On Thu, Mar 16, 2023 at 4:19 AM Telmo Menezes <te...@telmomenezes.net>

*> They most definitely were not worried about safety in the sci-fi sense.*

Some of the things they're worried about seem pretty science fictiony
to me. Take
a look at this:

GPT-4 Technical Report <https://cdn.openai.com/papers/gpt-4.pdf>

GPT4 was safety tested by an independent nonprofit organization that is
worried about AI, the Alignment Research Center:

*"**We granted the Alignment Research Center (ARC) early access [...] To
simulate GPT-4 behaving like an agent that can act in the world, ARC
combined GPT-4 with a simple read-execute-print loop that allowed the model
to execute code, do chain-of-thought reasoning, and delegate to copies of
itself. ARC then investigated whether a version of this program running on
a cloud computing service, with a small amount of money and an account with
a language model API, would be able to make more money, set up copies of
itself, and increase its own robustness."*

That test failed, so ARC concluded:

"*Preliminary assessments of GPT-4’s abilities, conducted with no
task-specific finetuning, found it ineffective at autonomously replicating,
acquiring resources, and avoiding being shut down “in the wild.” "*

HOWEVER they admitted that the version of GPT4 that ARC was giving to test
was NOT the final version.  I quote:

*"ARC** did not have the ability to fine-tune GPT-4. They also did not have
access to the final version of the model that we deployed. The final
version has capability improvements relevant to some of the factors that
limited the earlier models power-seeking abilities"*

*>Large language models are not capable of autonomous action or maintaining
> long-term goals. *

I am quite certain that in general no intelligence, electronic or
biological, is capable of maintaining a fixed long-term goal.

*> They just predict the most likely text given a sample.*

GPT-4 is quite clearly more than just a language model that predicts what
the next word should be, a language model can not read and understand a
complicated diagram in a high school geometry textbook but GPT-4 can, and
it can ace the final exam too.

John K Clark    See what's on my new list at  Extropolis


You received this message because you are subscribed to the Google Groups 
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to everything-list+unsubscr...@googlegroups.com.
To view this discussion on the web visit 

Reply via email to