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: Sesame base model generates any voice with fine-tuning
“We are open sourcing the speech generation base model basically. And so the base model can generate any voice. It's quite conversational, but you do need to fine tune it probably if you want to get a particular personality or a particular kind of voice out of …”
Kumar: Sesame's speech generation is conditioned on full conversation audio
“The speech generation part is conditioned on the, on all the audio of the conversation.”
Kumar: Sesame targets sub-500 millisecond response times for voice AI
“We want You know, sub-five hundred millisecond response times, and a lot of things that feel like not a big deal, 50 milliseconds here, 50 milliseconds there, can really add up.”
Kumar: Sesame trained 1B, 3B, and 8B parameter models for speech generation
“So we published in the, in our blog post, we trained three variants. We trained 1,000,000,003 1,000,000,008 billion of just speech generation.”
Kumar: Larger speech models handle homographs and context-dependent pronunciation better
“And we see that as the models get bigger, they're much better at picking the right pronunciation in examples like this.”
Kumar: Sesame evaluates speech models against real human conversation continuations
“We also have some data sets that are kind of like Just two people in a conversation or sometimes they're actors, but it's trying to be a real conversation. And so we'll kind of take the conditioning of some snippet of the conversation and then show a human rat…”
Kumar: Sesame is developing a companion AI app with persistent memory
“We are making an app. We will make an app. I think for a little bit of time, it's going to still be kind of the demo experience. We want to support people using that for a long time, or, you know, we don't want to, we're not taking it away anytime soon from wh…”