We could soon see reproductive AI systems capable of hap Ph.D. exams thanks to more “ durable ” memory and more robust reasoning operations , Microsoft CTO Kevin Scott revealed when he took to the stage with Reid Hoffman during a Berggruen Salon in Los Angelesearlier this week .

“ It ’s sort of weird mighty now that you have these interactions with agents and the storage is entirely episodic , ” he bemoan . “ You have a dealings , you do a matter . It ’s useful or not for whatever task you were doing , and then it forgets all about it . ” The AI system is n’t learning from or even remembering late fundamental interaction with the exploiter , he keep on . “ There ’s no way for you to refer back to a matter you were trying to get [ the AI ] to solve in the yesteryear . ”

However , Scott is optimistic that,”we’re seeing technically all of the thing fall in office to have really long-lasting memories with the systems . ” With more lasting memory , future AI systems will be able to respond more naturally and more accurately over the duad of multiple conversations rather than being limited to the current academic session .

OpenAI announced in Februarythat it was beginning to test a new persistent memory system , rolling it out to take free and Plus subscription user . Enabling the feature film earmark the AI to recall user tone , interpreter , and format predilection between conversations as well as make suggestions in new projects based on details the user cite in previous chats .

Scott was also buoyant about meliorate the “ fragility ” found in the reasoning of many AI systems today . “ It ca n’t solve very complicated math problem , ” he explained . “ It has to bail out to other system to do very complicated things . ”

“ Reasoning , I think , gets a lot better , ” he continued . He comparesGPT-4and the current generation of models to high schoolerspassing their AP exams . However , the next coevals of ai “ could be the thing that could pass your qualified test . ”

To escort , reproductive AI arrangement have outperformed their flesh - and - blood counterpartson a variety of test and task format . Last November , for case , GPT-4 pass the Multistate Professional Responsibility Exam(MPRE ) , better known as the legal community exam , with 76 % correct — that ’s six points higher than the state average for humans .

Scott was prompt to point out , however , that training productive AIs to pass Ph.D. exams “ probably sounds like a bigger bargain than it actually is … the actual test will be what we choose to do with it . ”

Scott was specially excited to see the barriers to entry falling off so quickly . He observe that when he mother into political machine con two decades ago , his work expect alum - level cognition , stacks upon stacks of “ very daunting , complicated , technical papers to figure out how to do what I wanted to do , ” and around six month of coding . That same labor today , he said , “ a high school scholarly person could do in a Saturday dawn . ”

These lour barriers to entry will in all likelihood speed up the democratization of AI , Scott concluded . Finding solutions to the myriad societal , environmental , and technological crises facing humanness are not — and can not — be the sole responsibility of “ just the people at tech companies in Silicon Valley or just the great unwashed who graduated with Ph . D.s from top - five university , ” he said . “ We have 8 billion masses in the humans who also have some idea about what it is that they want to do with powerful tools , if they just have access code to them . ”