OpenAI CEO: "Super Human Coders By End of 2025"

[John Clark]

Sam Altman, the CEO of OpenAI, has recently made a number of points which I think are super important:

1) Open AI has found that the amount of intelligence an AI has is roughly equivalent to the logarithm of the number of flops it uses, so a small increase in computer power leads to a big increase in intelligence.

2) The cost of using an AI falls by about 10x every 12 months, this is far faster than Moore's Law which is about doubling every 24 months.  

3) If the intelligence of an AI increases linearly with time (like the function 2n does) then the socioeconomic value it produces will increase super-exponentially (like the function 2^2^n does) 

4) By the end of THIS YEAR, Altman expects an AI Will be the best computer programmer in the world. 

OpenAI CEO: "Super Human Coders By End of 2025"

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

ldg

[Russell Standish]

On Tue, Feb 11, 2025 at 01:01:46PM -0500, John Clark wrote:
> Sam Altman, the CEO of OpenAI, has recently made a number of points which I
> think are super important:
>
> 1) Open AI has found that the amount of intelligence an AI has is roughly
> equivalent to the logarithm of the number of flops it uses, so a small increase
> in computer power leads to a big increase in intelligence.

If it is l
ogarithmic, then it requires an exponential growth in
computing power to generate a linear increase in
intelligence. Something about this statement of Sam's makes not sense.

>
> 2) The cost of using an AI falls by about 10x every 12 months, this is far
> faster than Moore's Law which is about doubling every 24 months.  
>
> 3) If the intelligence of an AI increases linearly with time (like the function
> 2n does) then the socioeconomic value it produces will increase
> super-exponentially (like the function 2^2^n does) 
>

Not sure what basis he has for saying this (other than helping his
share price). Usually, these things follow a quadratic dependence - eg
Metcalfe's law.

> 4) By the end of THIS YEAR, Altman expects an AI Will be the best computer
> programmer in the world. 
>

I'm sceptical of the timescale here, but they are already useful
assistants - maybe better than the "freshly minted CS intern" that
they're often portrayed to be.

--

----------------------------------------------------------------------------
Dr Russell Standish Phone 0425 253119 (mobile)
Principal, High Performance Coders hpc...@hpcoders.com.au
http://www.hpcoders.com.au
----------------------------------------------------------------------------

[John Clark]

On Tue, Feb 11, 2025 at 5:49 PM Russell Standish <li...@hpcoders.com.au> wrote:

> If it is logarithmic, then it requires an exponential growth in

computing power to generate a linear increase in
intelligence. Something about this statement of Sam's makes not sense.

Sorry, I mangled Altman's quote.  The actual quote is "The intelligence of AI models scales with the logarithm of resources used to train and run them. companies can spend virtually unlimited amounts of money and achieve continuous, predictable gains, which is a pattern that holds true across many orders of magnitude"

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

ldb

[Brent Meeker]

On 2/11/2025 10:01 AM, John Clark wrote:

Sam Altman, the CEO of OpenAI, has recently made a number of points which I think are super important:

1) Open AI has found that the amount of intelligence an AI has is roughly equivalent to the logarithm of the number of flops it uses, so a small increase in computer power leads to a big increase in intelligence.

The logarithm function is sublinear.  So a big increase increase in flops equates to a small increase in intelligence.

2) The cost of using an AI falls by about 10x every 12 months, this is far faster than Moore's Law which is about doubling every 24 months.  

3) If the intelligence of an AI increases linearly with time (like the function 2n does) then the socioeconomic value it produces will increase super-exponentially (like the function 2^2^n does)

It seems to me that (2) and (3) are properties of new technologies in general.  Once they reach the level of application they exhibit a burst of growth.

Brent

4) By the end of THIS YEAR, Altman expects an AI Will be the best computer programmer in the world. 

OpenAI CEO: "Super Human Coders By End of 2025"

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

ldg

--
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-li...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/everything-list/CAJPayv3_wqxY1gb%3Dy8DT8OR1OfxShtDYbM2cM6r0u98Zzsi0Vg%40mail.gmail.com.