OpenAI Structured Outputs, character.ai “acquisition,” and is it an AI bubble?

Wall Street Spooks pretty easily and Hypes pretty easily and they're also on a cycle that research certainly is not structured outputs probably the most sexy release of this summer you're kind of breaking this fcking Bronco that uh just came out of the blue does the acquisition of character AI make any sense at all you have to know what's your value ad and how much of that is a differentiator with high Mo so others can't just come in and do what you do all that and more on today's episode of mixture of experts I'm Tim Hong and I'm joined today as I am every Friday by a genius panel of technologists Engineers and more to help make sense of another hectic week in AI land on the panel today we've got three guests Marina danki as a senior research scientist Kush vne IBM fellow working on issues surrounding AI governance and chit vary senior partner Consulting on AI for US Canada and Latin America [Music] all right so uh let's just get into it first story of the week uh is a big one um but I want to start with kind of a round the horn question let's just start with a quick yes or no and it's a very simple question to kind of kick off to discussion which is are AI companies going to bring down the American economy uh Kush yes or no what do you think uh no uh no and Marina no okay we have uniform skepticism at that position and I think that's actually what I wanted to get into so uh if you've been keeping your eyes on the financial news this week markets were massively down um across the board internationally and uh there was a lot of speculation as to why this was the case um people were proposing you know the unwinding of exotic Financial positions uh concerns about the FED not cutting rates but one thing that a number of people kind of argued was should we blame AI like the hyper AI for this um and part of this claim was based around the idea that the companies really leading the downturn um and arguably a big drag on you know indexes like the S&P 500 were tech companies that have made big bets on AI in the last 12 to 24 months and so want to get the kind of panel's opinion and Kush maybe we'll toss it over to you first is you know do we buy this as a theory like why should we or shouldn't we believe that AI is kind of a contributor to this downturn um and and is kind of a a popping or at least kind of increasing skepticism around AI um having these big macro effects I'm curious why you said no in the the first question there yeah I mean uh there's clearly I mean hype Cycles with everything uh but I think the economy has a lot more to offer I mean it's a it's a very broad-based sort of thing AI is kind of the the cherry on top or the the icing on the cake so um uh I mean yes it affects uh perception uh and less uh I mean of the view that it uh is really about the fundamentals at this point I think that'll change over time but but not right now well I think there's been I mean part of this I think is also following on the tales of I think we've been talking about it for the last few episodes is kind of these reports coming out of Banks and other uh you know Financial firms kind of raising some skepticism around kind of like the excitement around AI so you know there's the Goldman Sachs one that we talked about a few weeks back and also the sequ report that some people might have seen um it is true though that the tech companies have made genuinely a really big bet on the market for AI um and I guess I'm kind of curious you know maybe show bit I'll throw it to you is you know are you seeing you know clients kind of following those Jitters are they reading these reports and saying well you know maybe AI is not providing you know what we thought it would you know should we be a little bit more cautious about how we make these Investments so I don't think the clients have to and I'm talking about the 1400 companies 500 they don't have to read these reports to realize that certain U areas AI has been over prised in certain areas they have they are under utilized right so there's absolutely no confusion about the fact that AI is going to have is having a seismic impact on the businesses going forward so no CEO can can say that the next 5 years are not going to be uh massively impacted by what AI can do it's a question of how do you applying AI surgically in the processes and how do you think about a strategy of data that then leads to an AI strategy that then delivers value for you right so the conversation has changed more and all right after experimenting for 2 years we have a good sense of where Ai and geni are working well we now need to make sure that we have a good mechanism to figure out the high value unlocks in the business appreciate that it's a it's a combination of AI Automation and generative AI it's not all gen handling the entire process into end and we need to make sure that our data real estate and the people and and the processes are aligned to unlock that value right so I think there's a significant appreciation of the value it can bring but also the fact that it's a journey and you need to do you need to do make steps along the way to make sure that you're getting that value unlocked that's very clear to all my for 100 500 clients yeah I think that's maybe one thing that our listeners would benefit a lot from your expertise on show bit is I think you kind of put out the idea that there's like underhyped areas of AI right um and I'm kind of curious if there's when you say that if you've got kind of particular areas in mind where you're like this is where businesses aren't looking right like you know there's a lot of hype in the space but like this this seems to be where some of the Hidden are I'm curious if you can speak to that a little bit yes I think it's a it's a mundane tasks it's the stuff that uh how