I love that your brain runs on like 2 watts and a junior science scout chemistry set but it takes enough energy to run a city to make a language model spit out Reddit madlibs.
It's funny that none of the AI firms in the limelight are making any attempt to actually copy the architecture of biological neutral networks. Analog and neuromophic neural network hardware acceleration chips, ie hardware that mimics how biological brains compute, have both crazy energy efficiency and high performance. The only two firms that have actually been making a serious effort at developing such hardware are Intel and IBM. For both companies neuromophic chips have been r&d projects that have already had over a decade of resources poured into and will probably require at least another decade before producing something commercially viable.
It's also evidence that firms like OpenAI, Google, Microsoft, and all the AI startups that are spouting off about working on creating actual AI are full of shit. None of them are working on developing hardware to run real neural networks able to emulate the type of intelligence human brains are capable of, ie dynamic, general, and realtime learning.
It's funny that none of the AI firms in the limelight are making any attempt to actually copy the architecture of biological neutral networks
The term "neural network" is primarily marketing term, as is "AI" itself.
It's also evidence that firms like OpenAI, Google, Microsoft, and all the AI startups that are spouting off about working on creating actual AI are full of shit. None of them are working on developing hardware to run real neural networks able to emulate the type of intelligence human brains are capable of, ie dynamic, general, and realtime learning.
Judging by the creepy tech occultist fucks in charge of these corporations, it's probably better they don't.
The term "neural network" is primarily marketing term, as is "AI" itself.
I will say as a neuroscientist, it is a term we use, literally referring to the organization of neurons in the brain and how different areas connect to each other
I know neural networks are a thing and they do exist, including in synthetic forms, but my point is that when LLMs get marketing hyped enough they get conjured up attributes assigned to them and that's bullshit.
Doing a quick search to find a source, couldn't find just brain but I found numbers from as low as 12W to as high as 100W for the entire nervous system, which is still super efficient compared to chat bots!
When some "this is just like the blade runner treats" bazinga shows up and talks about how the treat printers are that close to replacing human brains, they always blissfully ignore the sheer material cost of these fucking things. They want a "happiness monster" to destroy the planet if it looks enough like the cyberpunkerinos.
Stuff like this makes me wonder of it wouldn't be more efficient to have humans doing the work of "AI." Instead of having some kooky software, use a person to do the same thing with 2 watts of electricity.
Amazon's Mechanical Turk program is/was basically this. Farming out small editing or image recognition tasks to people in the global south who may or may not actually get paid a pittance for it.
man i remember one time I basically got paid $20/hr to, based on a series of screenshots, make a title for a porn video. My god it was easy money, but the tasks ran out and never came back 😔
I love that your brain runs on like 2 watts and a junior science scout chemistry set but it takes enough energy to run a city to make a language model spit out Reddit madlibs.
My brain runs on beer and sandwiches. Look at what tech bros need to mimic a fraction of my power
Unironically this. This needs to be brought up more. Why must the planet burn down that much faster for this fucking bullshit?
It's funny that none of the AI firms in the limelight are making any attempt to actually copy the architecture of biological neutral networks. Analog and neuromophic neural network hardware acceleration chips, ie hardware that mimics how biological brains compute, have both crazy energy efficiency and high performance. The only two firms that have actually been making a serious effort at developing such hardware are Intel and IBM. For both companies neuromophic chips have been r&d projects that have already had over a decade of resources poured into and will probably require at least another decade before producing something commercially viable.
It's also evidence that firms like OpenAI, Google, Microsoft, and all the AI startups that are spouting off about working on creating actual AI are full of shit. None of them are working on developing hardware to run real neural networks able to emulate the type of intelligence human brains are capable of, ie dynamic, general, and realtime learning.
The term "neural network" is primarily marketing term, as is "AI" itself.
Judging by the creepy tech occultist fucks in charge of these corporations, it's probably better they don't.
https://futurism.com/openai-employees-say-firms-chief-scientist-has-been-making-strange-spiritual-claims
I will say as a neuroscientist, it is a term we use, literally referring to the organization of neurons in the brain and how different areas connect to each other
I know neural networks are a thing and they do exist, including in synthetic forms, but my point is that when LLMs get marketing hyped enough they get conjured up attributes assigned to them and that's bullshit.
Oh 100% agree, and personally it’s really annoying they took our term that I have to use for real things
Oho someone else in the medical/medical science field
deleted by creator
They're gunna' start harvesting brains and using them as computers if you bring this up to them.
Already happening...
https://www.technologyreview.com/2023/12/11/1084926/human-brain-cells-chip-organoid-speech-recognition/
https://newatlas.com/computers/human-brain-chip-ai/
Comrades are eventually gonna have to do each other a solid and shoot each other in the head after they die like we're in a fucking zombie movie.
Doing a quick search to find a source, couldn't find just brain but I found numbers from as low as 12W to as high as 100W for the entire nervous system, which is still super efficient compared to chat bots!
clearly we need to develop mitochondria for machines (mitochondria are the real hero behind eukaryotic efficiency)
When some "this is just like the blade runner treats" bazinga shows up and talks about how the treat printers are that close to replacing human brains, they always blissfully ignore the sheer material cost of these fucking things. They want a "happiness monster" to destroy the planet if it looks enough like the cyberpunkerinos.
So you're saying we should harvest brains?
Stuff like this makes me wonder of it wouldn't be more efficient to have humans doing the work of "AI." Instead of having some kooky software, use a person to do the same thing with 2 watts of electricity.
Amazon's Mechanical Turk program is/was basically this. Farming out small editing or image recognition tasks to people in the global south who may or may not actually get paid a pittance for it.
man i remember one time I basically got paid $20/hr to, based on a series of screenshots, make a title for a porn video. My god it was easy money, but the tasks ran out and never came back 😔
Sounds more like it was hard earned money to me.