@thomasfbloom To me this sounds like the organic food labelling debate: You want to pay more for the same food and want anti-fraud assurance that it was indeed inefficiently produced in traditional ways.
@ole_b_peters I once had a few left over pieces after assembling that 3000+ pcs Saturn V rocket. I took a picture of the pieces and asked the AI where these could possibly be missing. It pin-pointed the exact places and manual pages where they were placed. 🤯 And that was about a year ago...
@ID_AA_Carmack The cost function should be their token use. Modularization pays off in terms of tokens when re-use of modules is maximized. Needs to be accounted for across model-life time scales.
@lukOlejnik It seems like group theoretic problems are easily encoded with linear number of variables making them vulnerable to the subexponential counting algorithm.
@rperezmarco@danieltanfh95@OpenAI@SebastienBubeck (3) the physicists preferred some gate keeping by the big wigs for whatever reason or consultation. Let's see what works and what is appreciated most.
@rperezmarco@danieltanfh95@OpenAI@SebastienBubeck Chill out. They are experimenting with 3 modes of community interaction after the negative feedback on their previous release. (1) the mathematicians' whatever comitee advised on immediate release, (2) the CS folks wanted immediate release, shepherding by leaders + 1 conference,