Today Bitcoin rebounds from a week of Selling pressure and coinbase's head of Machine learning breaks down how the Crypto exchange uses AI to prepare for Highly volatile Markets welcome to cnbc's crypto world I'm Lindsay Jacobson digital currencies Are bouncing back from losses earlier in The week by noon Eastern Bitcoin Reclaimed the $60,000 level after Falling as low as 57 ,000 yesterday Ether is also high this morning climbing Back to 2576 Solana meanwhile rising to $146 there were some signs that the Crypto Market was bracing for nvidia's Earnings and some investors were Concerned the chipmaker wouldn't be able To live up to the hype and potentially Push investors away from risk assets Well Nvidia did drop better than Expected earnings reporting 122% Increase in revenue from a year earlier But that's down from three straight Periods of year-over-year growth Exceeding 200% so this latest quarter Might not have met the sky high Expectations for some holding the stock That being said the chip makers quarter Seems to have been enough to soothe Concerns that the market was headed Sharply lower which is good news for Crypto investors Too all right let's jump to our main
Story earlier this week coinbase Announced a newly deployed machine Learning model that helps the crypto Exchange predict spikes in user traffic And prevent serious downtime during High Volatility crypto world's McKenzie Sagales spoke to the company's head of Machine learning Rari Gupta about this Launch so this week your team announced That coinbase has developed and deployed A machine learning model that's all About preventing downtime tell me more About the launch sure so you know uh It's been pretty normal in the industry To look at load that's on the system and Then scale up your CPUs up and down as Things go and this can be done because CPUs are very um performant and they can Quickly do this now we have the problem That uh there are we need to do the same Thing for our databases and there we Really didn't weren't able to find any Solution out there that did the same Thing for databases so we ended up Building our own machine learning model That kind of predicts ahead of time and And by ahead of time I don't mean 15 Minutes I mean like hours ahead of time That we think that the load in our System is going to go up because the Volatility in crypto is increasing and By doing that and so of when we think The volatility is increasing we are Going to be able to prescale the
Databases to some level as needed and Such that we can manage it and we've had It in production for a couple of months And you know in the last um and when These uh let's say these events of high Load happen we've been able to manage it And uh and and pre-scale so we we wrote About it in our blog you know we saw a Lot of volatility overnight a lot of of Uh price movements and and since you've Been testing out this product in the Last few months have you noticed a Significant difference in the way that The platform operates yeah so I think Look what if if you think about it like What is the main purpose right so the Main purpose is to make sure that our Customers get the correct level of of Support that they need but at the same Time we manage our costs because we Could obviously hold the databases high At a very very high level all the time That would ensure performance but it Would raise our cost cost so this is the Way of balancing those two things and by Getting using machine learning to Pre-scale the databases we are then able To manage our cost while ensuring that The customers get the right level of Service now more broadly speaking I know That the exchange has been leveraging AI To beef up security as well and I'm Wondering just how significant a role Machine learning has played in both
Enhancing company security and Efficiency for Customers um absolutely so we've been Using machine learning to protect our Customers for many years now so in fact Every single transaction that goes Through our system is protected by Machine learning models that tries to Make sure that the transaction is indeed The correct um is indeed the right Transaction versus it's talking to some Risky Business whether your account has Been taken over I mean to some level you Can think of it starts even as you log In because as you are logging into the App we have a model that tries to detect Whether it's it's you or your account Has been taken over by someone so I mean These models run billions of times now Your coinbase is head of machine Learning and it's a side of the business That we typically don't hear a lot about But you've been leaning into gen AI Applications machine learning more Broadly for a while now and on the Gen AI side of things you've got your Customer facing conversational llm Powered chat bot which just launched in June and I'm wondering if you can tell Me more about how you're putting this Tech to use and what are some of the Features that are maybe invisible to the Customer but hugely transformative to The end user and the way that your
Business operates yeah so I think look I Mean almost everyone in the tech Industry has recognized the fact that Geni is a really really powerful tool But now the the I would say the right Analogy is that it's not enough having a Powerful tool you actually need to have The people who know how to use the tool So that's where we have been leaning in And we have I'm very proud of the fact That I have a a very strong machine Learning team that knows how to wield Those tools and one of the things we did As you mentioned is that in June we Launched our customer facing chatbot so Every single one of our customer facing Uh conversations in US is now going Through this llm power chatbot so in Order to do that you have to take care Of several different things right so you Have to make sure that that there are Appropriate guard rails in place to make Sure that the llm is protected and and Doesn't end up saying things that it Shouldn't it's uh we also need to make Sure that it's performant so um you it Probably heard of the global GPU Shortage we have to really work with our Partners to make sure that we have the Right bandwidth available to make sure That our customers get the um get the Like the low latency responses that you Want and and naturally we have to keep Tuning and find and fine-tuning the um
The the responses that we are getting And and testing it continuously to make Sure that we improve so it's been a it's Been a very exciting journey and uh and You know fingers crossed it's been we've Been enjoying the experience and then The flip side of that how is blockchain Technology enhancing machine learning Applications and data Integrity yeah so I think that's true you know most of the AI is a really powerful tool and many Applications that you see are for AI Helping blockchain but there's also the Flip side of it which is blockchain Helping Ai and there's a couple of Really interesting use cases so one for Example is that blockchain is a way to Make sure that provence of anything is Proven and it's verifiable so protecting AI deep fakes from uh and making sure That what you are seeing is real and not Fake is something that with blockchain You can easily manage um another way is That distributed AI is something that People have been working on for many Years but in order to incentivize this You need to have a systematic way of Sharing payments across a number of Large number of distributed nodes and Blockchain is the perfect underlying Platform to achieve that so these are Just about two of these cases that uh um That that help blockchain for AI I mean In fact you know back in may we uh we
Organized a research conference called Blockchain and Ai and where this was a Big topic of discussion that we had About how AI can help blockchain and Blockchain can help AI last question for You talk me through what's next what's Your vision for coinbase's Gen and Machine learning strategy in the next Few months and in the next year yeah so I think if you look at it I mean in the Next few months it's of course we are Doing many of these things and you're Already hearing about them uh in future Though we basically feel that AI should Be a a helper to every employee and Every customer of coinbase so we we Already have an employee assistant that Is helping many of our employees in Their day-to-day jobs not just the Engineers not the ml people but the Finance people the HR people the legal People and so on um but we also want to Bring all that AI goodness to our Customers and and bring some sort of a Helper um um so that the customers are Able to utilize AI in their day-to-day Transactions and their user Journeys okay that's all for crypto World we'll be back again tomorrow and We'll see you then [Music]
