The Ledger
Every statement that passed quotation and attribution checks. Mix any filter with any other: certainty 1/5, debate potential 5/5, or both at once.
why aren't all 56 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 2 checkable ones are still open, waiting for their date. predictions held up or didn't; assertions are supported or contradicted. on every card: ▮▮▮▮▮ certainty · ▮▮▮▮▮ debate potential. speakers are clickable
Kumar: Voice AI models must decide every 100 milliseconds for natural interaction
“You need to make decisions at the hundred millisecond, let's say, Time segment so that if you're talking and the other person, you know, starts sort of making some noises that make it seem like they're trying to interrupt you or they want to say something, you…”
Kumar: Near-term AI voice models will still lack dynamic conversational fluidity
“And the models that we have today, like CSM, for example, And probably some of the models that we'll have in the short term that will make the experience better will still not be modeling the conversational dynamics because they're making decisions that kind o…”
Kumar: Theoretical gains won't unseat transformers without matching years of optimization
“But because of all the engineering work that the community has done around Transformers, it's like, you know, it's very good. And you're not going to just sort of unseat that, you know, just by an idea, right? There's a lot of work to be done.”
Kumar: Competitors will match Sesame's voice quality; there is no secret sauce
“The other companies, the other sort of chat products and so forth, they will get better voices. They're all, like, it's not gonna, we don't have some magical secret sauce on the technical side that is gonna be, like, impossible to replicate. They're gonna get …”
Kumar: AI voice interface success depends on product experience over model size
“We think that that interface layer, it's really not kind of a core bigger, bigger models, better, better reasoning question. It's really a product experience question. It's really a question of, can you make a system that people actually want to interact with,…”
Kumar: Transcription-free conversational AI models are coming soon
“A pretty clear path that a lot of, I think, labs are taking, and we're taking as well, and will be in kind of future versions, is just kind of transcription free. Just go straight into the text component which will kind of obviate transcription entirely. That …”
Kumar: Top AI labs under-invest in creative taste and humanities
“I do think that there is kind of an under-investment or an under-focus in the sort of Strong AI team world on product experience and sort of creative taste and kind of humanities maybe, in a sense, to kind of bring AI to experiences that kind of everyday peopl…”
Kumar: Open source won't solve AI voice and personality features
“But other parts in particular, kind of some of the personality aspects, some of the voice aspects, the speech generation, we didn't think and we still don't think will just be kind of done by the community. We think we will need to do it because that's kind of…”
Kumar: Sesame is actively developing smart glasses for its AI companions
“We mentioned on the website and we mentioned some of our launch content that we are working towards glasses as a form factor for Companions, or kind of this companion interface”
Kumar: Smartphones and laptops will not be replaced anytime soon
“No one's going to replace phones anytime soon, or laptops for that matter.”
Kumar: Voice cloning and prompting alone cannot create great AI personalities
“It takes more than just sort of, you know, voice clone plus change the prompt. Now you have a new character that's just as good as it would be if you spent a lot of time on it. It takes, I think today making a great personality voice interface system, we can't…”
Kumar: Sesame will build a unified audio-text transformer within months
“The path that we're going to take, I think, over the next few months is making a single transformer that does both audio understanding, content, text content generation, and speech generation.”
Kumar: Multi-step AI agents need 99% reliability for daily adoption
“Doing, especially challenging, kind of multi-step things, you know, agents, as people say, I think to make that part of your everyday habits, it has to be like, 99%, you know, and right now, you know, every extra step the thing needs to take, There's some perc…”
Kumar: Not enough AI teams are focused on the interface layer
“But I don't think there's enough companies and kind of teams working on the interface layer.”
Kumar: Builders underestimate current releases due to internal roadmap gaps
“When you build the thing, right, when you're building the product and using it every day, you know, there are some things that you work on that don't get into the demo because they're going to take longer and you want to ship the demo. You kind of know how big…”
Kumar: AI product optimization depends on hard-to-quantify qualitative user reactions
“But really, I think with some of these more product experience questions, there's something qualitative about it that is very hard to quantify. That is one of the big challenges internally, actually, is how do you hill climb effectively on what is really an ML…”
Kumar: Internal ML testing fails when teams exhaust fresh user reactions
“Misleading at times because you tried so much and you don't have, at least when we're trying it internally, you don't have such a diversity of users that you get kind of the first reaction over and over, right? You only get so many first reactions. And then wh…”
Kumar: Speech-to-text transcription misses essential non-verbal audio cues
“Humans, of course, convey a lot of information through Their speech that is not the words, the content of the speech and transcription misses that entirely.”
Kumar: Next Sesame AI models will feed audio natively into LLMs
“And so the kind of next versions of our models that will take audio and natively into the kind of LLM component will hopefully more and more pick up on those things.”
Kumar: Sesame trades complex AI reasoning for natural voice interaction
“So, you know, if you talk to Maya and Miles, you probably will not be able to get the same quality of like reasoning capabilities or intelligence as other As other systems, but in return, you're kind of getting this much more natural fluid interaction.”
Kumar: Storytelling and AI will merge into new AI-native media categories
“And I think that we will see a lot of not just Sesame, but other kinds of media, let's say, that are sort of AI native in a way that Bring some creativity, bring some like storytelling into AI, or maybe bring AI into those categories. And I think they'll make …”
Kumar: Sesame built its open-source voice models from scratch
“Like we had to build the models that we're going to open source from scratch in order to get them to a point where they can achieve this experience.”
