The goal of AGI is to automate human labor. It requires solving hard
problems like vision, language, robotics, art, and modeling human
behavior.

We don't need to define intelligence to solve these problems. If you
want to define intelligence as survival, then that is not what we want
to build. We want the type of intelligence that can produce
antibiotics, not the type that learns to evade them. It is also not
the Turing test, because that requires reproducing human weaknesses as
well as skills. If you want to define intelligence as the goal, then
the appropriate measure is dollars per hour.

Lossless compression is a useful way to measure language models, but
not the other human skills required to automate work. With regard to
language, it is more useful than a Turing test because it allows
machines to predict text faster and more accurately than humans
without being penalized for it.

On Thu, Feb 20, 2020 at 8:53 PM <[email protected]> wrote:
>
> Have you tried telling OpenAI to use Lossless Compression as the evaluation 
> for GPT-2 etc? They seem to have not used it yet got GPT-2. But with it, 
> perhaps they can achieve something better.
> Artificial General Intelligence List / AGI / see discussions + participants + 
> delivery options Permalink



-- 
-- Matt Mahoney, [email protected]

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Tc67faac3048278cf-M779eb76a4744a044efb5ef2d
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to