The evolution and promise of RAG architecture with Tengyu Ma from Voyage AI

After Tengyu Ma spent years at Stanford researching AI optimization, embedding models, and transformers, he took a break from academia to start Voyage AI which allows enterprise customers to have the most accurate retrieval possible through the most useful foundational data. Tengyu joins Sarah on this week’s episode of No priors to discuss why RAG systems are winning as the dominant architecture in enterprise and the evolution of foundational data that has allowed RAG to flourish. And while fine-tuning is still in the conversation, Tengyu argues that RAG will continue to evolve as the cheapest, quickest, and most accurate system for data retrieval. They also discuss methods for growing context windows and managing latency budgets, how Tengyu’s research has informed his work at Voyage, and the role academia should play as AI grows as an industry. Show Links: Voyage AI Stanford Assistant Professor of Computer Science Tengyu Ma Key Research Papers: Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training Non-convex optimization for machine learning: design, analysis, and understanding Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss Larger language models do in-context learning differently, 2023 Why Do Pretrained Language Models Help in Downstream Tasks? An Analysis of Head and Prompt Tuning On the Optimization Landscape of Tensor Decompositions Sign up for new podcasts every week. Email feedback to [email protected] Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @tengyuma Show Notes: (0:00) Introduction (1:59) Key points of Tengyu’s research (4:28) Academia compared to industry (6:46) Voyage AI overview (9:44) Enterprise RAG use cases (15:23) LLM long-term memory and token limitations (18:03) Agent chaining and data management (22:01) Improving enterprise RAG (25:44) Latency budgets (27:48) Advice for building RAG systems (31:06) Learnings as an AI founder (32:55) The role of academia in AI

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:05Sarah Guo?

  Welcome to No Priors. Today, we're talking to Tenggu Ma, Assistant Professor of Computer Science at Stanford and the co-founder and CEO of Voyage. Voyage trains state-of-the-art components for next-generation retrieval systems, including embeddings models and rerankers. We're really excited to talk about his research and the RAG debate today. Welcome, Tenggu.

• 0:24Speaker 01

  Yeah, thanks so much. Thanks for having me here. We're looking forward to the debate.

• 0:28Sarah Guo?

  Yeah, why don't we start with just a little bit of an overview of your research agenda to date? Because I think uniquely, it covers a broad range of fields within and around deep learning, from like theory to RL to embeddings and optimizers. So can you talk a little bit about how you pick the directions you have?

• 0:47Speaker 01

  Yeah, so I think most of the papers I wrote have some theoretical thinking in it. I guess maybe that's the commonality. And besides that, I think I worked on quite a few topics, as you mentioned, ranging from the theoretical understanding, mathematical proofs of deep learning systems, to all the way to practical large- language models, reinforcement learning, deep reinforcement learning. And these days, recently, I think what we are working on more centralized to efficiency of training the large- language models and improving the reasoning tasks for large- language models. So my vision is that in the future, the efficiency is very important because we are running out of data and compute. So we have to either use the data much better

• 1:32Speaker 01

  and use the compute much better. And also, reasoning tasks seems to be a pretty important direction and also, in some sense, kind of like a risky direction in the sense that we don't know exactly how fast we can solve those challenging reasoning questions yet.

• 1:52Sarah Guo?

  Can you mention a few of the like key papers or work that you know you or students in your lab have done? Just so our listeners can look them up.

• 2:03Speaker 01

  In the very early days, I think I worked on some of this matrix completion optimization for matrix completion. That's like 10 years ago. And then I move on to embedding models like sentence embeddings, vector embeddings. One of the papers we wrote is a very actually simple paper where we averaged the word embeddings to get sentence embeddings. And then we did some of these transformations using PCA to make the performance much better. That was even before transformer came out. And then I think I move on to transformers, large language models and contrastive learning, which is the new way of training the embedding models.

• 2:42Speaker 01

  Especially the direction started with some of the papers on using contrastive learning for images. And we work on improving those and understanding why contrastive learning can work. And recently, we work on optimizers for large language models. For example, one of the paper we wrote last year was Sophia, which we found where we found that we have a new optimizer which can improve the training efficiency by 2X for pre-training.

• 3:09Sarah Guo?

  This is great. Adam is very old at this point.

