onemetwo · HackerTrans

2 tahun yang laludiscuss

Why? I am using dumb as a low intelligence system. A more intelligent person can take advantage of new opportunities. Efficience variable: You are right that effectiveness could be better here because we are not considering resources like computer time and power.

onemetwo

2 tahun yang laludiscuss

An intelligent system could take more advantage of an increase of knowledge than a dumb one, so I should propose a simple formula: the derivative of efficiency with respect to knowledge is proportional to intelligence.

$$ I = \frac{partial E}{partial K} \simeq \frac{\delta E}{\delta K} $$

In order to estimate $I$ you have to consider that efficiency and knowledge are task related, so you could take some weighted mean $sum_T C(E,K,T)*I(E,K,T)$ where $T$ is task category. I am thinking in $C(E,K,T)$ as something similar to thermal capacity or electrical resistance, the equivalent concept when applied to task. An intelligent agent in a medium of low resistance should fly while a dumb one would still crawl.

onemetwo

2 tahun yang laludiscuss

In (1) the author use a technique to improve the performance of an LLM, he trained sonnet 3.5 to obtain 53,6% in the arc-agi-pub benchmark moreover he said that more computer power would give better results. So the results of o3 could be produced in this way using the same method with more computer power, so if this is the case the result of o3 is not very interesting.

(1) https://params.com/@jeremy-berman/arc-agi

comments