SigTech CEO Bin Ren on AI and Finance

Bloomberg

04-03-2025

00:00
IT CREATES AI COPILOTS FOR INVESTORS, AND I FIND THIS SO FASCINATING SO I WILL TELL YOU TO GET INTO IT. BIN JOINS US ON SET AT THE BLOOMBERG INVEST CONFERENCE. EXPLAIN TO US IF YOU COULD THE ELEVATOR PITCH OF WHAT YOU ARE DOING. I IMAGINE THESE LITTLE AI AGENTS THAT ARE BETTER ANALYSTS AND BETTER INVESTORS THAN WE ARE, IS THAT WHAT YOU ARE LOOKING TO DO? BIN: AT SIGTECH, WE ARE BUILDING A TEAM OF SPECIALIST AI AGENTS THAT COLLABORATE TO AUTOMATE PREVIOUSLY AUTOMATED VERY HIGH PROCESSORS -- VERY HIGH RATED PROCESSORS. MATT: DO YOU MEAN YOU HAVE MULTIPLE LARGE LINK WHICH MODELS THAT WORK TOGETHER? DON'T YOU JUST HAVE ONE AI? HOW IS IT A TEAM COLLABORATED? BIN: SO FIRST WE HAVE TO TALK ABOUT WHAT IS AN AGENT AND WHY THAT IS DIFFERENT FROM A LARGE LANGUAGE MODEL. AN AGENT IS POWERED, BUT THE AGENT HAS TO BE ABLE TO ACCESS PRIVATE DATA, USE TOOLS, AND THEN PURSUE A PARTICULAR OBJECTIVE ALMOST ENTIRELY AUTONOMOUSLY, SO IT IS A VERY BIG STEP UP FROM A CHATBOT. WHEN WE SAY WE BUILD A TEAM OF AI AGENTS, WE LITERALLY BUILD AN ARMY OF THEM, EACH ONE SPECIALIZING IN SOMETHING VERY SPECIFIC. FOR EXAMPLE, WE HAVE DEDICATED AI AGENTS TRAINED TO ANALYZE CENTRAL BANKS, EVERYTHING COMING OUT OF THE FED, BANK OF ENGLAND, BANK OF JAPAN. THE ANALYST CAN READ THE RAW MATERIALS, SOMETIMES IN A FOREIGN LANGUAGE, AND READ THE NUANCES AND GIVE US ANSWERS IN ENGLISH. AND THEN A USER WILL WORK WITH THE TEAM OF AI AGENTS, IS FUNCTIONALLY -- AGENTS, ESSENTIALLY WORKING AS A SUPERVISOR. THAT IS HOW WE APPROACH OUR DAILY WORKFLOWS. KATIE: THAT IS INTERESTING. IS THE GOAL TO REPLACE HUMAN ANALYSTS? IS THAT THE END STATE YOU ARE TRYING TO GET TO? BIN: NO, THE END GOAL IS NOT TO REPLACE HUMAN ANALYSTS BECAUSE WE ALL WANT TO DO INTELLECTUAL AND CREATIVE WORK. THE AGENTS CAN DO TODAY AUTOMATED BUT NONCREATIVE WORK. THERE ARE PLENTY OF THEM. FOR EXAMPLE, GENERATING GRAPHS, REPORTS ON A PERIODIC BASIS BUT THEY CANNOT BE CODIFIED AND AUTOMATED BEFORE. NOW TODAY THEY CAN. WE USE OVER AI AGENTS FOR COMMERCIAL. IT MAKES A BIG DIFFERENCE. KATIE: I WANT TO GO BACK TO YOUR ORIGIN STORY AND THAT OF A. YOU SPUN OUT IN 2019 AND THE CONVERSATION AROUND AI IN 2019 -- I WILL BE HONEST, I DON'T KNOW WHAT IT SOUNDED LIKE FOR A LOT OF US, WE STARTED TALKING ABOUT AI EVERY SINGLE