# Sam Altman en Stanford (repost de @precis0x)

- **Publicado en X:** 2026-08-03 por @precis0x
- **Duración:** 38:29
- **Fuente:** https://x.com/precisox/status/2084315686990442914
- **Origen del texto:** altman_stanford.en.vtt
- **Palabras:** 6144
- **Bloques:** 71

---

## Transcripción
**[00:00]** You with like an affordable amount of spend on tokens, you can do what a hundred person incredibly great engineering team would do as a startup. And that was just totally impossible. That was not in the set of options for a startup and now it is. So I think what you can take on, uh, the level of ambition you can have, the speed at which you can move, The amount stuff you could do at once Uh... It's just totally different And, um... Does that change the shape of problems you feel like You'd assign at end-of class for people to attack?

**[00:33]** Y'know. At the End Of That Quarter if you were teaching it again I don't think assigning Problems To Attack ever works because If i can think of a problem If I could think about a really great startup idea It's obvious enough to me Then Its probably obvious too A lot of people When we started OpenAI We were Like The Uh One maybe Generously speaking For 4 AGI efforts in the world Right And you want to find something like that and I'm sure That there exists Something today That just wasn't possible at all Pre Like automated coding era

**[01:07]** Uh It is totally not obvious There will be You know uh Multi-trillion dollar market soon Ah Only four companies are working on it right now But i don't know what that is Its much more likely Y'all Know What That Is Than I know What Thats Just My brain has taken over by OpenAI Um The kind of idea someone can assign To work On Is Probably not what you want Yep Um Okay So that That's fair Uh But I think it would be helpful Since this is a systems class To maybe uh Reason about A particular problem That You have to reason through

**[01:39]** so That They can then apply The shape of the techniques used To break down from Systems perspective That Problem Into solutions To their own problems Yeah um And A concept That Ah You had started At TEAS In Class Y'know Back in 2014 and Then Clearly We've Talked About Publicly Over Years Is Um Scale Right Scale is Its own Beast It's it's its You know Quantity Is Its Own Quality We What I Would Scale As A concept Has Been Something It Seems Like you've um Empirically Investigated In All Kinds Of ways Over

**[02:10]** The last 10 Years So Okay Um Could You Help On help Us First Unpack like what You mean By scale Now Ten years Later How would You Deconstruct That as A So I Don't Know Why The Following Observation Is True i Offer No Theory That Find Satisfying To Explain It And that Makes Me a Little Bit Nervous to Suggest You Follow it But Im Going Anyway

**[02:42]** Because Empirically It Does Seem Be true Which is All Of the Most Interesting Things Have observed In My Career And Watching Other Things Happen All Of The Most Interesting Ones Have Had Something To Do With Immersion Properties That Scale Or Scale Continuing to Provide Returns Far Beyond What the Consensus Think Will Work and This Obviously Happens with Like Scaling Loss For AI Models But this Happens With You

**[03:13]** Know Getting More Smart People Together To Think About One Problem This Has in A Research Setting This Happens With Companies And The Sort Of Economy Of Scale You Can Get In All These Different Ways I Really Learned This At Y Combinator When It Became Clear To Me That Everybody Was Saying Oh Why Commenters Gotten Too Big It Should Shrink We Should Found Less Companies Per Batch You Know The Best Times A White Common Day What When It Is Like 10 Companies Per Batch And A Lot Of Like Very Smart People Were Saying This And It Was Like Tempting

**[03:46]** Because It Would Have Been Much Less Work The Theory Is That You Know Best Companies Are Always Kind Of Obvious Then You Fund The Rest Not As Helpful But A Huge Part In Magic Of What Made YC Work Where Was Sort Of Network Effects Inside Batch And That Was An Emergent Property At Scale That Never Happened Upon This Observation Of When You Do That There's Something Important

**[04:17]** That Happens That Just Didn't Exist At All At The 110th Or 1100 Th Of Scale There Is A Bunch Of Other Examples Like This I Will Skip Them In The Interest Of Time But I Would Say Again Offer No Explanation For Why But Empirically Speaking When You Find A Time That Can Push On The Smaller Scale More Often Than Not That Seems To

