From Voice Agents to AI Avatars with Alexander Smola - #777

30:15Sam Charrington:Have you published or said like what your base model or models are, or is that proprietary? It depends. So in some cases, it's proprietary. So for instance, one model that we released last year was built on top of LAMA and we acknowledged them appropriately in our model card.

30:38So basically some of those design decisions are a little bit variable for, you know, whether we just build an audio app model or an ASR model or whether it's an audio understanding model. So you do need the full pipeline, but we can either, you know, build and train on top of an existing model or we can build our own. But that's then often an economic decision and also, you know, which features you need. So, for instance, if you need particular performance in a particular language, then we may very well decide to invest also significantly more still on the language ability there. So, it really depends on the use case.

31:29But, you know, obviously, you know, if somebody gives you free steel, you don't build a steel mill, you build a car, right? and you know as long as you know those models are available it makes sense to take advantage of it it would be economically foolish not to but I would say the work that's required in building the audio models is somewhat commensurate with the language model itself maybe it's half or one third but it's not one tenth so the other thing is for instance if you want to have models that are able to then have a high amount of intelligence in the background you need to specifically make sure that they can do this while they're still maintaining a conversation in the foreground and I mean humans are pretty good at that so for instance you may idly chat with somebody and in the back of your head you're thinking about something else like well did I add some coins to the parking meter or I really need to go or you're actually trying to solve a difficult technical problem so for instance in an exam you may be stalling for time while in the back of your head you're feverishly thinking about the solution, right?

33:08So what goes on there may be, you know, quite variable, but humans are pretty good at that. Machines are still improving, and that's actually also one of the really exciting frontiers there. How do you design architectures?

33:26Sam Charrington:Talk a little bit more about that. But that's a great segue to the question that I wanted to ask, which is really around how you do that. As you were talking about the challenges of audio AI earlier, one of the questions that arose for me is, you know, are we able to do this all with a single model? does getting into hierarchical models you know your system one system two kind of bolted together to create a fluid experience does that ever make sense like how have you come to think about you know overall architecture for these types of systems I mean there are some systems that are just you know end to end audio and you know they have their place for something that's you know super responsive fairly small fairly low latency, but also fairly dumb.

34:25The problem is if you were to try and make them really smart, that would become unaffordable in terms of compute costs. So you really don't want to do that. Now, what you can do is you can have something that's maybe a little bit more of a two-stage architecture where the understanding and reasoning happens in one end with appropriate tool calls being fired off in the back end And maybe then the answer is being received. So it's really just like you would also do in multi-threaded programming, right? Where you have a main thread and it may fire off other threads in parallel that may do things and then get the answer back such that you can get a nice data flow.

35:12And I think a lot of tool calls, agentic programming and so on, have actually been quite helpful there. The other thing is it also allows us to build models that you can instruct just like you would a language agent. The prompts that you write for our model look and feel the same as if you were just instructing a regular LLM, just that our model also talks. So that makes it much, much easier for humans to work with that. And of course, in the back, there is a lot of engineering going on, deciding when to do a tool call, when to answer it directly. And then you also need to figure out when is the question easy enough that the model can answer it?

36:05And when is it sufficiently hard that you probably need something else? And again, that depends on the model size, depends on the information that's required. so do you need a database query do you need to talk to an MCP server right and yeah that is actually something where we I think spent a lot of time and the reason why we spent a lot of time is because we want to make sure that the model is affordable so you can easily build a very big model and that then uses a massive GPU and unfortunately do the user in the end pays for that massive GPU. So right now, on benchmarks, we are better than, let's say, GPT and Gemini and CROC at a fraction of the cost.

37:05So we are better than OpenAI's models at one-tenth the cost. Caveat with that.

37:12Sam Charrington:On what metrics? Intelligence. Speech accuracy, all of the above. So this is, for instance, for, you know, Big Bench audio or complex function bench and so on. So there's a couple of corresponding benchmarks. Now, the little footnote is this is with thinking turned off in these models. Now you might wonder, you know, why on earth would you turn off thinking? Well, because you don't want to have the big pause where the model thinks and then it responds, right? Because that feels very unnatural. So humans don't do that either. I mean, they may say, hey, let me think, but that's now, you know, you need to correspondingly integrate that.

38:01Sam Charrington:I think that's what, that was actually the thought that led to kind of this hierarchical thing. Like I'm envisioning a model whose primary function is maintaining the conversation and it might say, oh, that's a really interesting question. I'll have to think about that for a second. And kind of like you alluded to, kind of stalling, you know, while the tool is completing to retrieve the information. So, for instance, you know, if you test our, you know, you can test it out actually on our demo live afterwards. So basically, for the Higgs live demo, this performs web search in the background. And depending on how quickly it gets the result back, it will just answer or it will actually tell you, hey, let me look for that.