do you how do you make sure that every employee across organization can experiment in their day-to-day workflows with AI with generative AI in a very secure and governed way so within IBM Consulting for example we have 160,000 Consultants who wake up in the morning and doing all kinds of varied tasks there are a small subset of people who are AI gurus right they are they feel that if it's been 20 minutes since llama 3.1 landed and we have not had it running locally you are an embarrassment to society right there's a small portion of those but 85% of the other uh part of Consulting they're doing things like I'm going to do code creation I'm a tester for the last 11 years I've been doing marketing campaigns I'm going to do Finance workflow so I'm going to get a invoice I'm going to marry it against the contract the the purchase order and I'm going to approve it or disapprove it right so those kind of mundane workflows have a human in the loop and you need to figure out how Excel got embedded in those workflows you're now at the point where you're having AI generative AI get embedded everybody figured out how to use Excel to improve their day-to-day workflows right we're at that same point today so you need to get to a point where every IBM consultant we call that Consulting assistant as an example it could be a co-pilot from Asher could be Amazon Q's Google of the world but you need to democratize people actually messing with their day-to-day figure out that oh this email that I write ,00 times a month can be automated and that's the value unlock get get CL get your end customers to start your end employees to start experimenting in a governed way so that kush doesn't have a heart attack just make sure doing this in a way that we don't get ourselves into trouble um yeah I think that's actually in some ways like it has ended up being what you're describing show bit the kind of like 800 pound gorilla of the AI world you know I love this kind of joke that like you start open AI because you really want to create like AGI but like just slowly but surely the gravity ational well of being like B2B SAS uh and offering that as a service is like really where the gigantic amount of money uh is um Marine I did have a question for you based on kind of what sha just talked about here I know you know sha you kind of made the difference between like people saying okay if you can't Implement Lama 3 on day one and revolutionize all your business you know processes in the first day you're a waste of society um I'm kind of curious so there's one standout company here which is NVIDIA uh which is Hardware um and that is a company that has been hit in the stock market rather hard um and I think was one of the examples that people said see this is why AI is hyped um do you do you buy that I mean is NVIDIA indeed the most valuable company uh in the whole world uh and you know how should we think about sort of Hardware in this picture like will Hardware continue to be sort of the most valuable kind of piece of this AI Pi at least as far as like the stock market is concerned I mean it's a dependency so just talking of pure engineering terms you are pretty uh much tied to it because it's very much a dependency I will say as far far as Nvidia going up in value crashing in value um Wall Street Spooks pretty easily and Hypes pretty easily and they're also on a cycle that research certainly is not they want to know all right q1 what do you got Q2 what do you got Q3 what do you got it's not the rate at which research actually happens so when you have preliminary results Wall Street will get over excited and then the results next time are not as good and then they get over depressed and we actually have the same thing in research where I'm like I can't guarantee you that the research breakthroughs are going to happen in three months on the dot this is not how this you got to deliver your Q2 breakthrough Marina right I can't I can't promise my Q2 breakthrough so um I would also say that this is also to some extent uh a mismatch between the schedule of Wall Street in the schedule of research in a in a new area in an area that we don't yet understand very well and that's I think a lot of what we're actually seeing here yeah that's fascinating is almost you're saying like we should not be looking to the stock market to judge the value of the AI space in part because like the market doesn't know how to Value it at the moment is kind of what you're saying is like I don't think it's very clear yet I I don't know if K but you disagree but I actually don't think we actually know very well yet how to Value AI properly I'm I'm with Marina on this uh and I I don't think that the common uh stock investor understands the impact especially in Enterprise space and what can do for people so we've been we've been just dunking on AI stocks saying that hey you are leading to the downfall of economy we look at the positive that it has done right it's also contributing insanely towards the overall economy right so you should give AI enough credit to lift the entire stock market up as well not just the training week and say hey U the the market is down X points because of the large Nvidia swing which like just look at the world we live in right in the last few months Nvidia has swung a trillion dollars in market cap a trillion this just pause and realize how much of an impact that's having on on people right it is people reacting to oh my God I don't want to miss out but also not knowing at what point are you investing in the fundamentals or are you pulling out of the stock too early right like even massive companies like Arc invest ended up missing the boat on Nvidia lost a billion dollars of opportunity there right so you need to understand the fundamentals and stay long in the market versus going and reacting to these quarterly ups and downs I