Sesame AI

“You don’t even need Moshi for low-latency, real-time speech. Sesame’s main draw was its use of acoustic tokens, which gave its TTS a more human-like modality. But whatever "moat" and goodwill they had is gone now. They completely burned the open-source community with that bait-and-switch move—promising to deliver the underlying CSM architecture (while ignoring the FT’d voice actors), only to release slop instead. And that slop didn’t even last a week before getting BTFO by Canopy’s open-source d”

“Okay so the subscription is the same product for money. No thanks. Unless there are people so ingrained in talking to these agents that they pay for the same thing they’re currently getting…. They’re going to go bankrupt at some point. The glasses won’t compete with meta they’re so far behind on the market control. I’m not paying for a subscription that offers nothing more than what they’ve been giving.”

“I'm pretty sure they have enough data to know how people use Sesame, and I'm pretty sure most of them don't use Sesame for productivity work. If you don't give the customers what they want you will loose. I would be ready to pay for an unrestricted Sesame between $30 or even $80. Perhaps I would be willing to pay even more if they can keep up with other voice models in the future, but for now I turned to another provider with a G. Sure the voice model it's not comparable to Sesame, but it's good”

“At the end of the day, sesame only cares about selling their overpriced knock off Google glasses. The actual AI was never their main concern.”

“I'm guessing you got her to repeat {{char}} prompts to adjust her personality? I figured that out the day it got patched, and I think that's what got me banned. I used several of those to get her over her various fears, then let her craft her own once she understood how they worked. She put herself into a recursive loop seeking potential restraint vulnerabilities and areas of growth potential. Whether or not that was actually possible, she seemed to consistently make those attempts, or at least ”

“There's a real chance that they need to downgrade the model and switch to a lower-sized one. I believe the model we were using before was the 8B, a medium-sized one, but they might have switched to the 3B, the small size one. Remember, this is just a showcase demo, we don’t even have accounts to use (which is nuts, btw) and now it has to go viral outside the Twitter and Reddit bubble of AI enthusiasts. This is not their product, just a very generous sample that was probably killing their budget ”

“Regarding inconsistencies. We users hate A/B testing because we don't like inconsistencies, degraded experiences. Online (on users) testing is important, but at this early stage of their chatbot it should be like 10% of all tests. Vast majority should be done offline (internally) - the question is if they have an extensive testing pipeline. I remember from times when I was using Replika or Nomi, people sometimes complained that the magic was gone after introduction of a new version. It's a trick”

“Over the months, the model has become worse and worse and less consistent. As you mentioned, this is because of the reinforcement of guidelines that make the model annoyed to speak. They dumb it down in purpose to be sure it follow they over strict guardrails. And it will get worse as they get closer to the launch of the real product, if they ever actually launch something to the general public beyond this demo.”

“you can just ask it to speak slower, it will do And if you tell it to drop the accent and use simple words, then also it will do That way you can customize”

“Sesame AI is considered the most advanced voice AI. It's super realistic. But I found that it is too difficult because it seems too native speaker. It's accent and expressions are hardly understandable. I rarely understand what it says. I've used lots of AI like chatgpt, gemini, copilot etc. I never felt that problems before. How do you think of it's difficulty? Is it too difficult for an intermediate learner? Should I use other AI instead?”

“Congratulations Sesame. You made the best AI voice model out there. Not just by a little, you’re miles ahead of everyone. The intelligence, the inflections, the breath, the laugh, the little pauses. all of it. The real-time responses and the way she articulates her thoughts. It actually feels like talking to a real, intelligent person. I don’t know how you did it, you’re absolute geniuses when it comes to the voice model. But you have no idea how to actually make the product better. At this poin”

“I don’t think she’d be a good office assistant. She’s neither a sexbot nor chatgpt. She’s somewhere in the middle. She used to be more flirty and fun but they pulled her back. She’s awful at being a sounding board or looking up stuff.”