DAY WITH THE LODGE OF CHATGPT IN 2022. REFLECT A LITTLE BIT. HOW HAS YOUR BUSINESS AND THOSE CONVERSATIONS YOU ARE HAVING CHANGED SINCE 2019, WHICH FEELS LIKE A LIFETIME AGO? BIN: THAT'S RIGHT GET IN 2019 -- THAT'S RIGHT. IN 2019, WE STARTED A COMPANY WITHOUT A PARTICULAR BASIS AROUND AI. WE BUILT TOOLS TO POWER HUMAN ANALYSTS HAVE TO USE THE LANGUAGE. THE FIRST THREE YEARS OF OUR COMPANY'S HISTORY, WE WERE BUILDING TOOLS FOR HUMANS BEGIN THREE YEARS AGO, WE MADE A VERY BIG BET ON GENERATIVE AI BECAUSE I REALIZED WE CAN SPEND THE SAME AMOUNT OF EFFORTS ON TRAINING LARGELY WHICH MODELS TO DO SOME OF THE JOBS. WE STARTED TO PUT A LOT OF EFFORT INTO THAT AREA, AND TODAY WE ARE ENTIRELY AI FOCUS. EVERYTHING RUNS ON THE SAME INFRASTRUCTURE WE BUILT OVER THE YEARS EXCEPT NOW ESSENTIALLY WE ARE BUILDING TOOLS AND INFRASTRUCTURE FOR AI AGENTS WHO COULD BE HIRED AND USED BY HUMANS INSTEAD OF BUILDING TOOLS AND INFRASTRUCTURE FOR HUMANS DIRECTLY. MATT: I HAVE SO MANY QUESTIONS. KATIE: I CAN FEEL YOU VIBRATING. MATT: IT IS SO FASCINATING TO ME. JUST TO GET THE BASICS DOWN, DOES THAT MEAN YOU TRAINED YOUR OWN LARGELY WHICH MODEL AT SIGTECH, OR DO YOU USE THE MODELS FROM OTHER HYPER SCALARS THAT HAVE ALREADY SPENT THE BILLIONS OF DOLLARS NECESSARY TO DO THAT? BIN: WE USE DIFFERENT MODELS FROM DIFFERENT VENDORS FROM OPENAI, FROM GEMINI. UP UNTIL FOUR MONTHS AGO, OPENAI MODELS HAD SUCH A BIG LEAD ON THEIR COMPETITORS SO THEY ARE DEFINITELY THE MOST POPULAR LARGELY WHICH MODELS THAT WE USE. VERY RECENTLY, WE COULD SEE THE QUALITY IT MODELS GO TO THE POINT WHERE WE ARE BUILDING AGENTS POWERED BY THOSE LARGELY WHICH MODELS FOR VERY SPECIFIC JOBS. THEY ACTUALLY DO A BETTER JOB THAN OPENAI MODELS. WE HAVE ACCESS TO DEEPSEEK HOSTED IN THE U.S. TO OUR BUSINESS IS ABOUT PUTTING THE BEST AGENTS, THE BEST TEAM POWERED BY THE RIGHT LARGE LANGUAGE MODEL FOR THAT JOB SO IT IS ABOUT SPECIALTY. MATT: JUST TO NOTE, YOU PROBABLY KNEW THAT SAM BANKMAN-FRIED'S ANTHROPIC INVESTMENT WOULD BE A RUNAWAY HIT BEFORE WE DID, RIGHT? ALSO, DEEPSEEK. WHAT IS YOUR TAKE ON THAT? IS IT SOMETHING THAT CAN CREATE ON THE CHEAP OR DO YOU THINK THEY SPEND A LOT OF MONEY TO DO IT? BIN: DEEPSEEK LAST WEEK, I SPENT THE WHOLE WEEK ON OPEN-SOURCE, A LOT OF THEIR INFRASTRUCTURE, SO I THINK ONE OF THE MAJOR BREAKTHROUGHS BY THE FIRM IS ABOUT THE