It is a curious and odd coincidence that the Yin and Yang mirrors the duality that C.S.Peirce identified in semiotics as icons and symbols.
Even more surprising (and useful) is that you can actually explore this duality by playing around with diffusion models (i.e., #stablediffusion, #dalle, #midjourney). This is because these models are both symbolic as well as iconic in nature. medium.com/intuitionmachi…
It's also intriguing that existing human civilization is structured primarily around the use of symbols. An insightful ancient historical accounting of this can be found in the book "Alphabet versus the Goddess". amazon.com/Alphabet-Versu…
In my proposed model of human cognition, I recognize the importance of these opposites to construct an architecture that involves the conversation of four specialized components. medium.com/intuitionmachi…
The derivation of this architecture originates from C.S. Peirce's formulation of knowledge discovery. medium.com/intuitionmachi…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
It's a fascinating coincidence that large language models have to solve the same problem as human cognition. More specifically, I am referring to the information bottleneck that's often described as "now or never" cognition. medium.com/intuitionmachi…
GPT-3 is limited to a window of 4,000 tokens to drive its subsequent output. A lot of engineering that goes into something like ChatGPT3 is in figuring out how you can stuff relevant information for the next step in the response.
The psychological explanation for humans is that we perform chunking of information. So when we remember a phone number, we chunk it into two pairs, remembering two "tokens" instead of 7 tokens.
It seems to perplex many that if you already know the solution, why can't you just program it into an AI? Why is it that current AI (i.e., Deep Learning) requires one to grow a solution and not program a solution?
At a conceptual level, current AI is simply not understood by the masses because their models of human cognition are based on a mechanical model that gets programmed (i.e., education is the distillation of facts) and not on an organic model that learns from engagement.
This is perhaps why society is completely unprepared for incoming AI technology. Deep Learning systems are intuition machines. However, centuries of substance philosophy have indoctrinated us to believe that we are logical machines. amazon.com/Artificial-Int…
AI Business Strategy Tip: Just use the most capable language model that you can get your hands on and scale your business before you go broke. Language models are just getting better as they get smaller.
If you optimize too early because of the cost, you increase the risk of an inferior product and hence slow down your customer growth. It may be a wiser strategy to take the financial hit early rather than wait for technology to become economically feasible.
Organic growth is rarely in the cards for AI startups. You have to prepare to lose money upfront and instead, anticipate were the AI puck is moving.
Tinkering with how to use both #midjourney and #stablediffusion in the same workflow. Here's a cyborg I created by first importing a SD generated image in MJ. I then took the MJ image, ran img2img with SD. The result is more lifelike images but preserving MJ details.
SD is nowhere near MJ in creating intricate details. The next step is to use Dall-E to outpaint.
Each tool has its strengths. I doubt you could have achieved the same results with a single tool. In the grand scheme of things, a strong workflow will use many different image generators to achieve a desired result. Become familiar with all tools to up your game!
Here are five recommend books to read that will give you an entirely new perspective on why civilization must change so that AGI can achieve human alignment.