**[04:49]** Be A Good Idea And It Also Seems Something Most People Don't Do Enough I Don't Offer An Explanation For This Either But Like In You Know When We Were Really Going To Scale AR Models All Of The Geniuses In The Field Most Of Them Where Oh This Isn'T Working You Know That's Barely A Scientific Result Its Not Interesting To Scale This I'm

**[05:24]** A Little Worried About It For Non Specific Reasons And Again Looking Back At Like A Huge Data Set Of People That Have Scaled Their Companies In All These Different Ways There's Almost Always Interesting Stuff There So I Think Directionally That'S Like An Interesting Thing To Push On And Severely Underexplored On The Systems Design Part Of That Uh I Don't Think One Reason People Don't Do It As Much Is Stuff Breaks Uh At An

**[05:57]** Accelerating Rate And In Unpredictable Way As You Scale It And If You Are Going To Really Scale Something Um Its Always Like A Little Bit Broken There Are Always Very Smart People Who Say Why You Shouldn T O This Y Know Dont Get Too Ambitious Dont Get Too Big Lets Try The Smaller And So Breaking That Down Is A Systems Problem I Use The Thing Of When We Were Like Scaling Up AI Models There Was Technically Can Do This At All Seems Crazy No One Ever Thought About Trying To Run Across 10,000 Or 100 Thousand GPUs And That Was Going Require Stacks

**[06:28]** Of Engineering Talent Um There Was The Capital Requirement What It Was Going To Take To Do This How Learn Something Why Not Divide It Up Among All These Projects And This Also Happens In Kind Of Every Area I've Looked At Almost Every Area For Scale And Breaking Down Into The Sort Each Difficult Are Or Each Reason Not To Do It Trying

**[06:59]** To Address Them One At A Time That's Been Really Important Um I'm Going Push On That Little Bit Because There Is Very Few People Who've Been Able To So Repeatedly Scale New Products And Systems The Way Uh The Open Ai Team Has Over The Years But It Seems Like One Of The Issues Is There Are All These Prior Conditioning Sort Of Mental Models And Expectations Humans Have And You Said Things Break And One Of The Things It Seems Often Breaks That's Hardest Re Factor

**[07:30]** Is Human The Human Side Of The Systems Design Right Wherever There's Human Implementers Or There'S Uh Human Participants In That And So What Have You Learned About Humans At Scale Like Organizing Humans At Scale To Participate In A System That May Not Be Like Just A Redo Of Some Past System That They Get Naively On At A Priority On First Blush I Think Like Clear A Clear Goal A Clear Plan To Get There And Like A Clear Answer

**[08:06]** To The Way That You're Going To Get There In Kind Of How Your Gonna Make Decisions Along The Way That's Very Important So Um If We Go Back To The Example When We Decided To Scale Up Models They Were A Lot People Who Are Like Uh This Isn't Really Gonna Work It'S Gonna Have These Problems Its Also Not You Know Can Be Like If We Get There Uh That'S Very Powerful And Then For Whatever Reason

**[08:40]** Um We Did Not Evolve To Be Good At Thinking About Exponentials People Have A Hard Time Imagining That Scaling Laws Are Going To Continue Exponentially The Revenue Will Grow Exponentally That An Organization Can Take On Exponential Complexity And In My Experience It Takes A Lot Of Time To Really Reason Through First Principles With People About Why That Can Happen Then We Take Two Examples To Walk Through The First Being Chat GPT And Second Codex You Know Both Of

**[09:12]** These Have Transformed Can Everyone Hear I'm Going Try Project It Yeah Okay So Let Me Put In A Frame And Challenge But The Assumption Is Then We Can Hopefully Reason To Example What Happened In The Case For Long Time Scaling Models A Big Mental Block That Seem To Be Prevalent In The Space Is What Are These Things Going To Useful For This You Know It's A Research Sort Of Solution Solution Chasing A Problem Research First Approach Is Not A Product And Then Chat GPT

**[09:44]** Came Out And Improved To The World That Chat Experience Was A Killer App For General Models At Scale For Consumers And Then A Couple Of Years Later Yes It's Clear That Coding Has Been The Killer Enterprise App So How Would You Compare And Contrast The Systems You Guys Used To Discover Those Use Cases Ship Them Scale Them Monetize Them Any Salient Learnings From Those Two Systems Yes So We Had Made GPT-3 And We Needed To Make Money Because

