMPANIES SPEND MONTHS OR DEVELOPMENT TIME TO BUILD COMPLEX MATCHING RULES AND SYSTEMS THAT ARE TEND TO BE VERY FRAGILE. SO WE WANTED TO MAKE THIS PROCESS EASIER. SO TODAY
I'M EXCITED TO ANNOUNCE THE GENERAL AVAILABILITY OF ENTITY RESOLUTION IN. ML POWERED SERVICE THAT HELPS COMPANIES EASILY MATCH AND LINK RELATED DATA RECORDS. WITH THIS NEW SERVICE, YOU CAN NOW SET UP EASY TO USE ENTITY RESOLUTION WORKFLOWS IN JUST MINUTES INSTEAD OF WEEKS. ENTITY RESOLUTION USES MATCHING TECHNIQUES LIKE RULE BASED MATCHING TO QUICKLY LINK RELATED INFORMATION INTO A UNIFIED ID. YOU CAN ALSO APPLY A POWERFUL ML MODEL TO MATCH RELATED DATA SETS WHEN THERE ARE INCOMPLETE OR CONFLICTING INFORMATION. ENTITY
RESOLUTION READS RECORDS DIRECTLY FROM S3 PROTECTING YOUR DATA BY MINIMIZING DATA, DUPLICATION AND SOON WE WILL ADD PARTNER INTEGRATIONS WITH LIVERAMP AND TRANSUNION AND AN INTEGRATION WITH OPEN SOURCE ADVERTISING FRAMEWORK LIKE UNIFIED ID 2.0 TO TRANSLATE OR ENRICH DATA RECORDS WITH COMMON INDUSTRY IDENTIFIERS. WHEN YOU COMBINE THE POWER OF ML WITH YOUR DATA, YOU ARE ABLE TO ACCOMPLISH SOME PRETTY AMAZING THINGS. NOW I'D LIKE TO WELCOME ANOTHER CUSTOMER TO THE STAGE
THAT IS USING THE DATA TO BUILD AN AI POWERED INFRASTRUCTURE AND BUILD INNOVATIVE GENAI APPLICATIONS. PLEASE WELCOME LINDSAY SILVER FROM FOX. THANK YOU ALL FOR BEING HERE. THANK YOU ALL FOR BEING HERE AND JOINING ME TODAY TO TALK ABOUT AI ON ONE OF THE MOST IMPORTANT DAYS, ONE OF THE MOST IMPORTANT YEARS, I THINK WE CAN SAY WE'VE HAD IN THE AI SPACE. I'M LINDSAY SILVER, THE HEAD OF DATA AT FOX AND I'M REALLY EXCITED TO TALK TO YOU TODAY ABOUT HOW WE'RE USING AI TO CHANGE THE FACE OF MEDIA AND GIVE SUPERPOWERS TO OUR BROADCASTERS, ADVERTISERS AND PRODUCTS. IF YOU DON'T KNOW FOX FOX IS ONE OF THE US'S LARGEST MEDIA COMPANIES. WE HAVE BRANDS THAT SPAN ENTERTAINMENT, SPORTS, NEWS AND NOW A TO BE AN INCREDIBLE PLATFORM FOR FREE CONTENT ON THE WEB. SO I INVITE
YOU TO TRY IT OUT. IT'S GREAT. WE HAVE SOME NEW MOVIES ON THIS YEAR THAT ARE REALLY COOL. WE HAVE A HUGE FOOTPRINT. WE HAVE OVER 300 MILLION MONTHLY ACTIVE USERS. THAT MEANS 300 MILLION PEOPLE SEE ONE CONTENT ON FOX EVERY MONTH. WE HAVE OVER 600 OR 6 MILLION PEOPLE WHO SEE OUR NFL BROADCASTS EACH GAME IN THE FALL, AND WE COLLECT OVER 50,000 DATA POINTS PER SECOND ABOUT OUR AUDIENCES, ABOUT OUR CONTENT, ABOUT OUR BUSINESS AND OUR OPERATIONS. THAT'S A MASSIVE FOOTPRINT. BUT WE'RE NOT HERE TO TALK ABOUT DATA AS AN INGESTION POINT. WE'RE HERE TO TALK ABOUT
HOW TO LEARN FROM DATA AND HOW TO USE AI TO MAKE A BUSINESS BETTER. AT FOX, I LIKE TO THINK ABOUT LEARNING AS A HUMAN PROCESS. SO I'D LIKE TO JUMP BACK TO THE VERY START OF TIME WHEN HUMANS STARTED DOCUMENTING CROP CYCLES, DOCUMENTING TAX ROLLS IN AN EFFORT TO AUTOMATE THE PROCESS OF LEARNING IN SOCIETY. WE'VE BEEN DOING THIS AS HUMANS FOREVER. BEAR IN 1923, NIELSEN CAME AROUND AND IN THE MEDIA INDUSTRY THEY STARTED DOCUMENTING HOW MEDIA WAS PERFORMED IN AN EFFORT TO HELP ADVERTISERS AND MEDIA COMPANIES LEARN AND GET FEEDBACK ON THEIR ON DATA IN REAL TIME. THIS WAS A HUMAN PROCESS. IT WAS SLOW. IT WAS BLOCKY. IT WASN'T UNTIL THE 80S THAT WE STARTED AUTOMATE ING
