Or at least won’t be the kind of “foom” that roomers think it will. The idea that RSI will lead to exponential improvement has to assume that not only are AI researchers the ONLY bottleneck in the equation, but also that there is infinite compute and that training models is instantaneous. Yeah sure, AI that is slightly smarter then every AI researcher will give you better tips to improve its successor, but it will also need to collect more data and train another trillion parameter model on all that data. And the idea that it will just generate higher quality data for itself and get into a loop is retarded assumption that kept medieval and Ancient Greek philosophy from technologically improving. You can’t just assume few axioms and rationalise your way into explaining everything because you will arrive at unknown unknowns and your logic is not perfect. Science unlike rationalisation needs feedback from the real world which is limited by the number of labs present to do experiments in, so the idea that RSI could give us AGI or ASI quickly is fucking stupid, yet everyone pretends that singularity like this could happen.
tldr
We have been in land of rationalizations since al gore invented floating point quantum fuzzy logic. What exactly are you saying? That there are not sufficient incentives? That training data will not be rigorous? In some ways a lack of rigor has heuristic utility. Honestly the only pain point is having to call it "ai".
Idiots have believed for the last 50 or so years that improvements in technology will "exponentially" increase the speed of new improvements in technology so that "by the time I'm 60, we'll all become immortal" (lol) or "in just 2 more years we'll have a man on the moon again" or "mars colonies by 2030" or "superpower by 2020" etc.The reality is that technology plateaus off and basically follows moore's law. When there's lots of low hanging fruit, improvements are fast. When there's no more low hanging fruit, everything stagnates and enshittifies as greedy corps try to squeeze more performance out of existing concepts by minmaxing and reducing reliability of everything.
>the business strategy
>>109399586People have believed all kind of stuff since ever, but you have to burry your head in the sand to not see that AI is huge and everything will change.
>>109401381this but unironicallyhttps://www.youtube.com/shorts/pLnyjxgFxew
>>109401650>everything will change.Two more weeks (since 4 years and counting)
>>109399506>>109399586It's been known for some time (in the defense/geospatial community) that 'fast RSI' (hard takeoff scenario) is impossible.