**[10:17]** We Want On Go Scale Up To You Know Billion And We Had Been Thinking Thinking We Just Couldn't Do It We Tried A Few Things They Hadn't Worked So When You The Models Bring Get Better But We Also Want To Like Start Revenue Engine Sooner And We Say Well Since We Can Figure Out What Product Build Which Is Going Put This Into An API And We Are Gonna Hope That Somebody Else In Fear Our Product Of Build And So We Launched In Like I Don't Know Something Of The

**[10:48]** Summer Of 2020 Did Gpd3 Api And Initially It Kind Got No Traction At All And Then About A Month Later Randomly As Far As We Can Tell It Went Viral On Twitter On The Same Day Uh Few Different Developers Kinda Found Gotta Do Something Cool Post That Other People Started Trying And And Then Like A Lot Of People Started Trying The Api Um But It Was Shockingly Bad If You Go Back And Use Gpd3 Or 3.5 Um you Will Be Astonished

**[11:21]** At How Bad The Models Were Then Relative To The Amount Of Excitement They Generated At Time Uh So People Tried All These Things And Really the Only Business That People Got To Work In A Significant Way With GPT Three A Better Model Although That Was The Only Business That Was Working Developers Had Figured Out How To Like Put In A Prompt And Get And Be Able Chat With It And We Saw This Alot

**[11:54]** Like More People Were Using They Couldn Get The Api Work For Their Business But They Were Using There API Key Just Chat I Said Well We Can Build A Good Chap Buck People Clearly Want That And We Had A New Model We Actually Had You Before Done But We Had An New Model We Were Ready To Release In Between Called 3.5 And We Could Figure Out A New Kind Of Post Training Where We Can Get The Models To Do Like Good Job With Instruction Following So It Make It Easier To Chat With And We Said Well You Know The Api Is Not Working Well We'll Build A Chat Bot Around It And We

**[12:36]** Put That Out And We Still Didn't Think Was Going To Do That Well There Was Really Meant As Like A Research Demo To Convince Other People They Should Build Chat Light Products In Past For The Api But That Went Like Crazy Viral Another Thing I Had Learned From YC Is When Something Starts Growing Not Very Good You Have Like A Guaranteed Hit On Your Hands And So We Had Like Five Days Where The Traffic Would Shoot Up Fall Off In Everybody Be Like Well That Was Just A Hype Cycle But Then Next Day It Would Get To Higher Peak Fall Off Again Later People Say

**[13:08]** That's A High Hype Cycle By The Fourth Or Fifth They I Was Like I Know How This Works I Know What's Going To Happen Like We Have The Potential Here At A Killer Product And We Knew We Could Make It Much Better When You We Could We Know He Had Gpt For We Knew We Could Keep Scaling But By That Fifth Day We Got Everybody Together And Said This Is An Emergency This Isn't Good Kind Of Emergency But We Have Build A Company And A Product All Once We Then Had Like Two

**[13:39]** Months Of Crazy Scaling And Then He Said You Know We Have To Figure Out A Business Model Later For Now Were Just Going To Charge People So That We Don't Run Our Compute Bills But That's Obviously Not The Long-Term Answer That Also Turned On Just Work Um And There Was A Story Of Tracking T And Then It Was So Much Utility That Before Chats Was That We Were Going To Go All In On Code Um We Knew These Models Could Write

**[14:11]** Code Uh We Knew They Be Really And We Knew It Would Like A Valuable Area But Then Had This Incredibly Exciting Thing Happen Um But Our Kind Of Internal Belief At The Time Was How These Models Would Control Things On Computers And Robots Where How These Models Were Robots And Then We Knew Intelligent To Do Stuff In The World So Uh

**[14:43]** It Took Us A While To Get There I Think Codex Got Really Good By Early This Year But With 5.5 Is When We Saw This Real Inflection Point Where People Are Now Like Doing Just Incredible Things With It Um You Know That Earlier In Class Talked About How The Capabilities Pipeline Is Starting To Look It Is Trying Become Somewhat More Legibly Standard Across Different Research Groups You Got Your Pre-Training Mid Training Post Training Then You Got The Rl And Supervised Feedback Loop Do