THESE STEPS. WE STARTED GETTING BETTER AT OBSERVATION PRODUCTS. WE STARTED GETTING BETTER AT SYNTHESIZING DATA WITH FORECASTING, WITH METEOROLOGY AND FINANCE. WE STARTED GETTING BETTER AT MAKING AUTOMATED DECISIONS IN ADVERTISING, AND THEN WE STARTED ACTING AGAIN. WE'RE TURNING A CORNER IN THE
LAST COUPLE OF YEARS FOR THE FIRST TIME EVER, LARGE LANGUAGE MODELS AND GANS HAVE ALLOWED US TO TAKE DATA AND GO STRAIGHT FROM OBSERVATION BACK TO A PRODUCT THAT WE CAN ACT ON. IT MEANS GENERATING CONTENT GENERATING RESPONSE PIECES IN CHAT, GENERATING IMAGES THAT WE CAN THAT WE CAN TAKE DIRECTLY BACK AND ACT ON AT. THIS IS A HUGE BOON TO OUR BUSINESS, BUT WE VIEW IT DIFFERENTLY THAN SOME MIGHT THINK AS A WAY TO SUPERCHARGE HOW OUR PRODUCERS AND OUR EDITORS AND OUR BROADCASTERS AND ADVERTISERS WORK WITH US IN ORDER TO DO THAT, WE NEED FOUNDATIONS. WE NEED A PLATFORM FOR DRIVING THIS CHANGE. AND THAT STARTS WITH OUR DATA SOURCES. FOX TAKES IN
HUNDREDS OF DIFFERENT DATA SOURCES ACROSS OUR BRANDS, ACROSS OUR OPERATIONS AND FEEDS THEM INTO AN INFRASTRUCTURE LARGELY BUILT ON AMAZON. AND WE'RE EXTREMELY PROUD OF THE INFRASTRUCTURE WE'VE BUILT. BUT ALSO PROUD OF THE RELATIONSHIP AND THE TECHNOLOGIES THAT AMAZON PROVIDES US WITH FROM DATA INFRASTRUCTURE THROUGH TO OUR AI AND ML TOOLCHAIN, AND THEN ALL THE WAY UP TO OUR APPLICATION STACKS. THESE LET US DO THINGS
THAT OTHERWISE WOULD NEVER BE POSSIBLE AT A COMPANY LIKE FOX, WE USE THIS INFRASTRUCTURE AS PART OF A LARGER PLATFORM THAT INCLUDES GROWING AI CATALOG OF PROPRIETARY FOX BUILT MODELS THAT SPAN SPORTS CONTENT AUDIENCE AND NOW GENERATIVE AI WE PLUG THOSE IN THROUGH PREDICTIVE APIS, TWO TECHNOLOGIES THAT WE'VE BUILT THAT SPAN OUR OPERATION SPAN, OUR ADVERTISING BUSINESS, AND ALSO OUR CONSUMER PRODUCTS. AND THEN WE BRING THOSE TO OUR CONSUMER INTERFACES THROUGH DYNAMIC RECOMMENDATION ACTIONS, THROUGH SPORTSCASTING APPLICATIONS AND OPERATIONAL CHANGES. IT'S A POWERFUL SYSTEM THAT THEN FEEDS BACK INTO OUR CORE DATA SETS AND THAT FEEDBACK LOOP BECOMES THE CORE OF HOW FOX USES DATA, AND POWERS AI.
SO LET’S SEE WHAT WE CAN DO WITH THIS PLATFORM. FORESIGHT. FORESIGHT HAS BEEN A PRODUCT THAT'S BEEN IN THE WORKS AT FOX FOR SEVERAL YEARS, IS NOW POWERING SPORTSCASTERS ACROSS MULTIPLE SPORTS THAT FOX BROADCASTS. IT'S A HEADS UP INTERFACE THAT ALLOWS US TO USE AI TO PROVIDE DEEP KNOWLEDGE AND INSIGHT TO OUR SPORTS CASTERS AND TO CREATE GRAPHICS ON THE FLY. THAT OVERLAY THE GAME WITH IMPORTANT CONTEXT ABOUT THE PLAYERS, ABOUT THE REFEREES AND THE MATCHUPS. IT'S REALLY EXCITING PRODUCT AS WE GET INTO OUR NEXT GENERATION OF FORESIGHT, WE'RE LOOKING TOWARD LMS AND GANS TO HELP US CREATE EVEN MORE DYNAMIC CONTENT WITH THIS PRODUCT. FOX IS A MEDIA
COMPANY. WOULDN'T BE ANYTHING WITHOUT OUR ADVERTISERS. WE TRY TO CREATE EXPERIENCES THAT ARE THE BEST, BOTH FOR OUR END USERS AND CONSUMERS, AS WELL AS OUR ADVERTISING CLIENTS. FOX ATLAS ALLOWS US TO DO THAT. ATLAS TAKES THE DATA THAT WE KNOW ABOUT OUR VIDEOS AT ANY GIVEN POINT IN TIME, MAKES IT ACCESSIBLE TO ADVERTISERS SO THAT THEY CAN TARGET VERY SPECIFIC POINTS IN OUR VIDEO WHERE THEIR CONTENT WILL ALIGN.