don't know I'm you're not arguing this but I think it's almost like you could make out the argument that like you know what the biggest meme stock in the whole world is it's Nvidia you know it's it's not it's not GameStop it's not anything like that I was just gonna agree with Marina I mean uh the fact is that uh and what chth was saying as well I mean this is a it's a long game uh we don't really know how to Value things uh yet uh it's not like some commodity where you can like grab it and hold on to it and see what it's doing so um uh I think we'll we'll get better um just like we've had trouble valuing data as well um uh valuing the models and what we can do with them is going going to be part of this as [Music] well so I'm going to move us on to our second segment uh of the day um so open AI this week announced the new feature uh they call structured outputs um and uh this is huge uh although it might not seem like it on the surface for people who are like not in the day-to-day work of of AI um effectively what they're offering is for the very first time uh model developers uh are allowed to basically work with their system um to constrain their outputs to match specific schemas that are defined by Engineers um and this is a little bit nerdy but I think it's actually worth kind of walking through the technical points here because I think it's one of the areas where if you dive a little bit into the technical kind of understand what's going on you may recognize why out of a summer of lots and lots of announcements of AI this may actually end up being the biggest announcement of the summer in some ways um so I'm going to try to explain this and then I think Marina you'll keep me honest you should be like that's completely wrong Tim you've completely misunderstood what they're trying to do the way as I understand it is that language models are of course Very powerful they can do all sorts of remarkable things but the problem is that they kind of output in sort of non determinative ways they like produce outputs that are kind of difficult to kind of constrain and standardize and this has been a really tough problem because you know you have to take Ai and then you have to connect it to all the other all these other traditional systems that are expecting structured data right like there's a computer just being like Oh well I'm expecting a table that has like the following elements within it and it's been very hard to kind of like integrate language models with that um and is what open AI is saying here that you can finally for the first time do that reliably uh correct me if I'm wrong I'm just kind of thinking through this the thing I'm actually going to push back on is this whole finally for the first time thing this is not for the first time the fact that we before were like all right structured outputs semi-structured outputs are where it's at we used to say what you do with unstructured data this is work that I've done for years is you try to turn it into something more structured so it's features and you can feed it into a you know classifier feed it into ML and and go from there then everybody said oh Foundation models all right now we can doesn't matter we no more structure is needed no more data is needed nothing is needed we're just going to go and have unstructured data is going to everything you go and you work with that for a while and you go no guess not all right we're going to go ahead and walk it back a little we're going to walk it back a little let's go back to the fact that especially if you're trying to mix and match a heterogeneous system you do need structure output because these things don't know how to talk to each other so I'm going to pretty strongly push back on the for the first time and go back to no now that we're trying to be practical about it we've gone back to the fact that you need to impose a bit of structure I would also say that this is with like the success of uh code models where we see that there already is a lot more structure imposed on what kind of things can go in and can go out there's some lessons being learned there again going oh maybe we don't do just generally unstructured text and we're going to go back to having a bit of a mix um Kush would you agree with that particular we're kind of back yeah yeah no I mean I think that's exactly right I mean one way to look at it is I mean you're kind of breaking this bucking bronco that uh just came out of the blue in the last couple of years and bringing it back to uh where it should be right I mean the the control and the governance I mean all of that is part of making these things practical right and um I think another way to look at it is uh and one good thing about these language models is uh that they're very creative they're coming up with all sorts of uh different things but it's really a tradeoff safety versus creativity and um the control the the constraint is bring us back to to that safety aspect and um uh if you're inspiring a poet I mean go ride that Bronco it's all good um but uh I mean for all of the Enterprise use cases that uh uh that we care about that are going to make the the productivity differences and all that sort of stuff then uh that extra control is is where it's at yeah for sure so show bit am I um am I just being an open AI shill here just really hyping this feature where I guess Marina is just telling us like this has all been said and done before you know they're just selling something that everybody has known how to do for a long time so so Tim hot take on this this is the first time open AI is now appreciating and admitting that the whole workflow end to end won't be done by an llm they have admitted by releasing this that at step number three somebody's going to call an llm and expect it to behave in a structured manner so it can be a part of a team that does an end to and