38:59and the challenge is now to make this all feel very organic such that the user doesn't feel any breakage of the entire interaction because you really want to maintain that illusion of a properly engaged other party. So you don't want to break that sense that the model is able to search and do all of those things in the background.

39:32And that's where, yeah, a fair amount of the engineering comes in.

39:38Sam Charrington:And for speech applications, are you able to use off-the-shelf MCP servers for most of the things that you might want to use? Yes and no. So, yes, we can. But the quantity of servers enabled at the same time right now is a little bit limited. the next iteration is going to support significantly larger numbers of them at the same time. Basically, this is all about context window sizes and so on. And again, cost of inference. So there's a trade-off because if you do quite a massive pre-fill and then you have a large KVCage that you need to log around with you, that, of course, makes the token generation more expensive.

40:29and this is really a little bit the trade-off. But what you can do is you can then have tool calls and the tools themselves have MCP configured and so on. So there are plenty of nice ways how you can do this efficiently. Of course, you can also do things where you build, you know, hundreds of billion parameters front-end audio model and that makes for gorgeous demos, but not so gorgeous, you know, cost as soon as, yeah. This is really, I think, where we took a slightly contrarian approach where we went for affordability first.

41:15Sam Charrington:That's very interesting. It's also a little different from what I was asking. And the experience that kind of that was the motivation for my question is, you know, as I build agentic systems with just kind of off the shelf MCP servers, internet, you know, based services, you know, personal data, they're slow, really slow. and it's annoying even in text. It's unimaginable in a voice scenario. And so really the question I was asking was almost, do you have to build everything from the ground up in order to make it work for voice? Like I'm imagining the way you would build a weather MCP server is a lot more efficient than, you know, in an ideal world.

42:09As a matter of fact, you can test out, you know, basically search news and weather and it feels very fluid in our application. So you can go to, you know, Pose on AI.

42:22Sam Charrington:And presumably you didn't have to build those yourself. You're just using... So you use a good server for that. And being able to use these, I mean, I think it's both a feature and a necessity. It's a necessity because you don't want to have to reinvent all the tooling again. it's also necessity because our customers don't want to have to relearn new techniques right and that that's that then also becomes a feature because it makes it easy to or at least easier to integrate within an existing system right but you know if you have a service that takes a long time I mean humans are pretty good if I ask you a question that you have to look up like if i were to ask you hey when was i last time on twimble and i don't think you remember the exact date you would probably tell me something like hey alex let me look it up and then you'll you'd open your laptop and do you know your search and maybe five or ten seconds or maybe a minute or two later you'll come back and say oh this was at that and that date, right?

43:40And it would feel totally normal for me, right? I would not be offended with you spending that time because you told me before, hey, this is going to take some time. So humans are very tolerant to delays if they are told that there's a delay. My favorite example is actually the boot screen from Apple. So Apple does something brilliantly sneaky there. And I guess we all have seen the boot screens on an Apple device, which goes nicely linearly. And then typically before it's even completely done, then suddenly it switches to, okay, it's on, right? And we've also seen the infuriating boot screen on a Microsoft device where it goes and goes and goes, and then it gets stuck at 99 % and you wait for two minutes for the last percent to complete, right?

44:41So you might wonder, you know, how does Apple do that, right? They, you know, after all, you know, is there any secret magic? No, actually, they lie to you. The Microsoft boot screen is the truth. What Apple does is very cleverly, they measure the boot time that it took last time. and they add a tiny amount to it and then they have a fake boot progress bar that's timed to go all the way to the end within the time that it took last time to boot. So in other words, you get this very nice progress bar that is eye candy. It means absolutely nothing. It just indicates how long it took last time. humans are perfectly fine with it right so i don't know whether that is still the very implementation now but it was the implementation for a long time and this is a brilliant slate of hand where i can create a very good user experience without having to solve an impossible technical problem right the other thing that you also need to do is you need to start actually finding benchmarks, like what makes for a good user experience for interaction for voice.

46:05Like, you know, is the model properly interruptible? Can the model recover properly from that? Right. Does it know that, for instance, in, you know, in Japanese, it's very common to say, hey, so basically you're back channeling the other person and saying, hey, I got it. Yeah. Oh, that. Right.

46:32Sam Charrington:But if the agent stops for every one of those, that's going to be infuriating. Exactly. On the other hand, if that person were to say, well, I don't understand, you want the model to stop. Right. So what that means is you need to make sure that the interruptibility is really scene dependent. So for instance, we released a benchmark exactly on that. That paper I think we put it up on Archive a month ago. We looked at the level of productivity in interaction. So basically whether the model actually can pick up if there's something missing and whether the model should also step in. So basically you do need to do extra user experience engineering and benchmarks and measurements and optimization to really make the model work really well in this context.

