Where is the Machine Learning component where GPT3 generates a semi-random prompt with (3) unique responses using different “settings” for each response and then providing the three completions to a random human and asking which they think is the best response, and possibly a generic follow-up question soliciting the reason for their selection with possible options being; “Best option”, “Least bad option” and “Hard to choose; multiple good answers”. Honestly, with Elon’s social media influence, I can’t fathom that it would be very hard to entice people to get on their phone’s to play some free games to help build out a new level of GPT3 intelligence; built on crowdsourced human opinion.
I feel that with these metrics, OpenAI should be able to create a secondary Logic Consideration Process that identifies correlations between question types and optimal settings for those types of questions. This kind of multi-layering of deep complexity from different sources is the path I expect to bare the most useful fruits in the pursuit of AGI, and I suspect will also provide the safest path forward due to the inherit bottleneck created by logically connecting two separate systems that can be independently monitored, instructed or even deactivated.
Another crowd-sourced machine-learning game I would love to see would be one based on DALL-E; where once again it generates a semi-random image prompt, but then chooses one of whatever levers are available to it when generating images and creates three images with one using the typical value for that lever; and then two others with offset values that can be used against other datasets to determine which levers have the greatest impact on the results depending on the style and subject matter being requested. The user would then pick the picture they liked the best and answer a follow-up question about why they chose their selection with answers like; “It speaks to my aesthetic”, “It seems hands-down the best” and “Hard to choose; multiple good images”. I don’t need to explain to the AI people how they could use this info, but it seems to me that it could help with emotional categorization (another feature that could be built in) as well as identifying optimal settings to use for different artforms and subject matter.
Anyways, anyone who has read this gets what I’m suggesting. I’m an OpenAI fan so I’d like to be whittling away my hours doing image and text recognition/categorization for OpenAI rather than Google. That is, unless the Google data is open-source and playing on the Crowdsource app helps all AI developers, but somehow an app identified as produced by Google LLC in the app store that is directly related to improving their AI products, doesn’t fill me with the butterflies & rainbows of hope & joy that run of the mill Google products typically do.
So yeah, turn me into a chimp; I’ll play crowdsourced games to earn points that have no value other than recognition of contribution, that’s enough for me.
PS: UBI with robotic workforce isn’t a pipedream unless the majority believe it is a pipe dream. We are part of the generations working towards giving our children’s children a better life than we could have ever envisioned; let’s not blow up the world or the economy on our way there.