CKX_Crypto

Hmmm.... Not having played with the o3 model myself yet, I don't want to say too much that's super definitive... My experience with previous models is, at first when I try them I'm like "Whoa! Incredible! This is super smart! Look what it can do!" ... then after trying to use it in practice I end up being like "Hmm well actually in some ways it's weirdly stupid... but once I learn the right way to use it's still super useful..." .... I am guessing o3 will be the same way... but maybe it really does solve significant aspects of coding more compellingly. I certainly hope so! However, that is still a quite different thing from AGI. Sam Altman says with models like this, the concept of AGI becomes less useful, but I don't actually think so. I think the concept of AGI becomes less useful for OpenAI, because they seem to be launching amazing and valuable stuff that is not AGI. I also think it's clear that tremendous economic and practical value can be gotten from these super large scale narrow-AIs (whose narrowness is a sort of halo around their incredibly huge training dataset), regardless of the fact that they're not AGI.... The theory of AGI is subtle and the concept of human-level AGI is ambiguous ... but: Basically until we have an AGI that can invent and create and build and communicate autonomously and creatively, based on its own values and inclinations, at the level of a smart innovative human ... we are not to HLAGI yet. We may well be almost there in a historical sense, could be just a couple years off, but that's still different than having it in hand... and still leaves lots of room for innovation in getting to the finish lin Also, it's important to point out that, even though I don't think LLMs are the path to AGI, I do think o3 and any further developments in that direction are very encouraging regarding my own team's quite different AGI efforts... Consider: o3 is probably clever enough to reliably map English into uncertain predicate logic expressions. And while it may be ahead of OSS models now, that won't last, the rhythm of the LLM world is that the OSS models don't lag the proprietary models by thaaaat long.... So before too long we'll have an OSS model that can reliably map English into uncertain predicate logic expressions... Then we can take whatever the best LLMs output (in response to any query or situation) and feed it into the scaled-up version of our Hyperon Atomspace (running spread across our decentralized networks...). We can use Hyperon's logical reasoning to go beyond what LLMs can do in terms of abstraction and generalization ... and we can use Hyperon's evolutionary and algorithmic-chemistry methods to do creative invention beyond what LLMs can do... fueled by the input from LLMs themselves, and using LLMs for guidance of the Hyperon method wherever helpful... We will be using the f**k out of the best OSS LLMs, but not letting them run the show, rather using them as a resource within a decentralized network leveraging Hyperon for advanced reasoning and creativity, and also for running the basic cognitive cycle of values, goals/motivations, perceptions, plans and actions... Hyperon doesn't need LLMs to do its thing. But it can certainly use them as accelerants ;) These next couple years are going to be insanely fascinating... What a time to be alive and working in the AGI space !! Wow !!