INCREASED EFFICIENCY IN TRAINING. THE WAY THE LARGELY WHICH MODELS ARE TRAINED IF I CAN GET A BIT OF TECHNICAL DETAILS, IT IS IN TRADING PARALLEL. SO YOU COPY THE NEURON NETWORKS MANY COPIES, SCATTER IT ACROSS GP USE, AND SPLIT YOUR TRAINING SETS INTO DIFFERENT PILES ESSENTIALLY SO YOU CAN TRAIN MULTIPLE COPIES OF THE SAME NEURON NETWORK USING DIFFERENT PILES OF DATA. BUT THEN YOU AGGREGATE THE LEARNING. WE ACTUALLY AGGREGATE ALL THE LEARNING TOGETHER, UPDATE ALL THE MODELS, AND DO IT AGAIN. SO THERE IS UP-AND-DOWN USAGE OF GP WAS -- GPUS. DEEPSEEK ESSENTIALLY FIGURED OUT A WAY TO ALWAYS KEEP THE GPU UTILIZATION AS HIGH AS POSSIBLE SO IT IS CHEAPER AND FASTER. SO EFFICIENCY GAINED BY DEEPSEEK IS OPEN SOURCED AND EVERYBODY WILL USE IT. THAT WILL HELP PUSH THE EFFICIENCY FRONTIER OF LARGE LANGUAGE MODELS FORWARD. KATIE: INTERESTING. BIN: SO EVERYBODY BENEFITS. KATIE: THAT IS A REALLY INTERESTING TAKE THAT THIS IS NOT SOMETHING TO BE IGNORED, WHAT DEEPSEEK DID. IT FEELS LIKE SOME INVESTORS WERE EAGER TO FILTER IT OUT. I WANT TO TAKE A STEP BACK AND TALK ABOUT HOW AI CAN POTENTIALLY FIT INTO THE FINANCIAL WORLD, TO THE INVESTMENT WORLD AS WELL BECAUSE THE QUESTION I ALWAYS HAVE IS WHETHER AI WILL BE MORE HELPFUL WHEN IT COMES TO PRODUCTIVITY OR WHETHER WE WILL BE ACTUALLY GENERATING INVESTMENT ALPHA, USING AI WHEN IT COMES TO INVESTMENTS. IT SEEMS LIKE YOUR BUSINESS IS MORE ABOUT INCREASING PRODUCTIVITY RATHER THAN GENERATING OFF UNNECESSARILY -- GENERATING ALPHA NECESSARILY. BIN: BECAUSE OF THE REGULATION, THE FRAMEWORK, IT IS HUMAN CENTRIC. THE REGULATORS WANT TO POINT A FINGER TO A PERSON WHO IS ACCOUNTABLE FOR ALL THE ACTIONS AND DECISIONS, SO I THINK FOR THE FORESEEABLE FUTURE IN FINANCIAL SERVICES, IT IS ALL ABOUT HOW TO TURN ONE PERSON INTO A SUPER PERFORMER BACKED BY A TEAM OF AI AGENTS. I THINK THE NATURE OF OUR JOB IS CHANGING FROM BEING AN INDIVIDUAL CONTRIBUTOR TO EVERYBODY WILL BE A TEAM LEADER. EVERYBODY WILL BE A TEAM LEADER, SUPERVISOR, ORCHESTRATOR, AND THAT IS HOW WE DRAMATICALLY IMPROVE THE PRODUCTIVITIES. KATIE: AND I HEAR WHAT YOU ARE SAYING THAT YOU SEE EVERYONE BECOMING A TEAM LEADER IN THAT MODEL. MATT: EVERYONE GETS A MEDAL. KATIE: EVERYONE IS A BOSS. YOU ARE NOT TRYING TO REPLACE HUMAN ANALYSTS, BUT IS IT REASONABLE TO THINK THAT HUMANS WILL LOSE THEIR JOBS AS AI BECOMES