>current LLMs have about 5 to 10 trillion parameters>dogs have about 2.5-5 trillion synapses>chimpanzees 28 trillion>humans 150 trillionhonestly it's starting to get a bit freaky isn't it?
so freaky and icky not gonna lie
A dog has more bandwidth
>>109987170Physical neurons are not equal to parameters in terms of working principles and complexity.
>>109987170>>current LLMs have about 5 to 10 trillion parameters>>dogs have about 2.5-5 trillion synapsesa dog is more performant than an llm with much less errors and is continuously learningyour point that neurons == parameters is self defeating
>>109987170and how many bits is a parameter vs human synapse?
>>109987170I can't wait when in a couple of decades all your woofers become so riddled with genetic defects from all the inbreeding they all die out from being born sterile
>>109987230I said synapses not neurons, they're different things>>109987238I don't know, but honestly even if it's 2 or 3 orders of magnitude off we're getting uncomfortably close IMO
>>109987262>*suddenly wolfdog in your path*heh, nothing personnel, kid
but for example this guy https://forum.effectivealtruism.org/posts/h7Rj8Y8YWZccYMy5J/it-takes-5-layers-and-1000-artificial-neurons-to-simulate-a says about 1000 artificial neurons for a single biological neuron at 99% accuracy. In the context of a huge network and not a single neuron that could be way off, probably in the direction of fewer artificial neurons though I'd say
>>109987232>dog eats cat shit>not an error.my vet bill is the error...? :( my dog is fine
>>109987281Woofers are mental, they constantly go apeshit for no reason and decimate the appartment/house
>>109987267You're right, but synapses and neurons are closely related and both are incredibly dynamic and complex. Your fears are justified but I wouldn't equate parameters and synapses.
>>109987328actually- no it isntdogs eat shit when theres still too much protein in itusually they eat their own shit, and that happens when their diet is too richid have to check to be sure but iirc its completely normal for a dog's metabolism to not eat for a couple daysmodern care oftentimes overfeeds them, ancient instincts kick in, dog eats his own shit in fornt of your family and/or neighbors then goen on to lick little timmy's face lmaosome breeds are more susceptible to that than others iirchuskies tend to do that, wolfdogs also tend do that
>>109987351yeah, you need to spend their energy if you live in an apartmentpick up biking, take your woofer with you so he runs a littlebeing enclosed in an apartment is like solitary confinment for a dogespecially their psyche is based on the notion of packexclusion/isolation is kinda traumatic for a wooferits kinda understandable they break a fuse in such a situation
>>109987230Blasphemy! How dare you bring logic and knowledge to a discussion about AI! AI has no limitation and will continue improving forever and ever and ever!
>>109987354>You're right, but synapses and neurons are closely relatedwhat kind of bot answer is thisget outta here if you can't understand shit
>>109987281Wolfdogs are dysgenic
>>109987419your moms dog is dysgenicjust look at the poor fucking thing
>>109987170ai will serve dogs therefore dogs will alway be above ai... I mean super intelligence
>>109987388I understand, and I'm sorry if my answer appeared YOU KNOW WHAT FUCK YOU!!! FUCK THIS THREAD!!! DIE!!!!!!
If you consider how freakishly tiny features they draw onto silicon its not that special that when you make silicon take dozens of square cm's it starts to compare to a brains number of features
>>109987238a single ejaculate has 50 petabytes of genetic information
>>109987498fuck u too BIIITCH
Dumb thread. Parameters are not comparable to synapses and the idea that biological brains have too much while LLMs have too little is ass-backwards. LLMs are the ones who are extremely energy inefficient, for example.
It's getting a bit chill. Even 10T parameter models are almost prohibitively expensive to experiment with and to train, then you have to pay for inference with long chains of thought on top of that.Can't really go much smaller than FP4 parameter size in training. All the easy wins are already implemented.Do you expect a 80T parameter model in two more years?
>>109987767They can cram artificially generated data in and bloat it intentionally, I dont think they want to maxx out parameters to begin withThey want to improve the architecture with more efficient algorithms and maxing out each parameters usefulness.
>>109987170The human brain uses analog signals instead of digital ones, making it vastly more energy-efficient. Human brain cells can form far more interconnections, never stop training, and continually form or prune connections—a level of dynamic adaptability current AI cannot replicate.
>>109987767It economically doesn't make sense for them to scale more right now. The Mythos/Fable tier will remain the biggest in the world for either a few years or ever, depending on whether if this is a bubble or not.
>>109987170LLMs aren't floofy tail wagglers thoughbeit
>>109987170just 2 more trillion parameters and agi will be achieved in 2 weeks trust me
Dogs are disgusting and thankfully AI and robots will replace them
>>109987899the kids now just say "trust" in place of "trust me"
>>109988091indicative that they do not trust themselves; they ask you to simply trust in the information they are presenting itself, not their integrity, for they have none. Third worldification
>>1099871702 more years.
>>109988297haven't thought of it like that, that's a good point
>>109987170>inefficient parameters >ass pull parameters >basic obvious shit paramaters Yea man we are so close to making a supermind.
