I've spent the last half hour trying to figure out what this company is, and I still have no idea. My first stop was their website and I could not find a single sentence providing concrete details regarding their value proposition. It's just a bunch of vacuous statements which paint themselves as some sort of iconoclastic visionaries. Please do check it out for yourself and tell me it's not just a bunch of nonsense.…
They're the ones that released Stable Diffusion (the ML model that can generate images from text prompts).
The investors are buying into a toolset for what will potentially become open source AGI that is already blowing the competition out of the water in terms of capabilities, speed of development, and access/reach. Add together large language models with text and code interpretation/generation + diffusion + vision + speech and sound + embodiment = what might you get? It's unlikely investors care more than an inkling abo…
Leading to AGI? I have been using deep learning professionally for about seven years (and neural networks since the 1980s with a commercial product and I wrote the model for SAIC’s bomb detector they did for the FAA). Personally, as awesomely useful as deep learning is, I don’t see the path to AGI. I think AGI will be a hybrid of DL, GOFAI, and things not yet invented.
Sure but the comment is still an appealing pitch to investors, and probably can be read broadly to include "things not yet invented".
According to the article, they plan to train private models. They claim to have done R&D that makes them more efficient at training models.
Does that sound like a $1 billion business to you? Even at 6 figure profit per model, they are going to need to build up a huge sales arm to do enough business for that valuation. I think they are expecting to become some sort of artist's tool, but since the model is open source and artist's tend to have good GPUs, Adobe can crush them by putting their own model (or a similar proprietary model from Adobe) into photos…
It does not imo but also seems like a cheap way to back a possibly asymmetric group. The valuation is not the hurdle for investors, as they will have seniority (likely with a multiplier) in any liquidity event.
Does it seem like this group is moving quickly and making waves in a potentially huge market? Yes. Is this still a lot of money and very risky? Also yes.
Leading to AGI? I have been using deep learning professionally for about seven years (and neural networks since the 1980s with a commercial product and I wrote the model for SAIC’s bomb detector they did for the FAA). Personally, as awesomely useful as deep learning is, I don’t see the path to AGI. I think AGI will be a hybrid of DL, GOFAI, and things not yet invented.
Sure but the comment is still an appealing pitch to investors, and probably can be read broadly to include "things not yet invented". Out of curiosity - what is GOFAI?
Good Old Fashioned (symbolic) Artificial Intelligence. And thanks, absolutely my clumsily worded investor pitch intent was to cover techniques not yet invented, thanks for picking up on that.
The investors are buying into a toolset for what will potentially become open source AGI that is already blowing the competition out of the water in terms of capabilities, speed of development, and access/reach. Add together large language models with text and code interpretation/generation + diffusion + vision + speech and sound + embodiment = what might you get? It's unlikely investors care more than an inkling abo…
Leading to AGI? I have been using deep learning professionally for about seven years (and neural networks since the 1980s with a commercial product and I wrote the model for SAIC’s bomb detector they did for the FAA). Personally, as awesomely useful as deep learning is, I don’t see the path to AGI. I think AGI will be a hybrid of DL, GOFAI, and things not yet invented.
Couldn't agree more - generalised task capable systems will likely be composable architectures that draw upon a grounding in variety of methods, not just DL by any means. Synthetic sensoria capable of integrating, interpreting and generating general task capabilities are likely to lead us somewhere interesting. DL / LLMs are akin to Wintermutes before waking; with whatever passes for consciousness instantiating and evaporating with each prompt input (which some might argue is a good thing for foom-resistant biospheres) and no continuum vector state pseudoawareness. However, strongly feel we'll have instruct-GPT models on the edge as an interface modality, but they're not the holy grail or be all end all.
I've spent the last half hour trying to figure out what this company is, and I still have no idea. My first stop was their website and I could not find a single sentence providing concrete details regarding their value proposition. It's just a bunch of vacuous statements which paint themselves as some sort of iconoclastic visionaries. Please do check it out for yourself and tell me it's not just a bunch of nonsense.…
> I've spent the last half hour trying to figure out what this company is, and I still have no idea
That's not unusual.
As long as the investors understand it, that's fine.
Even though the investors have no clue what it does, it's still fine as long as the investors are willing to write the cheque.