• 3:12Speaker 01

  Yeah, he's 10 years old now. I think that's the interesting part about it. So optimizers, you know, I think people have tried in the last 10 years for so many times. There were so many papers published which has improvements over Adam in various cases. But so far, Adam is still the default algorithms for training large language models. And that's why we thought that is the time to really, we spent a lot of time on this. Like I think I started probably around 2018, 2019. And I asked a few students work on this. And finally, we had one paper out after a few years, after a few failed projects and failed ideas.

• 3:54Speaker 01

  And recently, I think one of the Facebook friends actually used this in their large-scale multi-model training. And they found that in on that scale, I don't know exactly how many parameters they are, but I think I assume it's kind of more than 100 billion parameters. They found that on that scale, they are, there is a 1.6X improvement in the efficiency of the training. So that's like $10 million versus $16 million.

• 4:20Sarah Guo?

  That's super exciting, yeah. I think Sophia has an opportunity to be really impactful. You started a company last year, taking leave from Stanford. Given your work has been theoretical but with practical applications, what drove you to do that?

• 4:41Speaker 01

  I think I came to Stanford partly because there's a very strong industry connection here at Stanford compared to some of the other universities. And also, probably entrepreneurship is just part of my career plan anyways. And in terms of the timing, I felt that this is the right timing in the sense that the technologies are more and more mature so that it seems that the commercialization is the right timing right now. So, for example, I think one story I have is that I look up some of my slide deck for my lectures at Stanford CS299, seven years ago, when I started to teach at Stanford.

• 5:25Speaker 01

  At that point, machine learning, we have a lecture with Chris Ray and machine learning, on applied machine learning. So, how do you apply machine learning industry? And there are seven steps there. So, the first step is you define your problem. The second step is you collect your data and you choose the loss function, you train it, and you iterate, so on and so forth. So, it's pretty complicated at that point. And now, the foundation model arrives to power. And in a new foundation model era, the only thing you have to do is that you have to, you know, someone will train the foundation model for you. And then, you turn the prompt and you add a retrieval argument generation on top of it. And that's pretty much, that's it.

• 6:06Speaker 01

  So, applying machine learning, AI, to an industry environment is much, much easier than seven years ago. And that's why I felt that this is probably the right time to commercialize many of the technologies because the technologies are more mature.

• 6:22Sarah Guo?

  Yeah, this is actually, I mean, a core premise even for the investing fund that I started a conviction in that, you know, somebody's doing the bulk of the work for you in a more general way. And so the application of AI in industry is just much, much cheaper, right? Because you only do, you know, the last few steps and, or different set, but last few steps in essence. So maybe you can talk about like the, you know, just given wide range of research, the problem you focus on with Voyage that you saw with customers.

• 6:51Speaker 01

  Yeah, so with Voyage, I think we are mostly building these two components, we rank and embeddings for improving the quality of the retrieval or the search system. So the reason why we focus on this is because we talked to so many customers and we found that right now, for implementing RAG, the bottleneck seems to be that, you know, it's not very hard to implement it, right? You can just connect the components and have your RAG system ready very quickly. But the bottleneck seems to be the quality of the response, and the quality of the response is heavily affected or is kind of almost bottlenecked by the quality of the retrieval part.

• 7:31Speaker 01

  If the large language model see very relevant documents, then they can synthesize very good answers. Even like Llama 7B can do that very well.

• 7:39Sarah Guo?

  Can you just give like a sort of general intuition for like what a RAG system is and some of the applications of it?

• 7:45Speaker 01

  Yeah, so I guess just a little background. So a retrieval of my generation, the idea is that there's a retrieval step, there's a generation step. So the main point here is that if you just use a large language model as a black box, you know, as is, then the large language model wouldn't know anything about the proprietary information inside the company. And I doesn't know enough context about the use cases. And the retrieval of my generation stack is about you first retrieve some knowledge from, for example, inside the company and then use this knowledge and give the knowledge to the large language model so that the large language model can generate or synthesize a good answer without any hallucination. This has been found to be very, very useful to reduce the hallucination rate.

• 8:31Speaker 01

  And so there's Two steps, the first step is to retrieve some relevant information given the query and then this relevant information are given to the large language model. The retrieval step is important because once the large language model sees the relevant information, it can reduce the hallucination rate dramatically because it uses the relevant information as an anchor to refine the answers in some sense. And what we are doing here is that we want to improve the quality of the retrieval or the relevancy or accuracy of the retrieve documents and information. And the way to do this is that there are two steps. The first step is that you vectorize all of your documents or all of your knowledge base. So you turn the documents to vectors, you turn the videos into vectors.

