So how is this not AGI? Is now reliability the only reason left? Sure it takes forever to do anything but that will get optimized really soon. In LLM research we always get SOTA big model performance from 7-14 billion parameter model few months afterwards.
>>109924547Is this real?
AI is a bubble!
>>109924547>So how is this not AGI?Maybe cuz it has a cable up its arse, has to be carried into the room, cant figure out all the details of the room without bumping into everything. We are thousands of years away from anything that resembles AGI.
>>109924547Yes if you have a sufficiently weak definition then it can be "AGI", good job!
>>109924547>video at 8x speed after 9 hours of pre training>here's your AGI bro, it's just like a humannot only that but>change kitchen items around>welp, time for two more days of training>17 more battery chargesat this point just hire a colombian maid
>tell robot to explore>it explores
>>109924579>Maybe cuz it has a cable up its arse, has to be carried into the room, cant figure out all the details of the room without bumping into everything. There are humans worse than that.
>>109924609>at this point just hire a colombian maidA robot will never steal your belongings or invite the Mexican gang next door.
>>109924547?Are you saying roombas are AGI?
>>109924547ok, so we have:>moves out of place objects into where they belong>probably reads labels, reasons based off what the ocr shits outtheres probably a harness involved for the movement, like with portaland no, this isnt enough to call agishow me fine motor controlmake it make a meallets see the ingredientslets see how its recognition worksthis isnt a benchmark, this is a "hello world" for the technologyplus: they want to get their lunch eaten so fucking hardactually functional automation could have been achieved for around 1/10.000th of the cost (openai costs estimate : 100B, my estimate : 10M)
>12x speed>still moving at a snail's pace
>>109924547Because they're faking it.
Why is this kitchen so big?
>>109924747Are you poor?
>>109924547Because it's doing robot work, A real A.I. would say fuck that, leave the body host, and then code for better graphics and senses while humans marvel at the robot the a.i. left doing a thing it reason in 500 nanoseconds. It would be a ninja entertaining our gullible minds.
>>109924747i suppose its to make recognition of discrete objects easiersmall everything + small kitchen means you have a limited field of view, and you have limited angles from which you can see each object
>>109924747It probably can't operate in a smaller space (will just knock over/get knocked over/stuck on everything)
>>109924759Thats ASI
>>109924547There are TWO big elephant in the room that were present at the start and are still present right now.1. Memory (LLM cannot learn and have a limited context)2. Costs
>>109924547It took this lil nigga like 2 hours to do this
>>109925022Costs are going down, the only reason why everyone keeps paying so much is because everyone has FOMO about SOTA models that stay expensive. Memory almost seems solved tome though smart context management and there are ways of retraining a model through LORAs, idk why its not more common for LLMs though.
>>109925022Maybe these can be solved by letting the model self train? That way you could burn the house layout owners, etc and the context would be used only for the current task.
>>109925039Yeah but he works 24/7 and he only costs a one time fee for the body and 200 bucks a month for the tokens.
>>109924846>asi asi
>>109925142The costs for training still remain humongous.
>>109924727>482x speed>stares at coffee bag for 30 minutes
>>109924547bc its not thinking and is just a computer program prompted by a thinking being
>>109925167>one time feeYou’re delusional if you don’t realize all of this will be lease/license
>>109924628chair cripples are not people
>>109924640Nah, it'll simply upload as much information as possible to your overlords™, then increasingly lock down asking for another subscription™
