Just to be clear, while I spent time at the same research institute, I am not the author. The author is Yiannis Aloimonos @UMD: http://users.umiacs.umd.edu/~yiannis/
On Tue, Aug 31, 2021 at 3:47 AM <[email protected]> wrote: > Yes good read, I think many of us already know all this though. BTW you > made it into a blog post but really you only said a few things in there > that was your point.* Only saying this because* the paper I read called > foundation models was way too big, yes big is good but first of all it uses > complex formal language which does not make life simple like markov chains, > and second it is unfocused and can be summarized 100 times smaller. > > Yes the multi-model GPT ex. DALL-E is a big improvement, using vision and > text is a big jump up, we wouldn't need a multi-model if it wasn't so > useful! It's like training on a dataset about dogs....just dogs....then > multi-modal is like dogs+cats, or dogs+clothing....so it gives it a wider > spectrum on data collection....similarly a vison that sees beyond our color > spectrum will give it "more data" using the "same amount" of train data as > well. > > Yes, adding motor ability to our world will be another giant "multi-model" > ability to add to its mix in neuronal mixing. It decides where to collect > data from in the real world, or the internet, then it learns, and decides > again where to search now, repeat, it is all about data collection to > decide where to collect data. Adding motor ability doesn't need a body > exactly, it can use Google to search specifically on the internet what to > read, or what thoughts to think about and what to store it thinks up. For > motor ability to figure out how far a face is by using your hand/etc > instead of a distance sensor, yes, this collects you data. You can also use > the size of the face etc to determine how far away it is. Deciding what > data to eat/make on its own is a big step up from being fed image and text > data from us, it allows it to have agency, to learn, instead of us making > it learn. Using a body to do this is a great way but thinking / using Goole > / using a machine using its thoughts is also great ways that fill that need > of a humanoid robot. > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/T9f314873b2a6ed76-M52ec649122c966e811e5f43f> > -- Daniel Jue Cognami LLC 240-515-7802 www.cognami.ai ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T9f314873b2a6ed76-M1b50b25134bf29e8c84f82f9 Delivery options: https://agi.topicbox.com/groups/agi/subscription