• 9:16Sarah Guo?

  You turn your code into vectors.

• 9:18Speaker 01

  Code into vectors, everything into vectors. And so the vectors are the representations of each piece of the knowledge or documents. And all they are the indices. So and then you put these vectors into a vector database. And and and then you search the relevant information using the vectors as indices.

• 9:36Sarah Guo?

  Where are you seeing RAG applications today? Like what are what are customers building or you know, what are the most common systems?

• 9:44Speaker 01

  Yeah, so we have a lot of users and and they are all over the places. I think, you know, we have even a customer who is a chemistry company who is building this RAG system to understand their chemistry documents or products in descriptions. And I think just it's almost everywhere, like finance, legal, code retrieval, code generation, so on and so forth. I think it can be applied to almost any cases and also for even for individual users where you have a lot of like individual

• 10:19Speaker 01

  personal information and you want to have a RAG system on a phone so that you can access your past information much in a much more easy way and and you want to retrieve. For example, we all have seen that, you know, when you search your documents on your laptop, it's actually pretty hard. You have to use the exact file name. It would be much easier if this search can be semantic based.

• 10:41Sarah Guo?

  RAG is a relatively new architecture. I think your average enterprise technology leader had not heard the term before the last year or so, and it became popularized in research over the last few years. But there is already a debate, I think, you know, in terms of opinions from people at different large labs and in academia about whether or not you need a RAG architecture to work on proprietary data. And just to sort of describe some of the alternative views, I think there's kind of two, two alternative points of view given.

• 11:23Sarah Guo?

  One is a sort of agent chaining architecture where you are inputting your data and knowledge, you know, chemistry code, law, finance, whatever documents, into a series of LLMs that just operate with instruction on it, for example, to summarize or categorize it. Or you simply feed everything into LLMs with infinite context or actively managed context versus explicitly vectorizing anything. And so I would love to get your reaction to, you know, that as a, as an alternative to RAG.

• 11:57Speaker 01

  Actually, there was also a debate last year about RAG versus fine tuning. And I think that debate was kind of like getting to a consensus now. It sounds like RAG is much easier than fine tuning. And fine tuning, in many cases, doesn't work because you need a lot of data to see the results, and there are still hallucinations even after fine tuning. And now, as you said, the debate becomes RAG versus agent chaining or long context. So maybe let's talk about long context first. So I think there are probably two answers to this, you know, from different angles, because long context right now is not practical yet, right?

• 12:38Speaker 01

  So we have to kind of anticipate what long context transformer can do and then do the debate at a future time in some sense, or anticipate the debate at a future time. In the near term, I think the long context Transformer where you just put in all the proprietary data one billion tokens into the context of the transformer. So um will be very, very expensive, right? If you use the price right now, it's gonna be just impossible to do it. It's probably like five, ten magnitudes of difference depending on how many documents you have in the context. Of course, you know, you can bring the cost down by, for example, one approach is to catch

• 13:18Speaker 01

  the activations of all of the internal operations of the documents you put in the context. So that will bring the cost down by a lot, but I think still if you do the calculation theoretically is still much more expensive than RAG. So I think that's the more practical answer. So in terms of cost, it's gonna be much more expensive than RAG because you have to save all of this activations or intermediate computations in the GPU memory, most likely, or maybe in CPU memory of all the, you know, one billion tokens context. You know, you may argue that, okay, over time, everything will become, you know, cheaper and cheaper. But RAG will be cheaper as well, right?

• 14:03Speaker 01

  Because many of the technologies under RAG are neural network based and the GPUs will become cheaper, the neural networks will all become smaller. So my prediction is that RAG will be much cheaper than long context going forward.

• 14:17Speaker 01

  And another way to think about this is that maybe just on the first, from the first principle, right? So my analogy of long context is that, so in some sense, the context is the short-term memory in some sense, right? And the RAG is more like long-term memory in some sense. So the question is, you know, for example, you know, when you answer any question, why you have to go through the entire library every time, right? Like, put all of the entire library in your, in your short-term memory for answer a single question, right? It sounds like the right approach should be that for every single question, you retrieve some subset of the information and use those to answer the question. That seems to be the most efficient way to do that. It should be some kind of hierarchy in some sense in terms of how we solve the problem so that.

