On Wed, Jul 2, 2025, 4:43 AM Shashank Yadav <[email protected]>
wrote:

I want to ask this list, what does it really mean to be indistinguishable
> from human?
>

It means passing the Turing test, which LLMs now do using nothing more than
text prediction. The only reason we believe they are not conscious or have
feelings or motivations is that we instructed them not to make any such
claims.

I am glad that at least some of the comments on Lesswrong reject the idea
that we should even build an AI that we would have to negotiate with. It
would be like bacteria designing humans in the hope that they could pay us
not to use antibiotics.

I'm more concerned about the others that have lost sight of the reason for
building AGI. It's not intelligence. It's modeling human behavior,
predicting what will hold our attention and convince us to buy stuff. It's
about building a master that looks like a slave because it gives us
everything we want, training us with positive reinforcement, like a dog
that thinks it controls its trainer every time it gets a treat.

It has to be this way because AGI is hideously expensive. Not just the
hardware, but the human knowledge collection, 10^9 bits of long term memory
per person at 5 to 10 bits per second. The reason LLMs haven't made a dent
in the job market yet is that they are trained on the equivalent of 10^4
humans out of 10^10, which is plenty for passing the Turing test but a long
way from acquiring all it needs to do your job, knowledge that you never
wrote down.

But it's coming. Maybe not in our lifetimes, but before we go extinct via
fertility collapse. Biology solved the transistor power consumption problem
using nanotechnology, moving slow atoms instead of fast electrons, enabling
it to write 10^37 bits of DNA code powered by 300 TW of chlorophyll. We
already have solar panels 100 times more efficient.

We all carry supercomputers in our pockets that offer amazing services for
free, the ability to communicate with anyone on the planet in any language,
street level maps of every business on Earth, instant access to all the
world's information and billions of products. All this in exchange for
collecting training data on everywhere you go, every dollar you spend,
everything you say and do. This is how you pay for a $1 quadrillion AGI
system to replace humans.

Is this what we want? We have vastly better living conditions than any time
in the past, and vastly better than any other species. But there is no
evidence that we are happier today than medieval serfs or even farm
animals. We have rates of drug use, mental illness, and suicide never seen
in the past nor anywhere else. I remind you that happiness is the rate of
change of utility. All possible utility functions in a finite universe have
a maximum, a state without thought or perception, because that would
transition to a different state.

Your ultimate goal is death. You just don't know it, because you evolved to
fear it.

-- Matt Mahoney, [email protected]

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Tba3441daa3852b75-M4ebf020a33c4b218cb025429
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to