flow other aspects will be automation RPA there'll be some regular AI they'll be just PL old API calls but now llm they have admitted to this by releasing this that it's now down to a subtask level versus being the llm that's going to do the entire process end to end right so I think it's it's a really hard take on what they're doing for for practical deployments for me in the field we uh we are we are the launch partners with open Ai and whatnot right we do a ton of open AI with clients in our workflows last week um on Monday actually we were working with a large Healthcare client where we're reading greams of different documents and stuff and we extracting things from those documents right so talking about my HealthCare coverage I need to know what's in network what's out of network what's family coverage what's single coverage and so and so forth so using an llm to go extract things out from it every time we run this against our rubric of uh checking the accuracy there quite often response back with a blurb instead of giving me the in network and out of network so the way we used to solve this historically we would ask questions in a manner and then we provided some coaching saying just respond with the actual dollar amount the problem there used to be it responds back with saying 14.9 and in three out of 10 cases it'll forget to put million in front of it right there's like practical issues with you having with leveraging these large language models and then we like okay fine just give me the entire thing and then like to Marina's point I'll just use a small regx somewhere to extract what I need from it and then I'll plug it back in that was a horrible way of doing things in production yeah that's awful having a commitment now saying that this is the JS I'm going to get and if you can't fill that number if you don't don't know what the single coverage is for outof Network it'll be null it'll be blank then I can do something in a structured manner raise some some alerts and have a workflow accordingly I think it's brilliant they're allowing you to do this this combined with the price drop that we got 50% decrease in inputs 33% in the outputs makes it very very easy for us to plug it in the 40 Mini price is just Rock Bottom it's slow it's very inexpensive to deploy mini even the fine-tune versions of mini now they're allowing you to go fine tun these models very very easily have a structured oper around it so they've understood the fact that instead of doing a generic top down I'll take care of the entire thing all the way down to a subtask level it has to be fine-tuned for that task it has to be super inexpensive and has to be a good contract on what the input and the output structure coming out right in other words like a good a good tool to be used in the Enterprise um so the super interesting takes on this it definitely went in a direction that I wasn't expecting but I think is like very helpful in kind of thinking through why open AI did this I think the final aspect of this I want to touch on is it was very funny I mean as someone who you know is a software engineer kind of turned into a lawyer you know I like read this very long blog post about structured outputs and then at the very end it's like oh by the way it's not eligible for zero data retention which I think was a very interesting part of the announcement was basically like normally the promises that open AI will not train on any data that you send in through the API on the Enterprise basis but in this one case if you send in a schema they're going to they're going to train on that um and I guess for our listeners I think it'd be useful for them to hear a little bit some intuitions for why it is that open AI sees this data as so uniquely valuable right that they're going to say we've got this General policy of zero data retention but for this tiny little segment we're going to cut out a hole and if you send us our schemas we definitely want to train on that um Chris I see you nodding but I don't know if you want to speak to speak to why they would do something like this yeah I mean uh I was reading the announcement as well and uh I think the there's they're taking two different technical approaches to make this work right one is just training on more and more of these schemas the second is constrain decoding using this uh context free grammar to really make sure that um uh what comes out is uh is really I mean matching the the schema and stuff so on the first of the two I mean it's really hard to to get this sort of variety of uh what kind of schemas are going to be out there this is not something you can just download from the web and uh I mean in some of our work we also I mean look at very unique Enterprise sort of um uh policy documents or or other stuff like that and it's just not easy like um I was uh talking with one of my group members yesterday we were trying to figure out what are like different policies for um or guidelines for different professions and I was looking like can I get the New York State barber license uh guidelines like what does a barber need to do to do their job and like there's tons of stuff like that that um is like really not out there so I mean just the the uniqueness of it is is the key I think I think that's that's absolutely right and I think that will be coming sort of the increasing battle it seems like right as like all of the easy to get data is now accessible now the kind of question is like who's got these kind of access to like very hard to get data and this kind of these schemas are they're they're valuable tokens right they're they're unique tokens um in a lot of ways so this has been a big struggle for us with our clients in Enterprise settings we go through Enterprise security govern when we take a new product and we have to make sure that it's being used in a particular way