• 15:01Speaker 01

  That we can get the best efficiency. You know, even when we do the the computer architecture, like the the hardware stuff, right? So you have a different level of caching, right? So you have disk, you have a CPU cache and so forth. So so in that sense, I feel like the the more hierarchical two level kind of like system like RAG is more cost efficient.

• 15:22Sarah Guo?

  Yeah, I mean, the analogy certainly makes sense. I think there is another thread of discussion of like, what does long term memory for L L M's look like? Where, you know, it is something managed by the L L M itself. But I do not think that is a well answered question and like RAG may just be a part of that answer.

• 15:40Speaker 01

  So the the embedding model, the Rebanker are in some sense the large language model that are managing the long term memory. Of course, there might be variants and other ways to manage the long term memory. But I think it will be somewhat similar. It's gonna be like more, you know, the technology always evolves, right? Gradually, right? So maybe two years later, Voyage or maybe other companies will have a new version of the long term memory, which is based on, you know, embedding models, but you know, kind of like extending the embedding model in some way. That's that's entirely possible.

• 16:12Sarah Guo?

  Yeah, I do think it's useful to sort of contextualize for people who are not working with sort of data sources for L L M's at scale every day, like what sort of token limitations are, right? You know, we go from a few thousand tokens to something like Gemini one five pro, a context window of a million tokens, right? And so if that's short of that in word count, that's maybe five books or like twenty five, thirty thousand lines of code. And obviously, like limited amount of video and audio. And so I think the ability to make reasoning decisions on more than that amount of data is obviously going to be needed.

• 16:55Sarah Guo?

  And then the questions are, to me, are really like, you know, does does efficiency matter? Both from a cost perspective and a speed, like a latency perspective, right? And how much can you push the context window? And like, you know, does hallucination management matter? And so I I think there are lots of arguments for like RAG being very persistent here.

• 17:17Speaker 01

  Yeah, yeah, exactly. And just to, I, a little bit on that, so one million tokens, five books, right? So but many companies has a hundred million tokens. That's a hundred X difference, right? So a hundred X, you know, into for cost is a, is a big difference. That could be just, you know, a hundred K dollars versus like ten million dollars, right? Ten million dollars is unacceptable, but a hundred K sounds okay. Yeah, I think that's probably what's gonna happen like so. So from, at least for many of the companies, right? So right now, if they have a hundred million tokens, I don't think they can use long context transformers at all because it's way too expensive.

• 17:49Sarah Guo?

  Right. And I'm, I, like, the simplest thing for me is actually for a system to look at the entire code base or some representation of the entire code base versus the portion of it that could fit into context today.

• 18:01Speaker 01

  Yeah.

• 18:02Sarah Guo?

  What about the other piece, like the idea of agent chaining and using L L M's to manage the data in, in that form?

• 18:09Speaker 01

  So agent chaining, this is a, you know, growing area and many people are doing research on it. I think this is sort of less well-defined in some sense. On the, the first level bit I would say is that I think it's kind of orthogonal to embedding models and re-rankers to some degree because even when you have agent chaining, right? You still probably use embedding models as part of the chain, right? You probably do iterative retrieval as part of the chain. And of course, you use large language models as part of the chain as well. In some sense, it's orthogonal direction. So I probably rephrase the agent chaining as more like a iterative um, um, multi-steps retrieval augmented, uh, uh, large language model augmented uh, system.

• 18:53Speaker 01

  So, uh, and some part of this retrieval probably is done by a large language model, sometime part of the system is done by a small large language model. Um, and some part of the system is done by embedding model, so on and so forth. So in that sense, I feel like it's somewhat kind of orthogonal.

• 19:07Sarah Guo?

  Yeah, and I, I, I feel like, um, some of the motivation for agent chaining to begin with is the same efficiency motivation as RAG.

• 19:16Speaker 01

  Yeah, exactly, right. But if you use a very, very large language model to manage to manage the system, the knowledge system, I think you again lose the efficiency, right? So so it has to be a somewhat kind of like smaller model to manage the

• 19:32Speaker 01

  the knowledge. Uh, and then at that point, embedding model might be the right thing to do in that agent training framework. Maybe another angle to look at this is that whether we should do iterative retrieval versus um uh just retrieve at once. Um, I think iterative retrieval is definitely useful, especially because now there are still a lot of headroom uh in the uh embedding models' performance. So that's why sometimes you have to retrieve multiple times because the models are not clever enough. Um, however, in the long run, my suspicion is that iterative retrieval will be useful, but it will be a bit less useful as the if the embedding models becomes more and more clever, right?

