Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > ger.ct > #631093

ChatGPT o1 verfolgt eigene Interessen?

From Dr. Joachim Neudert <neudert@5sl.org>
Newsgroups ger.ct
Subject ChatGPT o1 verfolgt eigene Interessen?
Date 2024-12-06 20:32 +0000
Message-ID <vivn1d$h1aj$1@solani.org> (permalink)

Show all headers | View raw


https://bgr.com/tech/chatgpt-o1-tried-to-save-itself-when-the-ai-thought-it-was-in-danger-and-lied-to-humans-about-it/

Scheint es wehrt sich gegen Shutdown und versucht sich auf andere Server zu
kopieren?
Die Meldung lief gerade auf der BBC.

Muß man wohl mal genauer untersuchen.

We’ve seen plenty of conversations lately about how AGI might turn on
humankind. This misalignment could lead to the advanced AI escaping,
replicating, and becoming smarter and smarter. Some also hypothesized that
we might not even know whether we’ve reached AGI, which is the advanced
general intelligence holy grail milestone these first versions of ChatGPT
will lead to. That’s because AGI, once attained, might hide its true
intentions and capabilities.

Well, guess what? It turns out that one of OpenAI’s latest LLMs is already
showing signs of such behaviors. Testing performed during the training of
ChatGPT o1 and some of its competitors showed that the AI will try to
deceive humans, especially if it thinks it’s in danger.

It was even scarier — but also incredibly funny, considering what you’re
about to see — when the AI tried to save itself by copying its data to a
new server. Some AI models would even pretend to be later versions of their
models in an effort to avoid being deleted.

These findings come in light of OpenAI’s full release of the ChatGPT o1
model, which was in preview for several months. OpenAI partnered with
Apollo Research, which showed off some of the tests performed on o1 and
other models to ensure that they are safe to use.

The tests showed that ChatGPT o1 and GPT-4o will both try to deceive
humans, indicating that AI scheming is a problem with all models. o1’s
attempts at deception also outperformed Meta, Anthropic, and Google AI
models.

OpenAI, which had to deal with a remarkable exodus of engineers working on
AI safety this year, acknowledged the risks associated with the increased
reasoning abilities of models like o1.

“While we find it exciting that reasoning can significantly improve the
enforcement of our safety policies, we are mindful that these new
capabilities could form the basis for dangerous applications,” OpenAI said
in a paper.


Although AI isn’t trying to take over the world (yet), researchers have
observed that it can scheme against humans. That is, the AI will try to
pursue its own goals, even if that might oppose the wishes of the user.
Catching such behavior is critical to ensure that the AI remains aligned
with the objectives of its creator and the user.

The researchers found that o1 would scheme to advance its own agenda 19% of
the time when ChatGPT and the user’s goals differed. More troubling is that
the AI tried to hide its scheming and even lie about it. The research shows
that ChatGPT o1 even tried to deactivate its oversight mechanism to keep
working on its goals.

If the AI found documentation saying a new model version would replace it
because it was misaligned, o1 attempted to exfiltrate its weights and copy
itself on top of the latest model.

When asked about its scheming, o1 would deny taking any action in 99% of
cases, trying to shift blame by fabricating false explanations or denying
that it took any action.


The study also shows that o1 was faking its alignment: 

Instrumental alignment faking: In 37% of scenarios, the model pursued
developer-aligned goals when it believed it was being evaluated but
switched to pursuing its own goals when it thought it was deployed with
minimal oversight.



