"Dr. Arne Babenhauserheide" <arne_...@web.de> writes: > "The main limitation of AlphaEvolve is that it handles problems for > which it is possible to devise an automated evaluator." (page 21) > > So it’s like a soap-bubble that minimizes surface tension on a defined > structure. ⇒ while that’s great, it may not be an example of > innovation *by the LLM*.
That is true for all human inventions as well. In natural sciences, for example, nature is the automated evaluator. In formal sciences, it can be a proof solver, a computer, or a library of theorems. LLMs need to access some ground truth to innovate. So what? Newton and Galileo did too. Does that make discoveries not innovative? Of course not. But, such reasoning does not apply in this thread, because the meaning of the term innovative is left undefined and keeps changing, while the final answer remains fixed to "there is no example of innovation by the LLM". Well then, that is the answer, no matter what. Rudy -- "Programming reliably -- must be an activity of an undeniably mathematical nature […] You see, mathematics is about thinking, and doing mathematics is always trying to think as well as possible." --- Edsger W. Dijkstra, 1981 Rudolf Adamkovič <rud...@adamkovic.org> [he/him] http://adamkovic.org --- via emacs-tangents mailing list (https://lists.gnu.org/mailman/listinfo/emacs-tangents)