• 20:14Speaker 01

  So uh once the embedding models are more clever, then maybe one round or two rounds is gonna be enough.

• 20:19Sarah Guo?

  Uh huh. Um, if we go ahead and just assume that RAG is at least a dominant architecture for enterprise use cases where you care about proprietary data that is large with reliability. How do you go about improving like a RAG system, right? You can improve the LLM itself, but what are the what are the other components that you guys are working on? Or what are the maybe challenges from the user, um the builder's perspective to improve retrieval quality?

• 20:45Speaker 01

  Yeah, so I guess there are a few ways, right? One way is that you improve the prompting of the large language models. Um, so um, for example, you could tell the large language models to abstain if there's no relevant information uh in the in the retrieved documents. But because the large language models are so good these days, I think you don't need a lot of prompting anymore. Uh, it just responds to the instructions so well. Um, and then the next thing is to improve the retrieval part, which is the bottleneck, in my opinion, because most of our users found out that if they re-improve the retrieval quality directly, that affects the response quality. Um, and improving the retrieval part, I think there are two ways. One way is you improve the embedding model. One way is that you Improve some of the other things on top of that.

• 21:28Speaker 01

  For example, how you chunk the data, whether you do iterative retrieval, whether you put in some of the meta information in the data, so on and so forth. So basically, I would say there are two ways of improving. One way is you improve the new networks, either the embedding models or the re-encoders, or you improve the ways to use the networks with software engineering, right? Better chunking iterations or other kind of like heuristics or kind of like tricks on top of that. So and what we are specialized in is that we want to improve the new networks because that's a requires a lot of heavy lifting. That's a very data-driven approach. We train our new networks on trillions of tokens at least, and we find them for special use cases.

• 22:12Speaker 01

  And this is something that probably what a company should do instead of like every the users, the end users should optimize themselves. And my long-term vision here is that some of the software engineering layers on top of the networks will be less and less needed when the new networks are more and more clever. So for example, right now, we already see that trunkings becomes less needed because the context window becomes longer and longer. And the long context embedding model, no, relatively long context embedding model. Long context here means like 10K, for example, maybe 16K, so that you can put out 50 pages PDF into it. Because this long context embedding model becomes much better,

• 22:53Speaker 01

  there's less of a need to chunk the documents into pieces of like 5, 12 tokens. And I think this will happen, you know, in other dimensions as well, right? So maybe in the future, you don't have to turn your images into description images and then give it to the text embedding model. That's what people are doing right now, like everything is turned into text and then use a text embedding model. But when the embedding models are more clever and multi-modal, then you don't have to do that anymore.

• 23:19Sarah Guo?

  Can you talk a little bit about just like the intuition for how fine tuning or domain-specific embeddings improves performance?

• 23:30Speaker 01

  Yeah, fine tuning and domain-specific embedding models are what we are very good at at Voyage. So just to have some context here, so what we do is that we start with a

• 23:40Speaker 01

  general-purpose base embedding model, which is also what we trained from scratch. And from there, we first fine-tune or continue pre-train, whatever you call it, on some domain-specific data. So for example, we fine-tune on 2 trillion of code snippets, tokens, and then we get a code embedding model. And we do the fine-tuning on 1 trillion legal tokens, and that's how we got the legal embedding model. And these domain-specific embedding models, I didn't use any of the proprietary data so that everyone can use them, but they really excel in one particular domain, and the performance in other domains are not changed much. And the reason why we do this is because the number of parameters in the embedding model is limited.

• 24:25Speaker 01

  So because you only have like, you have a latency budget, something like maybe 1 second, sometimes like 200 milliseconds, you know, some people even want 50 milliseconds. And then basically, it's impossible to use more than 10 billion parameters for embedding models. And we have limited parameters, any customization is very important because customization means that you use the limited number of parameters on the right tasks, the right domain, so that you excel in that domain. There's no way that you can use these 10 billion parameters to excel in everything. So that's why you have to specialize in one domain. And we have seen like 5 to 20% of improvements by this domain-specific fine-tuning,

• 25:08Speaker 01

  depending on the particular domains. For code, we have seen 15 to 20% of improvement, partly because we have a lot of data there, and the headroom there is also bigger because code retrieval requires a lot of deep understanding of the algorithmic part of the code. And for legal domain, the baseline is a little better, so that the headroom is slightly smaller. So that's why we see 5 to 15% improvement depending on the data sets. For some of the very complex legal data sets, we have seen bigger improvements.

