Anthropic ​wants governments ​to regulate AI to avoid catastrophe​ 18 months

[John Clark]

Anthropic says AI is improving too fast, and claims government regulation is needed. It says "legislation should be easy to understand and implement".  I am EXTREMELY skeptical that regulation could work,  but they do provide some interesting examples concerning the rate of improvement of AI. 

"On the SWE-bench software engineering task, models have improved from being able to solve 1.96% of a test set of real-world coding problems (Claude 2, October 2023) to 13.5% (Devin, March 2024) to 49% (Claude 3.5 Sonnet, October 2024)"

"AI systems have improved their scientific understanding by nearly 18% from June to September of this year alone, according to benchmark test GPQA. OpenAI o1 achieved 77.3%; on the hardest section of the test; human experts scored 81.2%."

Anthropic wants governments to regulate AI to avoid catastrophe 18 months

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

esm

[John Clark]

And Donald Trump believes the things we should be worried about are illegal immigrants hogging all the juicy tomato picking jobs and transsexual men in women's sports. 

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

tms

esm

[Giulio Prisco]

On Sat, Nov 2, 2024 at 3:02 PM John Clark <johnk...@gmail.com> wrote:
>
> Anthropic says AI is improving too fast, and claims government regulation is needed. It says "legislation should be easy to understand and implement". I am EXTREMELY skeptical that regulation could work, but they do provide some interesting examples concerning the rate of improvement of AI.
>

Great. So China will develop AI at full speed. North Korea and Iran
will develop AI at full speed. And Russia of course. Rogue
corporations will develop AI at full speed. Terrorists and criminals
will develop AI at full speed. And we will do nothing.
If AI is outlawed, only outlaws will have AI.

> "On the SWE-bench software engineering task, models have improved from being able to solve 1.96% of a test set of real-world coding problems (Claude 2, October 2023) to 13.5% (Devin, March 2024) to 49% (Claude 3.5 Sonnet, October 2024)"
>
> "AI systems have improved their scientific understanding by nearly 18% from June to September of this year alone, according to benchmark test GPQA. OpenAI o1 achieved 77.3%; on the hardest section of the test; human experts scored 81.2%."
>
> Anthropic wants governments to regulate AI to avoid catastrophe 18 months
>
> John K Clark See what's on my new list at Extropolis
> esm
>
>

> --
> You received this message because you are subscribed to the Google Groups "extropolis" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to extropolis+...@googlegroups.com.
> To view this discussion visit https://groups.google.com/d/msgid/extropolis/CAJPayv2sZPdviChtTGvqr4RzdC3FddQzKcBK9jqLQVKwUoyPpg%40mail.gmail.com.

[Keith Henson]

On Sat, Nov 2, 2024 at 10:09 AM Giulio Prisco <giu...@gmail.com> wrote:
>
> On Sat, Nov 2, 2024 at 3:02 PM John Clark <johnk...@gmail.com> wrote:
> >
> > Anthropic says AI is improving too fast, and claims government regulation is needed. It says "legislation should be easy to understand and implement". I am EXTREMELY skeptical that regulation could work, but they do provide some interesting examples concerning the rate of improvement of AI.
> >
>
> Great. So China will develop AI at full speed.

That's not clear. Last I heard, AI was discouraged because they could
not get it to toe the party dogma.

I may be out of date on this. Anyone know?

Keith

North Korea and Iran
> will develop AI at full speed. And Russia of course. Rogue
> corporations will develop AI at full speed. Terrorists and criminals
> will develop AI at full speed. And we will do nothing.
> If AI is outlawed, only outlaws will have AI.
>
> > "On the SWE-bench software engineering task, models have improved from being able to solve 1.96% of a test set of real-world coding problems (Claude 2, October 2023) to 13.5% (Devin, March 2024) to 49% (Claude 3.5 Sonnet, October 2024)"
> >
> > "AI systems have improved their scientific understanding by nearly 18% from June to September of this year alone, according to benchmark test GPQA. OpenAI o1 achieved 77.3%; on the hardest section of the test; human experts scored 81.2%."
> >
> > Anthropic wants governments to regulate AI to avoid catastrophe 18 months
> >
> > John K Clark See what's on my new list at Extropolis
> > esm
> >
> >
> > --
> > You received this message because you are subscribed to the Google Groups "extropolis" group.
> > To unsubscribe from this group and stop receiving emails from it, send an email to extropolis+...@googlegroups.com.
> > To view this discussion visit https://groups.google.com/d/msgid/extropolis/CAJPayv2sZPdviChtTGvqr4RzdC3FddQzKcBK9jqLQVKwUoyPpg%40mail.gmail.com.
>
> --
> You received this message because you are subscribed to the Google Groups "extropolis" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to extropolis+...@googlegroups.com.

> To view this discussion visit https://groups.google.com/d/msgid/extropolis/CAKTCJydnjPB5dJ-%3DWBVXdK8AbnwLNXmCCTZ0CzxZoZDkYDL0hQ%40mail.gmail.com.