Kumar: Good ML taste means avoiding what APIs will soon commoditize
“I think from my perspective, good taste in ML today, because it's such a fast moving field with so many people working across, you know, open source and APIs and big labs and so forth. Really, you're trying to identify What part of the ecosystem or what part o…”
Kumar: Sesame is not building an API or developer-facing product
“We are not a developer facing business. We're not making an API.”
Kumar: Sesame is open sourcing its speech model, not the full demo
“We're not open sourcing the demo. We're open sourcing the speech generation model that is powering the voice of the demo.”
Kumar: Sesame achieves voice cloning via in-context learning prompt strings
“The model is this kind of, you know, it has kind of this in context learning style voice cloning. I mean, typically with some other kind of text-to-speech models, the voice cloning is kind of like an explicit feature. So it's sort of the model has dedicated ki…”
Kumar: No other open-source model generates multi-participant contextual audio
“At least to our knowledge, there's not another model out there that, that is open source that kind of is a sort of contextual thing where you kind of can put two participants in a conversation, even more, three, and generate kind of a conversation between them…”
Kumar: Traditional text-to-speech sounds flat because non-neutral tones risk sounding inappropriate
“And that's probably why, or it's one of the reasons why historically voice assistants feel so flat is that traditional text of speech, it's kind of like it can only be flat. Or in other words, if it tries to not be flat, it's very likely wrong.”
Kumar: Speech research community will shift toward contextual AI architectures
“So, so the speech generation research community is very likely, I think, to move to more and more contextual architectures basically.”
Kumar: Seemingly easy side products like APIs create massive engineering drag
“Sometimes it feels like an API or something like that is like relatively easy to do. And, you know, it's not like, it's not maybe as hard as some of the other things that we're doing, but everything is a drag on engineering, right?”
Kumar: Sesame will not build a one-size-fits-all AI companion
“So we're certainly not going to, we don't see our product as like one companion that's the same for everyone. People have different preferences and that has to be a part of this kind of product category for sure.”
Kumar: AI conversation is a distinct modality requiring core research
“I think that conversation, like human conversation, is kind of its own modality. And it is nowhere near done, right? There's so much more to do in the core research side to make it better.”
Kumar: True AI naturalness requires modeling turn-taking and backchannels
“I think to get these things to feel very, very natural and real, you do need to model the full conversation, the turn taking, the back channels, everything.”
Kumar: Early AI startups need flexible systems thinkers over niche specialists
“Especially when you're smaller, you know, you don't really want to harden, like, you know, you have this team that is super, super niche and doing only this thing, because you don't know exactly what the stack is going to look like tomorrow. Things change on t…”
Kumar: Word error rate metrics for AI speech generation are now saturated
“Earlier on in the speech generation world in the community, very often you'd look at like word error rate where you look at transcription, like you kind of have a sentence and you generate and you transcribe it and you see if it's the same. And those metrics a…”
Kumar: Achieving human realism in voice AI is harder than text
“I think you, I think it's much easier. It would be much easier to make a system that produces text chats with you that feels like you're texting a human because there's such a compression of like what the entity on the other side is into just like text. Wherea…”
Kumar: Adding generative modalities to AI models is harder than understanding
“It's much harder to add a modality to a pre-trained model than it is, add a generative modality, than it is to add an understanding modality.”
Kumar: Conversational AI will eventually rely on single models over heuristic pipelines
“I don't think you want to, in the long term, have those dynamics be like heuristics and so on, which they kind of are now. There are models involved in some heuristics and so forth. I think in the long term, it's just one model that is kind of naturally employ…”
Kumar: Sesame is developing diffusion-based audio generation models
“We are also working, by the way, on kind of ideas that make the audio generation part diffusion.”
Kumar: Transformers will remain the dominant AI sequence architecture short-term
“And I wouldn't bet against transformers, you know, not in the short term anyways.”
Kumar: Sesame intentionally includes speech imperfections to make AI sound natural
“My and Miles, they might sort of say the wrong thing or kind of like back up a little bit and say something else or something. And that's on purpose, of course.”
Kumar: Sesame will preserve AI companion personality as models improve
“They're making assistance. They're making utilities. I love those products. I use them all the time. They're great products. We want to make a companion. And so our prioritization of features and of, let's say, post training kind of personality, et cetera, wil…”
Kumar: Big tech companies will attempt to own conversational AI interface layer
“I think that over time, I think we will see more of these, you know, bigger companies trying to operate this layer. Like I said, I think that there is not enough effort on that right now, making these systems delightful to interact with, you know, and I think …”
Kumar: The ChatGPT plugin system I built failed to fully take off
“I did this chat to be plugin system before, and I think it's still there probably. And it kind of didn't fully take off really.”
Kumar: Developer plugins will be essential to future AI interfaces
“I think that the models still need to get better basically to utilize plugins essentially in a way that's kind of reliable enough that someone will go out and look for a plugin for, you know, their kind of downstream service of choice because they just want, y…”
Kumar: Sesame's current demo cannot detect user emotional tone
“The current demo does not sort of hear the user from the perspective of their paralinguistic kind of emotional tone and so forth.”