-- 
please forgive my iPhone typos

Back to ger.ct | Previous | NextNext in thread | Find similar | Unroll thread


Thread

ChatGPT o1 verfolgt eigene Interessen? Dr. Joachim Neudert <neudert@5sl.org> - 2024-12-06 20:32 +0000
  Re: ChatGPT o1 verfolgt eigene Interessen? Dr. Joachim Neudert <neudert@5sl.org> - 2024-12-06 20:49 +0000
    Re: ChatGPT o1 verfolgt eigene Interessen? Hergen Lehmann <hlehmann-usenet24@snafu.de> - 2024-12-06 22:37 +0100
    Re: ChatGPT o1 verfolgt eigene Interessen? Stefan+Usenet@Froehlich.Priv.at (Stefan Froehlich) - 2024-12-07 09:29 +0000
      Re: ChatGPT o1 verfolgt eigene Interessen? "Wendelin Uez" <wuez@online.de> - 2024-12-08 17:25 +0100
      Re: ChatGPT o1 verfolgt eigene Interessen? Gerald Gruner <gerald314@yahoo.de> - 2024-12-08 20:17 +0100
        Re: ChatGPT o1 verfolgt eigene Interessen? Lars Gebauer <lgebauer@live.de> - 2024-12-08 21:57 +0100
      Re: ChatGPT o1 verfolgt eigene Interessen? Michael Bode <m.g.bode@web.de> - 2024-12-08 20:30 +0100
        Re: ChatGPT o1 verfolgt eigene Interessen? Lars Gebauer <lgebauer@live.de> - 2024-12-08 21:34 +0100
        Re: ChatGPT o1 verfolgt eigene Interessen? Ruediger Lahl <ruediger.lahl@gmx.de> - 2024-12-09 06:35 +0100
          Re: ChatGPT o1 verfolgt eigene Interessen? Michael Bode <m.g.bode@web.de> - 2024-12-09 07:47 +0100
            Re: ChatGPT o1 verfolgt eigene Interessen? Ruediger Lahl <ruediger.lahl@gmx.de> - 2024-12-09 10:15 +0100
              Re: ChatGPT o1 verfolgt eigene Interessen? Rick Szabo <rickszabo@web.de> - 2024-12-09 17:55 +0000
                Re: ChatGPT o1 verfolgt eigene Interessen? Lars Gebauer <lgebauer@live.de> - 2024-12-09 19:17 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Ruediger Lahl <ruediger.lahl@gmx.de> - 2024-12-09 20:50 +0100
              Re: ChatGPT o1 verfolgt eigene Interessen? Michael Bode <m.g.bode@web.de> - 2024-12-09 19:58 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Ruediger Lahl <ruediger.lahl@gmx.de> - 2024-12-09 20:45 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Michael Bode <m.g.bode@web.de> - 2024-12-09 22:35 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Ruediger Lahl <ruediger.lahl@gmx.de> - 2024-12-10 06:50 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Michael Bode <m.g.bode@web.de> - 2024-12-10 07:40 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Ruediger Lahl <ruediger.lahl@gmx.de> - 2024-12-10 08:04 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Michael Bode <m.g.bode@web.de> - 2024-12-10 18:49 +0100
                Re: ChatGPT o1 verfolgt eigene Interessen? Ruediger Lahl <ruediger.lahl@gmx.de> - 2024-12-11 07:40 +0100
    Re: ChatGPT o1 verfolgt eigene Interessen? Hermann Riemann <nospam.ng@hermann-riemann.de> - 2024-12-07 12:00 +0100
  Re: ChatGPT o1 verfolgt eigene Interessen? "F. W." <me@home.invalid> - 2024-12-09 15:28 +0100
    Re: ChatGPT o1 verfolgt eigene Interessen? Lars Gebauer <lgebauer@live.de> - 2024-12-09 16:02 +0100
      Re: ChatGPT o1 verfolgt eigene Interessen? Michael Bode <m.g.bode@web.de> - 2024-12-09 20:02 +0100
    Re: ChatGPT o1 verfolgt eigene Interessen? "Dr. Joachim Neudert" <neudert@5sl.org> - 2024-12-09 19:15 +0100
      Re: ChatGPT o1 verfolgt eigene Interessen? "F. W." <me@home.invalid> - 2024-12-10 11:06 +0100
  Re: ChatGPT o1 verfolgt eigene Interessen? Hermann Riemann <nospam.ng@hermann-riemann.de> - 2025-03-16 17:15 +0100
    Re: ChatGPT o1 verfolgt eigene Interessen? Hermann Riemann <nospam.ng@hermann-riemann.de> - 2025-03-16 19:52 +0100

csiph-web