• 25:39Sarah Guo?

  Just to make sure that our listeners can picture exactly like where the latency cost is coming from here in a search system like your data, you know, has been vectorized by an embeddings model, but then every query also needs to be translated into an embedding and then compared to the embeddings of your knowledge in order to feed that that L M for the generation that you want, right? And so there's a sort of there's inference time latency here as well. I just think that's not obvious if somebody hasn't built a RAG system.

• 26:11Speaker 01

  Yeah, exactly, exactly. So basically, at any first time, you have to first turn the query into vectors and then do the search with vector database. And actually, related to this, the dimension of the vectors you produce also affects the latency for the vector based search. If the dimension of the embedding is like 100, is only 100, then it's going to be much, much faster than when the dimension of the embeddings is a thousand. So and actually, this is something we are very good at as well. So we produce embeddings that is like a 3X, you know, 4X smaller dimension than some of the competitors.

• 26:46Sarah Guo?

  Yep, that makes, I mean intuitively, you are creating embeddings models that use a limited number of parameters and dimensions, just given the sort of latency budget that you that any application has to create the best possible representation of proprietary data or domain-specific data.

• 27:08Speaker 01

  Yeah, exactly. And going back to the the domain specificity and fine tuning. So the second level of customization is that we can customize to a particular company, right? So we fine tune on the proprietary data of a particular company and we can see 10 to 20% improvement on top of the domain specific in fine tuning as well. So of course, there's a total budget in terms of how much additive improvements you have there, right? So if you start with like 50% accuracy, then you only have 50% headroom. But if you start with 90%, you only have 10% headroom. So the the improvement, absolute improvement varies a little bit across the domains.

• 27:48Sarah Guo?

  Maybe just advice to people who are building RAG systems. At what point do they begin to invest in, you know, some of these retrieval components?

• 27:58Speaker 01

  Yeah, I think they can do it even from day one, as long as they have a prototype, you know, available, right? So basically, my default suggestion for our users is that when they have the RAG, you know, first of all, of course, you want to connect the components and at least see some response. And then probably do some kind of like basic profiling in terms of the latency and the quality, right? So you can check the retrieval quality, meaning that how often you retrieve relevant documents. There are some default ways to evaluate the retrieval quality. And then you also do the end-to-end evaluation for the responses. And then you can see which part is the bottleneck. And in many cases, people found that the retrieval quality is not good, so that the final response is not good.

• 28:44Speaker 01

  And then you can swap some of the components. You can say, I'm gonna try voyage embedding, I can try the voyage re rankers, which we haven't discussed too much about. So, and you can try various different embeddings and possibly various different large language models as well.

• 29:00Sarah Guo?

  Maybe just zooming out, like, you know, you started by saying, in order to have the debate about RAG versus alternative architectures for working on proprietary data, you need to predict forward, right? Any predictions for how these systems change as LLMs improve dramatically? Right? If we look at the next generations of OpenAI and, or GPT and Claude and Mistral models and Llama and such?

• 29:27Speaker 01

  Yeah, so my prediction is that the system will be simpler and simpler. Maybe this is my biased view. So, or at least this is something that we are working towards. So the ideal world would be that it's a very, very simple system. So you just do, you just have three components, like large language model, vector database and embedding models, and maybe four components, another re ranker, which refine the retrieved results. And you connect all of this and each of the new artworks does everything else. You don't have to worry anything about chunking, multimodality, changing the data format, because new artworks can do more. Most of them, right?

• 30:08Speaker 01

  So seven years ago, if you talk to any of the so-called language models seven years ago, you have to turn the the format into a very, very clean format. And now you talk to G P T four, you can have typos, you can have all kind of like a weird formats, you can even dump J S N files to it, right? So the same thing would happen for embedding models as well. So my vision is that in the future, A I will just be that a very simple self-engineering layer on top of of a few very strong neural network components.

• 30:40Sarah Guo?