WELL, THIS COULD BE ANYTHING FROM A MOMENT IN THE SUPER BOWL WHERE WE HAVE A WHERE WE HAVE A TOUCHDOWN TO A MOMENT IN A SHOW OR A PROGRAM THAT MENTIONS THEIR PRODUCT. IT'S AN EXTREMELY EXCITING PRODUCT. IT'S IN PRODUCTION NOW AND CONTINUALLY EVOLVING AS WE GO. LASTLY ONE OF MY FAVORITES AND ONE I'M EXTREMELY PROUD OF ARE CATCH UP WITH HIGHLIGHTS PRODUCT. THIS IS A DIRECT TO CONSUMER PRODUCT THAT WE LAUNCHED LAST YEAR AT THE WORLD CUP AND IT'S RUNNING AGAIN AT THE WOMEN'S WORLD CUP. NOW THE CHALLENGE WITH ANY SORT OF SPORT IS THAT PEOPLE OFTEN TUNE IN MID GAME. HERETOFORE, OUR EDITORS WOULD HAVE TO
PRODUCE SUMMARIZED VERSIONS OF THAT GAME AND CATCH UP MOMENTS ALL THROUGHOUT THE GAME. AND IT WAS AN EXTREMELY LABORIOUS PROCESS. NOW, USING AI THAT SPANS BOTH SPORTS AND COMPUTER VISION, WE'RE TAKING AN AUTOMATICALLY COMPRESSING OUR GAMES AT ANY GIVEN MOMENT IN TIME TO THE MOST IMPORTANT POINTS, ALLOWING OUR CONSUMERS TO CONSUME A SHORT SUMMARY OF THE GAME BEFORE THEY TUNE IN ACROSS ALL OF OUR APPLICATIONS.
SO NEXT TIME YOU WATCH A GAME ON THE WOMEN'S WORLD CUP, IF YOU TUNE IN LATE, THIS WILL BE THE FIRST THING YOU SEE. AND IT'S AN EXTREME, EXCITING MOMENT FOR US ON OUR DATA TEAMS AND OUR USING THE AI. SO WHAT'S NEXT ABOUT GENERATIVE? AI HAS ALLOWED US TO START THINKING ABOUT CONTEXT MORE DEEPLY. FOX CARES ABOUT MOMENTS. WE CARE ABOUT CREATING CONTENT THAT REALLY MATTERS TO PEOPLE. AND WE UNDERSTAND THAT THAT CONTENT IS EXTREMELY RICH. THE END OF THE ARGENTINA - FRANCE GAME LAST YEAR AT THE WORLD CUP WAS ACTUALLY THE CULMINATION OF A HUGE AMOUNT OF CONTEXT. EVERYONE WAS ASKING, WHAT WOULD MESSI DO NEXT?
EVERYONE WAS SAYING, WOW, THAT WAS ONE OF THE GREATEST SHOOTOUTS IN HISTORY. WHEN WAS THE LAST TIME THAT WE HAD A WORLD CUP END? THAT WAY EVERYONE WAS WONDERING ABOUT THE REFEREES WAS ABOUT THE STADIUM. THERE WAS A HUGE AMOUNT OF CONTEXT THERE. THAT'S CONTEXT THAT OUR TEAMS REALLY WANT TO COVER. THEY WANT
TO BRING THAT TO TWO FOUR, BUT IT'S EXTREMELY LABORIOUS TO DO THAT USING GENERATIVE AI FOR THE FIRST TIME, WE'RE ABLE TO START CREATING CONTENT ABOUT THOSE THAT CONTEXT IN REAL TIME. AS MOMENTS HAPPEN SO THAT OUR EDITORS, OUR PRODUCERS AND OUR PRODUCTION TEAMS CAN FOCUS ON THAT CORE MOMENT WHILE HAVING THAT THAT SUPER POWER OF CREATING ADDITIONAL CONTENT. IT'S EXTREMELY COOL. AND WE'RE ROLLING THIS OUT ACROSS OUR BRANDS, ACROSS OUR SPORTS, AND YOU'LL SEE THESE PERCOLATE IN NOT NOT AS A REPLACEMENT TO CONTENT. WE'RE REALLY FOCUSED ON
CONTINUING TO UP THE QUALITY AND THE SCALE OF THE CONTENT. WE CAN PRODUCE. BUT AS A COMPLEMENT TO THAT CONTENT AND YOU'LL SEE THIS THROUGHOUT EVERYTHING THAT WE DO IN THE FUTURE, WE'RE REALLY EXCITED ABOUT IT. WE'RE REALLY EXCITED ABOUT THE TECHNOLOGIES THAT SWAMI AND THE TEAMS HAVE TALKED ABOUT TODAY. WE CAN'T GET ENOUGH OF IT AND WE LOOK FORWARD TO THE FUTURE. THANK YOU SO MUCH FOR HAVING ME. THANK YOU. LINDSAY IT'S GREAT TO
SEE HOW CUSTOMERS LIKE FOX CAN LEVERAGE A STRONG DATA FOUNDATION TO BUILD AI APPS. SO FAR WE HAVE EXPLORED HOW YOU CAN CUSTOMIZE THE FOUNDATIONAL MODEL IN BEDROCK AND LEVERAGE YOUR DATA FOUNDATION TO BUILD POWERFUL NEW CUSTOMER EXPERIENCES. BUT MANY OF OUR CUSTOMERS WILL GET VALUE FROM GENERATIVE AI WHEN IT IS BUILT DIRECTLY INSIDE OUR SERVICES AND APPLICATIONS. BY BUILDING THIS TECHNOLOGY INTO MORE OF OUR SERVICES, WE ARE MAKING THEM EVEN EASIER TO USE. INCREASING