47:27So that's something that I think is quite different and new relative to what we had in text. And, you know, that's, I mean, to some extent, that's, you know, what AI for humans really means to optimize in a way that it's pleasant for humans. So this is a slightly different optimization track rather than models for code generation, right? they worry about task completion. In our case, we worry about human happiness. You know, you still need task completion and all of that, but you also want to make it enjoyable.

48:06Sam Charrington:What was the name of that benchmark? So ProactBench is the one for productivity. And then there's another one also by the same team. This is basically my Toronto team that has released it. If you look at my blog, there is a fairly detailed analysis and discussion. So IHBench, which explains exactly, you know, the various metrics that we use, how you then also go and make this nicely reproducible TLDR. I mean, we do use third party audio models in the defined benchmark. So what happens is you actually need to measure appropriately all the interruptibility and responsiveness and whether, for instance, the audio matches what's being said.

49:01My favorite example is from the Hitchhiker's Guide to the Galaxy. I guess I'm showing my age here. There's the BBC series. and they have this notoriously cheerful computer on a spaceship. And there's a scene where the entire crew is about to die in two minutes because the spacecraft is going to crash into something. And the computer very cheerfully announces,

49:32Sam Charrington:you will crash in two minutes, we will all die. And of course, this is for comedic relief and spoiler alert, no, they don't die in two minutes, obviously. That's an EQ point that you were mentioning earlier. Exactly. So what I'm trying to say is that for voice and then also for video, you need to really care about how humans feel rather than just doing text only. And I mean, yeah, you want models that are not as dumb as a brick, But there is a difference between IQ and EQ and the latter is also important for humans. I was going to ask, there's quite a bit of research on EQ from a generic AI perspective, you know, probably text focused to, you know, but I'm imagining part of the kind of the through line in our conversation is that things are just different for text.

50:35Sam Charrington:the challenges tend to be more systematic as opposed to we're just going to solve this model and put a text front end there are a lot of moving pieces i'm imagining that same thing is true for eq solving eq from a text perspective doesn't necessarily get you you know eq voice ai it will get you to parts of it because so from a text transcript for instance for interruptability, you can figure some things out, right? But then in some cases, it's also a matter of, you know, what's a good point of reference. So let me give you examples of terrible points of reference. So you could think about, you know, maybe movies are a really great source of how humans should interact with each other.

51:29Well, take the romantic comedies and usually, you know, the obsessed guy who in the end gets the girl essentially stalks her, right? If you did that in reality, the police would show up and lock you up, right? likewise in some other movies people are quite liberal with their fists or kicks or whatever and in the end you know they will you know kiss and make up so to say or you know be best friends again again in reality if you do this you end up in jail so it's very clear that a lot of social behavior in movies isn't quite so realistic. I mean, there are more realistic things in terms of talk shows and so on, but even there, I mean, I sincerely hope that nobody will train a model on Dr.

52:27Phil and assume that this is normal human behavior. I mean, you know, it's entertaining TV, But to make it very clear that you do need to, you know, be more careful in terms of, you know, how interaction should really work. The good thing is that LLMs by now have a decent theory of the mind. So that gets you somewhere. And then, yeah, I mean, we're now entering, you know, the exciting new world where we can actually go and, you know, explore how humans really interact. We have to be very careful about it because, you know, ethics matter and you don't want to, you know, experiment on humans. But, you know, this is, I think, a really exciting time, at least not unless you do this appropriately.

53:23So I think it's a really exciting time of where this is going.

53:26Sam Charrington:And by that, you mean the technology is sufficiently far along enough that we can start putting in front of real people and engaging their reactions and that kind of thing? Is that where you are going? That is effectively, I think, what will happen where those models will get better by learning how to interact with humans. And yeah, I mean, this is an exciting new world. Yeah. And that makes me think of, you know, there's a lot of conversation right now about recursive self-improvement, right? And, you know, we, it's commonly discussed from the perspective of, you know, as the models that we use to build tools or to build things, you know, get better, you know, we can build more things, the models, you know, build better models, et cetera, et cetera.

54:15Sam Charrington:But is there an extent in which like the learning loop can become kind of continuous and the model can, you know, in a conversation, you know, learn what works for a particular person like that? I think that brings in a lot of things, this kind of self-improvement, like personalization. But it's something that I don't think we see very, you know, we don't see that in interactions today. This is starting, I would say, and you have a number of avenues. The closest thing I think we see is like memory in, you know, your traditional LLM. Okay, so let's actually go over a couple of pieces there. So the recursive self-improvement, it's cheap if you can do it in, you know, so to say fully on a computer by just interacting, let's say, with an LLM that has a decent theory of the mind.

