Raw LLM Responses

Open source finally catching up to the voice to voice models. I've mostly only been seeing TTS and STT. That being said the internal model for Sesame is quite small, so it's nowhere near "intelligent" as GPT 4o, nor is it actually fully multi modal. Now consider the fact that OpenAI had GPT 4o internally around a year ago. The completely uncensored version. We know how good it is based on their demos (while we got a heavily nerfed version). Given the "mind blowing" or "THE moment" reactions to Sesame, what do you think the OpenAI researchers' / testers' first impression of 4o was? Then consider that internally they probably have a fully multi modal version of GPT 4.5. Very quickly you can piece together what those OpenAI employees meant when they were "feeling the AGI" given they had access to a FAR more intelligent version than Sesame's Maya a year ago. This is also why I think regardless of when these companies achieve AGI, we the public will not know about it until a year later. If we get access to AGI in 2029, they probably developed it in 2028, and likely a much more powerful version than the one that we get. Anything "mindblowing" that we see now, those internal have already seen a year ago. There is quite a big disconnect between those privy to that and us the general public. Yes a lot of tweets are just hype and some are possibly even fake. However, a lot also probably SEEM like hype because they are so different compared to the models that WE have access to that we cannot wrap our heads around it.