PRODUCTIVITY ACROSS YOUR ORGANIZATION, FOR EXAMPLE, GENAI HAS ENORMOUS POTENTIAL FOR ANALYSTS AND BUSINESS USERS THAT WANT TO QUICKLY ACCESS THAT DATA FOR MORE STRATEGIC DECISION MAKING. EVERY ORGANIZATION I WANT I MEET WITH WANTS TO BECOME MORE DATA DRIVEN COMPANIES THAT CAN TAP THE VALUE OF THEIR DATA CAN BUILD AND INNOVATE FASTER. FOR EXAMPLE, WHEN A SALES TEAM CAN BETTER UNDERSTAND CONVERSION RATES FROM FREE TIER TO PAYING ACCOUNTS, THEY CAN OPTIMIZE MARKETING AND SALES PROGRAM SIMS CUSTOMERS HAVE TOLD US THAT GETTING THESE TYPE OF INSIGHTS FROM THEIR DATA CAN BE REALLY DIFFICULT, EVEN AFTER HUGE AMOUNT OF INVESTMENT IN THEIR ANALYTICS INFRASTRUCTURE. WE
BELIEVE IT SHOULD BE EASIER FOR ORGANIZED OCEANS TO EXTRACT INSIGHTS AND SHARE THEM ACROSS THEIR ORGANIZATION. THAT'S WHY WE BUILT AMAZON QUICKSIGHT QUICKSIGHT IS OUR UNIFIED BUSINESS INTELLIGENCE SERVICE THAT ALLOWS INSIGHTS TO BE SHARED ACROSS THE ORGANIZATION FOR DATA DRIVEN DECISION MAKING . IT HELPS COMPANIES QUICKLY ANALYZE THEIR DATA THROUGH TOOLS LIKE DASHBOARDS. PAGINATED REPORTS AND EMBEDDED ANALYTICS. ONE OF THE MOST POWERFUL CAPABILITIES OF QUICKSIGHT IS Q . Q ENABLES USERS TO ASK ANY QUESTIONS OF THEIR DATA USING NATURAL LANGUAGE AND GET ANSWERS IN JUST SECONDS WITHOUT HAVING TO WRITE SQL QUERIES OR LEARNING BI TOOLS. WE HAVE BEEN USING
GENERATIVE AI MODELS TO POWER Q SINCE ITS LAUNCH IN 2020. SINCE THEN, WE HAVE LEARNED QUITE A BIT ON HOW USERS WOULD LIKE TO USE NATURAL LANGUAGE TO GET VALUE FROM THAT DATA, AND ALSO THESE FOUNDATIONAL MODELS HAVE BECOME MORE AND MORE POWERFUL. THESE TWO THINGS MADE US ASK, SO WITH ALL THE INNOVATION IN GENERATIVE AI, HOW WOULD WE REINVENT EVERY ASPECT OF A BI SYSTEM RIGHT FROM DATA PREPARATION TO DATA ANALYSIS TO DASHBOARD AUTHORING TO CREATING THESE DATA, PRESENTATION AND STORY SHOWS, HOW CAN WE REINVENT BI? THAT'S WHY TODAY I'M EXCITED TO ANNOUNCE S NEW GENERATIVE BI CAPABILITIES AND AMAZON QUICKSIGHT. NOW LET'S TAKE A QUICK LOOK AT A COUPLE OF EXAMPLES OF HOW GENERATIVE BI SIMPLIFIES AND ACCELERATES GETTING INSIGHTS FROM YOUR DATA FOR ANYONE WHO HAS USED BI TOOLS, YOU KNOW, DASHBOARDS ARE EXTREMELY POWERFUL TO SHARE DATA INSIGHTS, BUT BUSINESS ANALYSTS SPEND A LOT OF TIME AND STRUGGLE TO DEVELOP THE RIGHT DATA VISUALS. THEY SPEND SIGNIFICANT
AMOUNT OF TIME EXPLORING DATA AND IDENTIFYING THE RIGHT FIELDS AND FILTERS TO CREATE VISUALS. IF YOU HAD TO CREATE A CALCULATION ON SOMETHING LIKE MONTHLY PERCENTAGE INCREASE IN SALES REVENUE, THEY NEED TO LOOK UP THE SYNTAX AND IDENTIFY THE RIGHT DATA SETS AND TEST THEIR CALCULATION. ONCE THEY BUILD THE DATA SET AND THEN BUILD THE VISUAL THEY SPEND EVEN MORE TIME FINE TUNING THE FORMAT TO MATCH THE DESIRED STYLING WITH A NEW GENERATIVE BI CAPABILITIES AND QUICKSIGHT BUSINESS ANALYSTS CAN DO ALL OF IT IN NATURAL LANGUAGE AND FINE TUNE VISUALS IN JUST SECONDS AND ADD THEM TO THEIR DASHBOARDS IN THIS NEW AUTHORING EXPERIENCE ALSO ENABLES THESE ANALYSTS TO CREATE CALCULATIONS FOR JUST SIMPLE NATURAL LANGUAGE. JUST LIKE WHAT I SAID , MONTHLY PERCENTAGE INCREASE IN SALES AND THEY DO NOT HAVE TO LEARN ANY SPECIFIC SYNTAX LIKE CREATING A NEW DASHBOARD OR CALCULATION IS NOW AS SIMPLE AS ASKING A FEW QUESTIONS. AND ALSO TO HELP OUR IDEAS REALLY TAKE FLIGHT. WE NEED TO USE OUR DATA TO TELL STORIES. FOR INSTANCE, A MARKETING MANAGER MAY NEED TO