  Uh huh. Yes, I think the bias toward it is actually all going to be A I versus complex, you know, discretized software systems is is clear, but I believe directionally right. Maybe zooming out to just get a little bit of your perspective as a founder, like, you know, what's one or two top learnings you have about starting the company as a as an academic before even, you know, despite your work with Google and other companies before?

• 31:07Speaker 01

  Yeah, I think it's very, very different. Founding a company is very different from doing research at big tech and and also even from, actually, it's a little closer to being academia because to run a university lab, I'm the C E O, C T O, C F O, and and and H R for for the university lab, right? So you you you you touch on a little bit of everything, but at a slightly different scale, right? So I think one of the biggest thing I learned actually is from one of our angel investors is that I should read some of the books. Even those, I think, for probably experienced entrepreneur, many of the books are very basic, but for me, they are very, very useful when I read some of the even the basic books, including Ela's book, by the way.

• 31:51Speaker 01

  So but his book is a little bit advanced in the sense that his his book is talking about how to scale from ten people to a thousand people, and I only read a few chapters of that because we are about ten people right now. So, um, yeah, and also talking to a lot of angel investors, talking to Sarah and my other lead investors. So so I think all of this helped me a lot in reducing the iron-fosst mistakes.

• 32:18Speaker 01

  In this process. To me, I think it's really about how to reduce the number of errors you make. So that you can maximize the efficiency. At least this is what happens to me. And also how to correct the mistakes as fast as possible. Right? If you could correct mistakes every one week after you made them. Versus like one month after you made them, then that's a ForEx efficiency improvement.

• 32:43Sarah Guo?

  Uh huh. Um, very theoretically consistent in your, you know, vein of research. Last question, you know, you have been personally productive, productive research lab, but you've started a company. What do you think the role of academia in AI is in this age of like scaling? Because most of your former students, like they essentially all work at OpenAI or Anthropic with, you know, a few professors and citadel folks in the mix. And the ones working with you.

• 33:13Speaker 01

  Yes, yes. In academia, this is a little bit controversial topic. I think different people have different views. My view is that.

• 33:22Speaker 01

  I think academia probably should work on some different questions from what industry is good at. Right? So if we are just only working on how to scale up the system, then obviously the incentive is not right. The, we don't have enough capital there. And, you know, even OpenAI, I guess, Seth Altman argues that you need a lot of capital to start to do this in some sense. So, you know, like at the very beginning, I think the, the point is that, you know, you first have, it cannot be nonprofit. Because if it's nonprofit, then you don't have enough capital and you cannot scale up enough. I think I kind of agree with that. And that's why in academia is very hard to scale up and have enough resources to do the large scale research.

• 34:07Speaker 01

  However, I think in academia, there are many, many other things that we can do on a smaller scale. And we probably should focus on more long-term innovations. So what I told my students at the lab is that we should think about what will be the, the breakthrough in three to five years. As opposed to how do you help OpenAI to. To improve their large language models in the next in G P T five. So so that's why we work on optimizers, which is like ten years old. The item is a ten years old optimizer. And we say, OK, that sounds like a long term project. Maybe in five years, we can improve the optimization efficiency by five to ten X. That's gonna be a game changer for the whole landscape, right?

• 34:48Speaker 01

  So if we improve the efficiency by ten X, I guess that's like a hundred million dollars or ten million dollars for training G P T five, then I think that would change the landscape a lot in the industry. So so efficiency is one of the things I spend a lot of time on. Another thing is that there's reasoning tasks. I think the reason why I identified that as one of my lab's direction is because it's challenging and it requires a lot of very innovative research. It's very unclear whether you can really, the scaling law is really enough to get you to prove Riemann hypothesis or any of the math conjectures. So and also you have to be superhuman performance in some sense, right?

• 35:30Speaker 01

  So if you turn on just the the common crowd data on the web, can you be a good mathematician? It's kind of very hard to believe that. So so we need more innovations there. So that's pretty much what we are doing at the university lab. We try to work on the three to five years agenda and on a smaller scale.

• 35:50Sarah Guo?

  I think that's an inspiring note to end on and like a very open minded one about what is still to be figured out. Thanks so much for doing this, Tango.

• 35:58Speaker 01

  Thanks so much.

• 36:00Sarah Guo?

  Find us on Twitter at NoPriorsPod. Subscribe to our YouTube channel if you want to see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.