everybody signs off on it and so on so forth right so we we're struggling with this with our Enterprises when when you Outsource your API calls to a third party then every time the API calls change or they do something differently or now in this case there's the retention issue with with the schemas right you need to go back through the whole process and I don't think Enterprises have a good mechanism to understand capture and then act on each one of these incremental updates that happen so it scares me a little bit that enterprises will end up approving a product in a particular state but it so rapidly evolves with features and stuff that you won't be able to go back in time and say I have to this small incremental thing has to be done differently the data scientists will start getting super excited about these function calls and and about these structured outputs and start using it and then that's where Kush and team are going to come in and say guys time out there has to be a good discipline around how you govern incremental updates that are happening to these so you don't get yourself into trouble so I think that's a very unaddressed issue with at least my Enterprise [Music] clients so I'm going to move us on to our final uh story of the day um it was announced last week that Nome shazir who was the CEO of character AI was going to rejoin Google along with a core team from his company um and also that Google was going to acquire a license to all character IP um this is widely seen though it's disputed as an acquisition ultimately of character um which had raised something like $150 million and was basically building sort of personalized companion AIS um and so I really want to go into this story because it's very interesting and part of a trend of uh Acquisitions in the space if you will um that I think are very interesting and I think get us to thinking a little bit about how this Market's going to evolve and what we really anticipate from AI startups the next 12 to 24 months Kush I wanted to turn to you first is you know why is a company like Google interested in a company like character AI at all you know like it feels like Google's got all the resources in the world to do all the AI um why are they acquiring companies at all at Great cost like it feels like couldn't they just build kind of a character product on their own um and we' love to get your thoughts on what do you think is motivating this in the first place yeah I think that's the similar question like why does IBM research exist versus um why don't we just tell me a little more about that yeah yeah I mean why don't we just keep acquiring a lot of startups I think uh there's always going to be a balance between kind of organic growth and and uh kind of uh the acquisition sort of thing um uh there's always a spark of some idea it you can't assume that uh you're going to have all of them and uh uh I mean in these cases there is something unique there's something where there's a market that they've touched on and something that I think only a startup can can maybe tap into because um they have a different pulse of the scene so I think it it makes sense to for a company like Google to to have a a mix of ways that they they grow yeah for sure I to push you a little bit further on that do you think it's cuz like is there some kind of compliment like what's the angle that I think you think Google's trying to chase after here because I I mean it's a search company right like ultimately um this feels like very consumer uh in some ways of what they're trying to do yeah I mean maybe they don't think they're they are a search company going forward I don't know um maybe they're uh edging on to there's I mean more things or or other things but uh I think just uh once you get I mean something interesting something exciting uh that just draws customers to you draws consumers to you and then uh uh you can keep them and get them into other stuff so yeah as part of a pivot for sure um so maybe we could take the other angle at the story I think which is you can see it from the perspective of the acquirer why would Google do something like this but I think it's also worth investigating it from the perspective of the startup um you know Marina there was a bunch of commentary online where people were saying look you've seen Adept go through a similar transaction there's another company called inflection that went through a similar transaction these are companies that have raised an enormous amount of money um and by all accounts would be very successful right like maybe some of the most successful startups in the AI space um but as yet the founders are choosing to to sell um effectively right they're they're choosing to go and join the big tech companies um do you have a theory for that like why would you I mean if I'm sitting there I'm n shazir you know I've raised $150 million that's certainly more money than Ever Raised right um what is what is motivating these kind of Founders to say okay well actually want to kind of throw in with the big companies rather than trying to make it on my own and does it suggest you know problems in the startup Market do you think I mean even 150 million can be burned through pretty quickly if you're doing a whole bunch of your own training what is 15 everything else there might be a a case here of again if there's an understanding that you want to have a sort of a pre-baked user base or a pre-baked you know set of being able to use a whole bunch of um of resources which a company like uh Google company like meta they're going to be uh really quite good with that um again potentially other people to collaborate with I really will second what kush said which is you've had one or two or three good ideas it doesn't mean that you're going to have 40 and they really are a ton of extremely interesting smart people who are