CREATE A DATA DRIVEN PRESENTATION TO REQUEST ADDITIONAL BUDGET FOR HOSTING AN EVENT IN NEW YORK TODAY. THEY WILL NEED TO SPEND A LOT OF TIME EXTRACTING DATA SUMMARIZING RESULTS AND CREATING PRESENTATIONS BECAUSE THEY ARE OFTEN EXPORTED INTO A PDF OR POWERPOINT, THEY CAN QUICKLY BECOME OUTDATED AND THEN RECREATING THEM AS A SLOW MANUAL PROCESS WITH THE NEW GENERATIVE BI CAPABILITIES IN QUICKSIGHT, IT CAN HELP WITH THIS TO NOW AGAIN BY BUSINESS USERS CAN AUTO AUTOMATICALLY GENERATE A STORY OR A VISUAL PRESENTATION OF THEIR QUICKSIGHT DATA USING NATURAL LANGUAGE PROMPTS SIMPLE ONLY TYPE THE DESCRIPTION OF A STORY IN ENGLISH TO CREATE A COMPELLING VISUAL WITH DATA FROM RELEVANT DASHBOARDS. AFTER THE STORY IS GENERATED, THEY CAN MODIFY IT AS NEEDED AND SECURELY SHARE IT WITH THEIR BUSINESS TEAMS. EXCITED TO BRING MORE GENERATIVE BI CAPABILITIES TO QUICKSIGHT IN THE FUTURE. FOR NOW, WE TALKED ABOUT BI. HOW ELSE CAN WE USE GENERATIVE AI
ACROSS OUR SERVICES TO IMPROVE PRODUCTIVITY? ONE EXAMPLE IS AMAZON, OUR AI CODING COMPANION FOR DEVELOPERS CODEWHISPERER GENERATES CODE RECOMMENDATIONS FROM NATURAL LANGUAGE BASED ON CONTEXTUAL INFORMATION SUCH AS DEVELOPERS RECENTLY OPENED PRIOR CODE AND COMMENTS, AND IT IS THE ONLY CODING COMPANION WITH BUILT IN SECURITY SCANNING FOR HARD TO DETECT VULNERABILITIES AND A BUILT IN REFERENCE TRACKER THAT DETECTS WHETHER A SOURCE CODE RECOMMENDATION MAY BE SIMILAR TO A PARTICULAR TRAINING DATA. THIS MAKES IT EASIER FOR DEVELOPERS TO DECIDE WHETHER TO USE THEIR PARTICULAR RECOMMENDATION IN THEIR PROJECTS DURING PREVIEW, WE RAN A STUDY A PRODUCTIVITY CHALLENGE, THAT SHOWED PARTICIPANTS WHO USE CODEWHISPERER WERE 27% MORE LIKELY TO COMPLETE THEIR TASK SUCCESSFULLY AND THEY DID IT 57% FASTER ON AVERAGE, CUSTOMERS AND PARTNERS LIKE INFOSYS, PUBLICIS SAPIENT AND HCLTECH ARE EMPOWERING THEIR DEVELOPERS TO DRIVE INNOVATION WITH CODEWHISPERER AND PYTHON DEVELOPERS WITHIN OUR OWN AMAZON ADS TEAM HAVE FOUND THAT CODE WHISPERER MAKES IT SEAMLESS FOR THEM TO PICK UP NEW LANGUAGES LIKE JAVA, AND THEY DON'T HAVE TO CONSTANTLY LOOK UP DOCUMENTATION AND SYNTAX NOW BECAUSE WE KNOW HOW MUCH TIME OUR BUILDERS HAVE SPENDING AND WRITING CODE. WE ARE ALSO LOOKING TO INTEGRATE CODEWHISPERER INTO MANY OF OUR SERVICES. ONE OF THOSE SERVICES IS AWS GLUE, WHICH MAKES IT EASY TO INTEGRATE DATA FROM MULTIPLE SOURCES FROM ANALYTICS MACHINE LEARNING AND APPLICATION DEVELOPMENT. GLUE PROVIDES MULTIPLE INTERFACES FOR CUSTOMERS TO BUILD DATA INTEGRATION JOBS AND THE COMMON INTERFACE IS GLUE STUDIO, WHICH ENABLES DATA ENGINEERS, ANALYSTS TO BUILD DATA INTEGRATION JOBS. HOWEVER, WHEN YOU THINK ABOUT AUTHORING THESE COMPLEX ETL PIPELINES, USERS STILL NEED TO UNDERSTAND GLUE AND SEVERAL SERVICES, WHICH MEANS THEY SPEND A LOT MORE TIME DOING ON THE APPROPRIATE SYNTAX AND BEST PRACTICES AND LESS TIME ACTUALLY SOLVING THEIR PROBLEM TO HELP USERS BUILD DATA INTEGRATION JOBS FASTER. I'M EXCITED TO ANNOUNCE A CODEWHISPERER
INTEGRATION WITH GLUE STUDIO. THIS IS AN AI POWERED ETL CODING ASSISTANT FOR DATA ENGINEERS ANALYSTS AND DEVELOPERS TO AUTHOR GLUE JOBS WITH THIS INTEGRATION, SIMPLY WRITE CODE OR COMMENTS AND NATURAL LANGUAGE IN GLUE STUDIO NOTES BOOKS TO RECEIVE CODE SUGGESTIONS AND SYNTAX CORRECTIONS FROM CODEWHISPERER IN REAL TIME. YOU CAN EASILY ACCEPT SUGGESTIONS OR DECLINE AND KEEP WRITING YOUR CUSTOM CODE. AND THIS INTEGRATION IS OPTIMIZED FOR DATA SOURCES LIKE S3, RDS AND REDSHIFT. IT IS THE BEST CODING COMPANION TO BUILD APPLICATIONS ON AWS. SO WHILE GENERATIVE AI IS ENHANCING PRODUCTIVITY FOR DEVELOPERS, DATA SCIENTISTS AND ANALYSTS, THESE TECHNOLOGIES ARE ALSO BUSINESS PROCESS IMPROVEMENTS FOR FRONT EMPLOYEES ACROSS A VARIETY OF INDUSTRIES, FROM HEALTH CARE TO AUTOMOTIVE.