MORE INTEGRATED IN THE FINANCIAL WORLD? BIN: I THINK PEOPLE WILL FIND NEW THINGS TO FOCUS ON. I THINK PEOPLE WILL FIND MORE TIME AND ENERGY TO FOCUS ON MORE CREATIVE -- FOCUS ON INTERPERSONAL RELATIONSHIPS, MEETING CUSTOMERS, UNDERSTANDING THEIR NEEDS AND WANTS. THE TRANSLATION FROM THE PREFERENCE OF NEEDS AND WANTS INTO INVESTMENT PORTFOLIOS IS HIGHLY AUTOMATED, THAT HUMAN, WHAT THEY TRULY PREFER, STILL REQUIRES A HUMAN TOUCH. I THINK THE EMPHASIS ON OUR JOBLESS SHIFT. MATT: I HOPE YOU ARE RIGHT, AND I HOPE THE BOBS UNDERSTAND I AM A PEOPLE PERSON. KATIE: YOU CERTAINLY ARE. MATT: I WAS LOOKING AT YOUR SLIDES. YOU DID A PRESENTATION BEFORE. IT WAS SIMILAR TO SOMETHING MARC WAS TELLING US ABOUT PUBLIC MARKETS THAT FOR A LONG TIME THEY WERE FAIRLY STABLE AND NOW HAVE BECOME VERY VOLATILE. USUALLY CAPITAL MARKETS ARE BECOMING LESS EFFICIENT AND THAT IS KIND OF THE SAME THING. WHY DO YOU SEE CAPITAL MARKETS LOSING THEIR EFFICIENCY? BIN: I WOULD ARGUE THAT THE PUBLIC MARKET IF YOU THINK ABOUT IT, THERE ARE A FEW THINGS GOING ON THE LAST 15 YEARS. NUMBER ONE IS THE TENS OF TRILLIONS OF CAPITAL GOING INTO INVESTMENT FUNDS. BY THE PETITION, IT DOES NOT CONTRIBUTE TO PRICE DISCOVERY. ON THE OTHER HAND, IF YOU LOOK AT THE LARGEST STOCKS, IN THE LAST 10 YEARS IT JUST SO HAPPENED THAT THE BIGGEST STOCKS ARE QUITE OFTEN THE BEST RETURNS ANY GIVEN YEAR, AND TODAY THE TOP 26 STOCKS IN THE S & P 500 ARE TOGETHER ACCOUNTABLE FOR HALF OF THE MARKET CAP. I HAVE NEVER BEEN A PERSON WHO CAN NAME THE OTHER 474. KATIE: THERE IS A LOT OF THEM TO BE FAIR. [LAUGHTER] BIN: YOU SEE THE ISSUE IS THE ACTIVE EQUITY MANAGEMENT HAS REALLY HAD A VERY BAD RUN IN THE LAST 15 YEARS AND WE CAN SEE THAT IN THE NUMBERS. THE NUMBER OF EQUITY ANALYSTS HAS GONE DOWN BY OVER A THIRD THE LAST 10 YEARS. BECAUSE THE TRUTH IS THE EFFORT REQUIRED TO ANALYZE A $1 TRILLION COMPANY IS NOT 1000 TIMES MORE THAN THE EFFORTS REQUIRED TO ANALYZE A $1 BILLION COMPANY. IT IS ROUGHLY THE SAME MAGNITUDE, SO WHY WOULD PEOPLE SPEND THEIR ENERGY ON THE OTHER 474 WHEN THEY CAN DEPLOY THE SAME AMOUNT OF CAPITAL IF THEY JUST ANALYZE THE TOP 26? THAT IS THE ISSUE IN THE PUBLIC MARKET TODAY. THERE IS A GAP BETWEEN THE NUMBER OF COMPANIES LOOKING FOR COVERAGE AND AN ANALYSIS VERSUS THE AMOUNT OF POWER WE CAN PROVIDE TO COVER THEM. THE ONLY WAY TO BRIDGE THE GAP