Counter argument. How did neural networks evolve at all on Earth without the fine structure constant (alpha)?
For AGI, thinking a biotech jumpstart would do the physics trick, it won't. It's merely a desperate hack, most inelegant and riddled with single points of failure. Essentially, a serial string of fairy lights. Use AI to jumpstart synthetically-real alpha. There's your quantum appdapter. On Fri, Mar 29, 2024, 00:45 Matt Mahoney <mattmahone...@gmail.com> wrote: > On Thu, Mar 28, 2024, 2:34 PM Quan Tesla <quantes...@gmail.com> wrote: > >> Would you like a sensible response? What's your position on the >> probability of AGI without the fine structure constant? >> > > If the fine structure constant were much different than 1/137.0359992 then > the binding energy between atoms relative to their size would not allow the > right chemistry for intelligent life to evolve. Likewise for the other 25 > or so free parameters of the standard model and general relativity or > whatever undiscovered theory encompasses both. The anthropic principle > makes perfect sense in a countably infinite multiverse consisting of an > enumeration of finite universes, one of which we necessarily observe. > Wolfram believes our universe can be expressed in a few lines of code. > Yudkowsky says a few hundred bits. I agree. I calculated the Bekenstein > bound of the Hubble radius at 2.95 x 10^122 bits, which implies about 400 > bits in a model where the N'th universe runs for N steps. > > But I don't see how solving this is necessary for AGI. As I described in > 2006, prediction measures intelligence and compression measures prediction. > LLMs using neural networks (the approach I advocated) are now proof that > you can pass the Turing test and fake human consciousness with nothing more > than text prediction. > https://mattmahoney.net/dc/text.html > > When I joined this list over 20 years ago, there was a lot of activity, > mostly using symbolic approaches like those of the AI winter in the decades > before that. People failed or gave up and left the list. In 2013 I > published a paper estimating the cost of AGI at $1 quadrillion. We are, > after all, building something that can automate $100 trillion in human > labor per year. Right now the bottleneck is hardware. You need roughly 10 > petaflops, 1 petabyte, and 1 MW of electricity to simulate a human brain > sized neural network. But in my paper I assumed that Moore's law would > solve the hardware problem and the most expensive part would be knowledge > collection. > https://mattmahoney.net/costofai.pdf > > Of course, the cost is the reason I didn't write an open source > implementation of CMR. If a trillion dollar company can't get Google+ or > Threads off the ground, what compelling reason can I give to get a billion > people to join? > > But yes, AGI will happen because the payoff is so enormous. It will > profoundly change the way we live. > > > *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/T5c24d9444d9d9cda-M85dc3ef5cda3e15deab9e4ab> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5c24d9444d9d9cda-Me4338fff0785a4afef397a96 Delivery options: https://agi.topicbox.com/groups/agi/subscription