LET'S DIVE DEEPER ON HEALTH CARE FOR A MOMENT. AWS OFFERS A PORTFOLIO OF SERVICES THAT EMPOWER OUR HEALTH CARE AND LIFE SCIENCES CUSTOMERS, INCLUDING PROVIDERS, PAYERS AND IT VENDORS TO MANAGE AND TRANSFORM HEALTH CARE DATA AT SCALE. THIS ALSO INCLUDES VENDOR THAT BUILT HIGHLY CRITICAL HIGH PERFORMANCE APPLICATIONS FOR CLINICAL SETTINGS AND TELEHEALTH APPS TO HELP STREAMLINE WORKFLOWS AND REDUCE ADMINISTRATIVE TASKS FOR PHYSICIAN PATIENTS AND PROVIDERS . WE HEAR THAT ONE OF THE MOST COMMON HEALTH CARE INDUSTRY PAIN POINT IS THE AMOUNT OF TIME IT TAKES FOR CLINICIANS TO WRITE DETAILED DOCUMENTATION FOR EACH PATIENT VISIT, WHICH OFTEN TAKES AWAY THE TIME FROM THE FACE TO FACE INTERACTIONS WITH THE PATIENT. TO SOLVE THIS PROBLEM, HEALTH CARE SOFTWARE VENDORS ARE EXPLORING HOW TO LEVERAGE GENERATIVE AI TO AUTOMATICALLY GENERATE THESE CLINICAL NOTES. BUT BUILDING THESE TYPE OF APPLICATION REQUIRES SIGNIFICANT AMOUNT OF ENGINEERING RESOURCE AND EXPERTISE TO TAKE THESE FOUNDATIONAL MODELS FOR MEDICAL USE CASES ON TOP OF ALL OF THESE CHALLENGES ARE HEALTH CARE HAS NEED TO MEET STRINGENT INDUSTRY SECURITY REQUIREMENTS. WE WANT TO EASIER TO ADDRESS THIS PAIN
POINT FOR CLINICIANS THAT IS WHY TODAY I'M VERY EXCITED TO ANNOUNCE THE PREVIEW OF HEALTHSCRIBE. THIS WILL SERVICE PROVIDES AUTOMATIC NOTE GENERATION FOR CLINICAL APPLICATIONS. HEALTHSCRIBE LEVERAGES AUTOMATIC SPEECH RECOGNITION AND GENERATIVE AI TO AUTOMATICALLY ANALYZE CONSULTATION AUDIO IDENTIFY SPEAKER ROLES FOR PATIENTS AND CLINICIANS, EXTRACT THE RELEVANT MEDICAL TERMS AND GENERATE A PRELIMINARY CLINICAL NOTES WITH HEALTHSCRIBE HEALTH CARE SOFTWARE VENDORS CAN CREATE APPLICATIONS THAT REDUCE THE BURDEN OF CLINICAL DOCUMENT STATION. IT SUPPORTS THE RESPONSIBLE USE OF AI BY INCLUDING REFERENCES TO THE ORIGINAL TRANSCRIPTS FOR EVERY SENTENCE IN THE AI GENERATED CLINICAL NOTES. THIS REALLY MAKES IT EASIER TO VALIDATE THE
ACCURACY OF IN THEIR APPLICATION BEFORE THEY ENTER IT INTO THE EHR, SECURITY AND PRIVACY FEATURES ARE ALSO BUILT IN TO ENSURE ANY AND OUTPUT TEXT ARE NOT STORED IN HEALTHSCRIBE. CUSTOMERS LIKE 3M HEALTH INFORMATION SYSTEMS ARE LEVERAGING HEALTHSCRIBE AS THE FOUNDATION TO HELP EXPEDITE, REFINE AND SCALE THE DELIVERY OF THEIR CLINICAL DOCUMENTATION AND VIRTUAL ASSISTANT SOLUTIONS. IT IS CLEAR THAT GENERATIVE AI HAS THE POWER TO TRANSFORM HEALTH CARE AND LIFE SCIENCES INDUSTRY IN MANY WAYS. LET'S SEE HOW ANOTHER COMPANY MERCK AI TO SOLVE A COMMON PROBLEM IN THE PHARMA INDUSTRY. OUR GLOBAL PHARMACEUTICAL COMPANY MERCK, HAS BROUGHT HOPE TO HUMANITY THROUGH DEVELOPMENT OF IMPORTANT MEDICINES AND VACCINES. AND
TODAY WE'RE AT THE FOREFRONT OF RESEARCH TO DELIVER INNOVATIVE HEALTH SOLUTIONS THAT ADVANCE PREVENTION AND TREATMENT OF DISEASES IN PEOPLE AND ANIMALS. A COMMON PROBLEM ACROSS PHARMACEUTIQUE INDUSTRY IS THE OCCURRENCE OF FALSE REJECTS DURING DRUG INSPECTION PROCESS TO INVESTIGATE THIS PROBLEM HOLISTICALLY, WE NEEDED TO INGEST PRODUCT GENEALOGY DATA PROCESS DATA QUALITY DATA FROM VARIOUS MANUFACTURING SYSTEMS AND REAL TIME DATA FROM INSPECTION MACHINES, AND THEN CONTEXTUALIZE AND HARMONIZE THIS DATA THAT'S WHY WE TURN TO AND TO SOLVE THIS CHALLENGE. WE USE AWS GLUE AND KINESIS TO INGEST, TRANSFORM AND CONTEXTUALIZE PROCESS AND REAL TIME DATA AND RUN ANALYTICS ON THAT DATA LOAD THAT DATA INTO REDSHIFT, WHICH IS THEN USED BY OUR ANALYTICS DASHBOARDS, OUR AI, ML PLATFORM IS BUILT ON AMAZON SAGEMAKER. WE LEVERAGE AWS DATASYNC TO INGEST DEFECT IMAGE DATA FROM INSPECTION MACHINES ACROSS SITES. OUR INFERENCE PIPELINES LOOK UP OUR MODEL REGISTRY FOR
