5 ms·

Factually inaccurate results = unsafety. This cannot be fixed under the current model, which has no concept of truth. What kind of "safety" are they talking abo

Factually inaccurate results = unsafety. This cannot be fixed under the current model, which has no concept of truth. What kind of "safety" are they talking about then?

Meekro 3y ago
In the context of this thread, "safety" refers to making sure we don't create an AGI that turns evil. You're right that wrong answers are a problem, but plain old capitalism will sort that one out-- no one will want to pay $20/month for a chatbot that gets everything wrong.
- resource0x 3y ago
  How the thing can be called "AGI" if it has no concept of truth? Is it like "60% accuracy is not an AGI, but 65% is"? The argument can be made that 90% accuracy is worse than 60% (people will become more confident to trust the results blindly).
spaceman_2020 3y ago
If factually inaccurate results = unsafety, then the internet must be the most unsafe place on the planet!
- resource0x 3y ago
  The internet is not called "AGI". It's the notion of AGI that brought "safety" to the forefront. AI folks became victims of their hype. Renaming the term into something less provocative/controversial (ML?) can reduce expectations to the level of the internet - problem solved?
  - autoexec 3y ago
    > The internet is not called "AGI" Nether is anything else in existence. I'm glad that philosophers are worrying about what AGI might one day mean for us but it has nothing to do with anything happening in the world today.
    - resource0x 3y ago
      I fully agree with that. But if you read this thread or any other recent HN thread, you will see "AGI... AGI... AGI" as if it's a real thing. The whole openai debacle with firing/rehiring sama revolves around (non-existent) "AGI" and its imaginary safety/unsafety, and if you dare to question this whole narrative, you will get beaten up.
s1artibartfast 3y ago
Truth has very little to do with the safety questions raised by AI. Factually accurate results also = unsafety. Knowledge = unsafety, free humans = unsafety.
- resource0x 3y ago
  But they (AI folks) keep talking about "safety" all the time. What is their definition of safety then? What are they trying to achieve?
  - s1artibartfast 3y ago
    I dont think it has a fixed definition. It is an ambiguous idea that AI will not do or lead to bad things.

Factually inaccurate results = unsafety. This cannot be fixed under the current model, which has no concept of truth. What kind of "safety" are they talking about then?

Meekro 3y ago

In the context of this thread, "safety" refers to making sure we don't create an AGI that turns evil. You're right that wrong answers are a problem, but plain old capitalism will sort that one out-- no one will want to pay $20/month for a chatbot that gets everything wrong.

resource0x 3y ago
How the thing can be called "AGI" if it has no concept of truth? Is it like "60% accuracy is not an AGI, but 65% is"? The argument can be made that 90% accuracy is worse than 60% (people will become more confident to trust the results blindly).

spaceman_2020 3y ago

If factually inaccurate results = unsafety, then the internet must be the most unsafe place on the planet!

resource0x 3y ago
The internet is not called "AGI". It's the notion of AGI that brought "safety" to the forefront. AI folks became victims of their hype. Renaming the term into something less provocative/controversial (ML?) can reduce expectations to the level of the internet - problem solved?
- autoexec 3y ago
  > The internet is not called "AGI" Nether is anything else in existence. I'm glad that philosophers are worrying about what AGI might one day mean for us but it has nothing to do with anything happening in the world today.
  - resource0x 3y ago
    I fully agree with that. But if you read this thread or any other recent HN thread, you will see "AGI... AGI... AGI" as if it's a real thing. The whole openai debacle with firing/rehiring sama revolves around (non-existent) "AGI" and its imaginary safety/unsafety, and if you dare to question this whole narrative, you will get beaten up.

s1artibartfast 3y ago

Truth has very little to do with the safety questions raised by AI. Factually accurate results also = unsafety. Knowledge = unsafety, free humans = unsafety.

resource0x 3y ago
But they (AI folks) keep talking about "safety" all the time. What is their definition of safety then? What are they trying to achieve?
- s1artibartfast 3y ago
  I dont think it has a fixed definition. It is an ambiguous idea that AI will not do or lead to bad things.