OpenAI CEO: When will AGI arrive? | Sam Altman and Lex Fridman

so one of the main concerns here is uhsomething called AI takeoffor a fast takeoff that the exponentialImprovement would be really fast towhere like in days in days yeahum I meanthere's this isn't this is a prettyserious at least to me it's become moreof a serious concernjust how amazing Chad GPT turned out tobe and then the Improvement of gbt4almost like to where it surprisedeveryone seemingly you can correct meincluding you so gpd4 has not surprisedme at all in terms of the receptionthere chat GPT surprised us a little bitbut I still was like advocating we'd doit because I thought it was going to doreally great yeah um so like you knowmaybe I thought it would have been likethe 10th fastest growing product inhistory and not the number one fastestthey're like okay you know I think it'slike hard you should never kind ofassume Something's Gonna Be Like themost successful product launch everum but we thought it was at least manyof us thought it was going to be reallygoodgvt4 has weirdly not been that much ofan update for most people you knowthey're like oh it's better than 3.5 butI thought it was going to be better than3.5 and it's cool but you know this islikesomeone said to me over the weekendyou shipped an AGI and I somehow likeI'm just going about my daily life andI'm not that impressedand I obviously don't think we shippedan AGIum but I get the point and the world iscontinuing onwhen you build or somebody Builds anartificial general intelligence wouldthat be fast or slow would weknow what's happening or notwould we go about our day on the weekendor not so I'll come back to the would wego about our day or not thing I thinkthere's like a bunch of interestinglessons from kovid and the UFO videosand a whole bunch of other stuff that wecan talk to there but on the takeoffquestion if we imagine a two by twoMatrix of short timelines till AGIstarts long timelines till AGI startsslow take off fast takeoffdo you have an instinct on what do youthink the safest quadrant would be so uhthe different options are less next yearyeah say the takeoff that we start thetakeoff period yeah next year or in 20years 20 years and then it takesone year or 10 years well you can evensay one year or five years whatever youwantthe takeoff I feel like nowis uh is saferso do I so I'm in longer no I'm in theseslowtake off short timelines is the mostlikely good world and we optimize thecompany tohave Maximum Impact in that world to tryto push for that kind of a worldand the decisions that we make areyou know there's like probability massesbut weighted towards thatand I thinkI'm very afraid of the fast takeoffsI think in the longer timelines it'sharder to have a slow take off there's abunch of other problems tooum but that's what we're trying to do doyou think gpt4 is an AGII think if it isjust like with the UFO videosuh we wouldn't know immediatelyI think it's actually hard to know thatwhen I've been thinking I was playingwith GPT forand thinking how would I know if it's anAGI or not because I think in terms ofuh to put it in a different wayum how much of AGI is the interface Ihave with the thingand how much of it uh is the actualwisdom inside of it like uh part of methinks that you could have a modelthat's capable of super intelligenceand uh it just hasn't been quiteunlocked this is what I saw with ChadGPT just doing that little bit of Rowith human feedback makes the thingsomehow much more impressive much moreusable so maybe if you have a few moretricks like you said there's likehundreds of Tricks inside open AI a fewmore tricks and also in holy