ALL ML DEEP LEARNING MODELS FOR THESE IMAGES RUNS THOSE SCALE CLASSIFIES THE IMAGES AND SAVES THOSE INFERENCES INTO DYNAMODB, WHICH IS THEN SERVED UP BY OUR CONSUMPTION APPS. WE USE GENERATIVE AI APPROACHES AND GENERATIVE MODELS LIKE GANS AND VARIATIONAL AUTOENCODERS TO DEVELOP SYNTHETIC DEFECT IMAGE DATA FOR COMPLEX DEFECTS WHERE WE HAVE LIMITED TRAINING DATA. THE INSIGHTS GAINED HAVE HELPED US TO UNDERSTAND THE ROOT OF REJECT, OPTIMIZE PROCESSES AND REDUCE OVERALL FALSE REJECTS ACROSS VARIOUS PRODUCT LINES BY MORE THAN 50. IT IS GRATIFYING AND MOTIVATING TO KNOW THAT THE WORK WE ARE DOING HAS A DIRECT IMPACT ON PATIENT LIVES IN TERMS OF IMPROVING THE AVAILABILITY OF LIFE SAVING MEDICINES AND VACCINES. WHAT A GREAT STORY. WHETHER YOU'RE BUILDING APPLICATIONS WITH FMS IN BEDROCK OR SAGEMAKER OR USING OUR ML POWERED SERVICES, WE WILL CONTINUE TO INVEST IN MAKING PLACE TO HARNESS THE POWER OF GENERATIVE AI AND ML FOR ALL TYPES OF INDUSTRIES, BUT MAKING THE MOST OF AI AND ML REQUIRES MORE THAN JUST TOOLS. A
SUCCESSFUL GENAI STRATEGY INCLUDES A STRONG INFRASTRUCTURE LAYER FOUNDATION THAT CAN SUPPORT THE MASSIVE SCALE OF POWER, SECURITY AND RELIABILITY NEEDS OF ENTERPRISE APPS. WHAT OUR CUSTOMERS ARE TRYING TO DO WITH THESE FMS IF THEY ARE BUILDING THEM, CUSTOMIZING THEM, THEY NEED THE MOST PERFORMANT, COST EFFECTIVE INFRASTRUCTURE. THAT IS PURPOSE BUILT FOR MACHINE LEARNING. AWS HAS BEEN INVESTING WITH OUR PARTNERS AND
IN OUR SILICON FOR MORE THAN TEN YEARS TO OFFER A BROAD CHOICE OF HIGH PERFORMANCE, LOW COST ML INFRASTRUCTURE. THAT IS WHY AWS IS UNIQUELY POSITIONED TO HELP OUR CUSTOMERS ACCELERATE THEIR INNOVATION WITH GENAI. I, FOR EXAMPLE, WE HAVE A BROAD CHOICE OF ACCELERATORS FOR CUSTOMERS, INCLUDING OUR GPU BASED SOLUTION . AWS WAS THE FIRST TO BRING NVIDIA GPUS TO THE CLOUD MORE THAN 12 YEARS AGO AND WE ARE COLLABORATING WITH THEM TO DELIVER LARGE SCALE, HIGH PERFORMANCE GPU BASED SOLUTION FOR APPLICATIONS SUCH AS AI, ML GRAPHICS, GAMING AND HPC COMPANIES HAVE BEEN USING THESE GPU BASED INSTANCES TO SPEED UP ML AND THEY HAVE SCALED THEIR ML TRAINING WORKLOADS UP TO 10,000 GPUS. BUT OUR CUSTOMERS ARE CONTINUING TO PUSH THE BOUNDARIES OF THESE LARGE SCALE , LARGE LANGUAGE MODELS. THEY ARE TRAINING AND DEPLOYING MORE SOPHISTICATED MODELS. AS A RESULT, THEY ARE VERY EAGER FOR THE MOST UP TO DATE SOLUTIONS THAT HELP THEM GAIN A STRATEGIC EDGE TO MAKE THEIR MODEL TRAINING FASTER AND AT SCALE. THAT'S WHY TODAY I'M EXCITED TO
ANNOUNCE THE GENERAL AVAILABILITY OF AMAZON EC2 P5 INSTANCES. THESE INSTANCES ARE OPTIMIZED FOR GENERATIVE AI AND POWERED BY NVIDIA H100 TENSOR CORE GPUS. THEY ARE IDEAL FOR TRAINING AND RUNNING INFERENCE FOR THESE INCREASINGLY COMPLEX LLMS THAT HAVE MORE THAN ONE HUNDREDS OF BILLIONS OF PARAMETERS. P5 INSTANCES WILL PROVIDE THE HIGHEST PERFORMANCE IN OUR PORTFOLIO, ACCELERATING PERFORMANCE BY UP TO SIX X AND REDUCING TRAINING COSTS BY UP TO 40% AS COMPARED TO EC2 P FOR INSTANCES. P5 INSTANCES ARE DEPLOYED IN SECOND GENERATION