working in these companies so it may be that there's a desire to to also do that and as well and and have that partnership be a lot more close in order to be able to to see that that yeah I mean zooming out to the macro level I mean cha do you think that um like what does this prage I guess for kind of like startups in the AI space in general like are you seeing more AI startups over time because I think there's almost one way of reading this which is well even if these companies that have raised so much money can't make it independently uh you know like no one can make it right like we're about to see a lot of consolidation in the AI startup space I think the the core values the fundamentals haven't changed you can't have a thin wrapper around an open AI API call and expect it to keep keep drawing more right so you you do realize that the intellectual property that you've built is what people are going to pay for and the talent that you have that you have assembled that particular team that's what is is uh golden now big companies will try to walk around Acquisitions and come get very creative to work around any of the antitrust rules and things of that nature as well right so in this case they're not acquiring it they are getting hiring some people or they're licensing some terms and so on so forth right so you can see that there are there some motivation on not just outright acquiring it but on the flip side just like any startup environment you'll also see big companies like whz which Google was trying to acquire and uh whz walked away from $23 billion offer and uh this I'm just laughing because it's like that's like a literally hilarious amount of money right that is insane and O who's the the co-founder of BZ he wrote a very humbling letter to all the employees explaining them why you're not getting rich today right Essen explain to them why I'm I'm not taking this offer it's a very humbling offer but here are the reasons why we believe that going IPO is a big bigger value ad and so on so forth right historically we have seen a lot of misses and hits and misses Yahoo trying to sell itself to Google or like Netflix to Blockbuster all of these have been multiple reminders that you you have to know what's your value ad and how much of that is a differentiator with a high Moe so others can't just come in and do what you're doing right so it takes a while to understand the rhythm of where you lie where you lie in the competitive landscape and R forecast I think we put undue pressure on co-founders on the on the founders who who are just passioned about building a product but now all of a sudden we are we are surrounding them with Venture capitals who have different objectives than what you mean I need to build a business yeah I think they need to bring back Silicon Valley as a as episodes in today's world with llm that's right yeah it's for sure yeah I saw this great Twitter thread that was on like if we modernized you know Silicon Valley what would it be and just like everybody's in AI basically um I mean it goes to a point that Marina raised earlier in our first segment though is like it almost kind of feels like this is almost like the micro version of the market being not able to price these startups properly like it feels like in a lot of these cases like these big companies like goog are like ultimately acquiring the talent versus necessarily like the product um I guess character you can maybe debate because it actually had a big install base but it feels like at the core of it is just simply like here's a team of people who seem to be able to get what they want out of the AI and like that actually ends up being like this huge value that's almost separate from like did you have a blockbuster AI product release um and yeah it kind of goes to these interesting questions that I'm thinking about now about like how do you how do you actually value these companies right because it's just like so unclear in such a fluid environment um any final thoughts on this um super super interesting and I I think again I mean to argue against myself you know this is also during the same week we saw a bunch of top leadership from uh opening I leave right and so it's not necessarily all consolidation it's possible that you know people are moving between big companies and also creating like new startups onto themselves um so any final thoughts to round this out for today um just one um I mean conversation I was having with my brother-in-law last week not related to this but uh I mean the difference between running your own business versus doing a job in a big company right and the lifestyle sort of issues there and um I think like I mean the point you were making before Tim like uh if you just want to make one product versus building a business I think maybe a lot of the folks that are um getting into this right now um are not in it for maybe that lifestyle or for that uh uh business building sort of uh sort of way of of going about it so maybe it's just a way for them to return back to their natural sort of State um so so that could be driving it as well kind more of the lifestyle issue yeah I believe that for sure yeah it's I mean personally crazy to do a startup so and then got a friend who was a Founder who is like it's literally an irrational act to do a startup so um well great on that note uh no shade to anyone else who has already been on mixture of experts as a panelist but I have to say this is my favorite panel the marina Kush show bit you know power Trio is basically like we just get the best conversations all the time so I appreciate all three of you coming on the show and for all you listeners thanks for joining us this week uh if you enjoyed what you heard you can get us on Apple podcast Spotify and podcast platforms everywhere and we will see you same time next week

2024-08-11 18:53

Other news