55:19So for instance, if I ask an LLM, well, what happened? Yeah, so like, and they're a user simulator. So for instance, NVIDIA did a great job at releasing some digital personas and some scenarios and so on. And we use that. And we also wrote papers about it. So it's no secret. And what that does is it also allows us to build models that work well for humans that are not always cooperative, that are maybe a little bit abrasive, that are a little bit unpleasant to deal with, right? Or people who might want to break the system, right? So no human is perfect and people will try to have fun with those things.

56:07But, okay, so there's the, you know, given that those models by now have a decent idea of how humans work, you can use that for, you know, RSI to some extent. Another extent and direction, though, is then personalizing things. So, for instance, knowing that Alex likes, you know, equations and facts and technical details and maybe has a little bit of rough edges on the social side or whatever. However, you know, these are things that I can use for personalization such that, you know, next time I interact with Alex, I can probably, you know, personalize better for him. Right. Or, you know, knowing which devices I have, what my preferences are, what my language background is and all of that.

57:04this all can be used to improve direct personalization by just having models reason over it. In a similar way how, for instance, if you look at Hermes agent, which then looks at past interactions and past things that it's done and it reasons and it commits that into memory. So think of this as a glorified CRM, but now for everybody. But at the same time, you also want to learn across all the interactions. So, for instance, knowing that if I insult a user, the user will not take kindly to that. I mean, okay, sure, it's an egregious example. And of course, nobody would actually do that. But the point being, you know, learning such things across is something that you can then overall use for an improved model.

57:57So you basically get global improvements. So it feels very similar to recommender systems just now in 2026, where you have an overall behavioral improvement, but you also have personalized improvement. The latter improving as the model gets to know more about, you know, let's say Sam or about Alex. But then overall, the model getting better at dealing with those pesky humans or maybe dealing with those pesky humans in North America versus, you know, there's certain appropriate behavior in some parts of the world that's perfectly inappropriate elsewhere. so like eating with your left hand in India is seriously frowned upon because you use the right hand for eating and the left hand goes behind your back right so that or if somebody sees the bottom of your feet in Thailand it's also not a good thing but basically there's a lot of cultural weirdness and you know weirdness that really matters for the people that you know live in that culture and learning that will allow probably also to personalize a little bit better for you know specific regions specific cultures but then also for individuals within that culture and, you know, at the other end, you know, behavioral improvements for, you know, agents interacting with, you know, just those squishy humans.

59:38So that I think is a really exciting future. And by now we will be able to do this at a scale where we can actually build things rather than just positive theories, right? There's, you know, glorious work by, So for instance, if you look at Maya Mataric, so she builds those robots that interact with babies and get babies to move their legs and whatever. And there's a lot of engineering and science brilliance in building this stuff. Brilliance is good, but brilliance is not repeatable and automatable. what I think the future is going to be is going to be systems that automatically learn this stuff such that in the future you don't need quite as much brilliance anymore and you know you can just let data speak you know for movie recommendations this is by now a solved problem mostly but for overall human behavior interactions and improvement, I think even humans are not particularly good at it.

1:00:55If I was trying to aggravate somebody and to just harass them, a lot of people might at some point lose their cool and respond in kind, right? Because not all humans are nice. but for the purpose of you know achieving a given task for an agent you actually want the agent to keep his cool and to to handle this right and i mean this is one thing when you you know watch you know school teachers and how great they are with kids and to get them to behave even though those kids are very uncooperative maybe initially, right? There's a skill. Most humans don't have that. But, you know, we can learn those skills eventually with these organic systems.

1:01:54I think this is an exciting world.

1:01:56Sam Charrington:Well, there's a bit of a paradox in that if most humans don't have them, they're not in the data, and therefore we won't be able to easily train models to follow those patterns. Well, you get it from interactions, right? And basically, you can think of each interaction as another new experiment, a new data point. And so eventually, you know, by some randomness of exploration and so on, you'll be able to learn what works and what doesn't work.

1:02:35I mean, the question then is, you know, how you use that appropriately. But for instance, if you need to have a difficult conversation with somebody, I mean, the fun example is the millennial shit sandwich, right? where you have praise, critique, and praise. And by doing that, you know that you can deliver the critique without people getting too upset, right? And this is a strategy of communication, right? Or when you learn how to become a manager, you take all those training courses. How do you deal with people who are happy, who are unhappy? How do you, you know, basically deal with people in a way that, you know, they can be productive, that they can enjoy?

1:03:41I mean, there are techniques. And, you know, these can be learned. They can be learned from instructions. You can take Cialdini's books and they probably contain some good instructions. but we can also then eventually go and learn from data. This is, I think, really going to be quite an exciting revolution for the next maybe two to three years. I think it's going to move fast.

1:04:08Sam Charrington:Well, Alex, it's been wonderful to catch up with you and get into a little bit of what you're seeing and working on with regards to voice AI. Thanks. Thanks for having me. Awesome. Thank you.

1:04:24Thank you.