EC2. ULTRACLUSTERS WHICH SCALE UP TO 20,000 GPUS INTER CONNECTED WITH PETABYTE SCALE NON-BLOCKING NETWORK. THIS ENABLES US TO DELIVER 20 EXAFLOPS OF AGGREGATE COMPUTE CAPABILITY. NOW, BEYOND OUR COLLABORATION WITH NVIDIA, WE HAVE MORE THAN A DECADE OF EXPERIENCE DESIGNING AND BUILDING SILICON AND WE HAVE APPLIED OUR OWN LEARNINGS TO CREATE INNOVATIVE PURPOSE BUILT SILICON FOR YOUR ML AND AI WORKLOADS. WE RECENTLY ANNOUNCED GENERAL AVAILABILITY OF EC2, EC2
INSTANCES POWERED BY OUR INFERENTIA CHIP, A CUSTOM BUILT CHIP FOR ML INFERENCE. THESE INSTANCES DELIVER UP TO 40% BETTER INFERENCE PRICE PERFORMANCE THAN COMPARABLE EC2 INSTANCES AND OUR TR AND ONE INSTANCE DELIVERS UP TO 50% SAVINGS ON TRAINING COSTS WHERE WE OFFER HIGH PERFORMANCE FOR DEEP LEARNING, TRAINING AND INFERENCE WITH SINGLE NIFICANT COST SAVINGS. OUR CUSTOMERS ARE SEEING IMPRESSIVE RESULTS AS THEY BUILD DEEP LEARNING MODELS AND FOUNDATIONAL MODELS USING THESE INSTANCES FOR INSTANCE, FINCHAI DEPLOYED THEIR TRANSLATE MODELS ON AWS INFERENTIA. AND SAVED 80% ON COST WHILE MAINTAINING THE SAME THROUGHPUT . OUR ABILITY TO DELIVER HIGH
PERFORMANCE AT THE LOWEST COST IN AMAZON EC2 IS WHY CUSTOMERS LIKE AIRBNB, SPRINKLR AND AUTODESK USE OUR PURPOSE BUILT ACCELERATOR FOR THEIR MOST DEMANDING ML WORKLOADS WITH THESE PURPOSE BUILT ACCELERATORS AS WELL AS THE LATEST EC2 P5 INSTANCES, OUR CUSTOMERS HAVE A HIGHLY PERFORMED AND DIFFERENTIATED SET OF TOOLS TO ADDRESS THE LARGEST ML CHALLENGES. THESE INFRASTRUCTURE INVESTMENTS WILL SUPPORT YOUR GENAI STRATEGY AS YOU EXPERIMENT AND QUICKLY SCALE YOUR APPLICATIONS FOR A WIDE VARIETY OF YOUR USE CASES. NOW OUR CUSTOMERS HAVE ACCELERATED THEIR USE OF GENERATIVE AI IN JUST THE PAST FEW MONTHS, AND WE KNOW MANY OF YOU ARE EAGER TO GET STARTED AS WELL. AND BECAUSE WE KNOW THIS TECHNOLOGY IS NEW TO
SO MANY, WE ARE ALSO MAKING IT ACCESSIBLE THROUGH A VARIETY OF TRAINING OPPORTUNITIES. AWS IN ADDITION TO OUR BROADER LIBRARY OF DIGITAL COURSES ON AWS SKILL BUILDER AS WELL AS PROGRAMS LIKE AWS ACADEMY, WE RESTART AND EDUCATE, WE OFFER NOW A NEW COLLECTION OF FREE AND LOW COST TRAININGS TO HELP PEOPLE UNDERSTAND, IMPLEMENT AND BEGIN USING GENERATIVE AI FOR EXAMPLE, WE RECENTLY LAUNCHED A TRAINING ON COURSERA CALLED GENERATIVE AI WITH LARGE LANGUAGE MODELS, A COURSE BUILT BY EXPERTS WITH DR. ANDREW NG FROM DEEPLEARNING.AI. THIS IS A GREAT OPPORTUNITY TO LEARN ABOUT LMS AND GET HANDS ON EXPERIENCE FROM SELECTING, TRAINING, FINE TUNING AND DEPLOYING THESE FOR YOUR APPLICATIONS. WE ALSO OFFER FREE COURSES USING CODEWHISPERER FOR EXECUTIVES WHO WANT TO ADDRESS THEIR BUSINESS CHALLENGES AND FOR PARTNERS WHO WANT TO BUILD GENAI APPS. NOW TO SUM IT ALL UP, WE HAVE COVERED A LOT OF GROUND TODAY AND SHARED SEVERAL NEW TOOLS AND INNOVATIONS TO HELP YOU START WORKING WITH GENERATIVE AI ON AWS, WITH A WIDE SELECTION OF FOUNDATIONAL MODELS AND TOOLS AND AMAZON BEDROCK NEW SET OF GENERATIVE AI POWERED SERVICES TO HELP DRIVE PRODUCTIVITY ACROSS YOUR ORGANIZATION. PASSION AND PURPOSE BUILT MACHINE LEARNING
INFRASTRUCTURE AND GPUS FOR BETTER PERFORMANCE AND LOWER COSTS AND ACCESS TO EDUCATION AND TRAINING OPPORTUNITIES. ENTITIES WITH ALL OF THESE AWS HAS EVER EVERYTHING YOU NEED TO ACCELERATE YOUR GENERATIVE AI JOURNEY. THERE IS SO MUCH DESERT TO LEARN MORE AND EXPERIMENT WITH THIS TECHNOLOGY. AND THIS IS JUST THE BEGINNING. WE HAVE A LOT MORE COMING THIS YEAR AND I'M EXCITED TO SEE WHAT YOU CAN DO TO REIMAGINE AND TRANSFORM MIME YOUR APPLICATIONS WITH GENERATIVE AI ON AWS.
THANK YOU FOR YOUR TIME TODAY.
2023-08-01 08:23