**[15:14]** You Think That's Roughly Like The Shape Of The Pipelines That Allowed Codex To Go Through A Capability Jump And That Will Basically Stay Stable Now In Consistent Or Are We Going To Go Through A Major Rewrite Of That Pipeline I Think It Is Definitely The Current Pipeline I Expect Will Go Through A Major Reright And Know When It Will Happen Or Exactly How But It Is Little Odd To Me That So Happens As A Pipeline Doesn't Quite Feel Like The Optimal Solution What Would Be An Optimal Solution In Your Head

**[15:46]** I Think That's A Research Problem For The AI's To Figure Out I Think We Are At A Point Where And We Set This Goal That By September Of This Year Will Use 500,000 800 Equivalent Gps Like A Lot Of Computing Power As An Ai Research Intern And By March Of 20 28 That We Have A Full End-To End Very Talented Researcher Like Figuring Out Complete Architectures So I Think We Are Going To Get Like With The Current Pipeline In The Current Architectures I Think Were Gonna Over The Line Of When Ai's Can Do Incredible Work You Know One

**[16:19]** Of Things That You Just Described There Is Your Way Even Talking A Lot About Systems Frameworks And Analogies To Make Concepts From One Domain Legible To Other People Who May Not Have All Context In Another And But Sometimes Because Of The Translation Problem You Know Reasoning By Analogy Is Not Helpful Because Then Errors Compound Yeah Right There You Said Our Goal To Use It As An Ai Intern Which Obviously A Very Useful Metaphor Within The Context Of Silicon Valley Class That Understands

**[16:52]** How These Pipelines Work And So On And Then As You Scale Actually That Matter For Globally People Who Might Have All That Context Go Start Analogizing These Models In Ways They Shouldn't Be Like How Should We Think About The Limits Of Of That What Are The Limit To Scale Of Or One Of Product Analogy Is Research An Algae You Find Most Useful Within Valley And Which One Have Found About Found About The Limits Of Those Analgy Scaling Now How Do Navigate Between Two Problems I've Been In Of

**[17:30]** Creating A New Utility This Doesn't Happen Very Often You Know Electricity Is Utilities Internet So Util It There Water I Guess There's Not A Lot Of These And So They Are Not A Lot Of Examples That We Can Study For Good Metaphors Or Learnings About How To Explain Us To The World But I Was Recently Looking At What Happened When Electric Became A Problem In The World Before And You Know It

**[18:10]** Feels Sort Of Like Very Different Than The World Before Maybe They Tried To Sell Electricity Or Market Electric At First I Don't Know But If Any Case That Didn't Work And Then What They Started Marketing Selling People Was Light At Night We Are Going To Get From Us Is Not Electric City Its Light At Night By The Way You Can Use Same Thing That Lets You Get Light For All These Other Things But People Are Like Well Why Would I Want That Now Like Well You Know It Will Wash Your Clothes For You Someday And No No One I Can't That's Too Far Of A Jump From Me Right Um

**[18:41]** So I Don't Know What Our Anal Alogy For This Should Be But I Suspect That Even If We Are Totally Right And Intelligence Is Going To Become This New Utility That Every Company Every Customer Every Government Just Needs Access Too In All Sorts Of Incredible Ways You Will Have Like A Open AI Token Subscription That You Will Plug Into Everything And Use To Access Everything And You Have Running For You All The Time Doing This Amazing Stuff I Kind Of Don't Think

**[19:12]** At Least Right Now The Right Way For Us To Analogize That Is We Are Selling Intelligence Because People Just Like Somehow Not Resonating I Don't Know What Our Equivalent Of Were Selling You Light At Night It's Going To Be But If We Will Become A New Utility We Need To Find A Way Explain To The World What It Means To Have This Like Intelligent Pike That You Can Just Do Whatever You Like With It So Question That Has Emerged An Immersion Property Of This Class Is Having A Diversity Of

**[19:44]** Different Speakers As The Utility Analogy Has Come Up Several Times But In Reference To Different Things So Jensen Likened Your Tip Like Compute To Our Utility And Why There Should Be Access And Talk About How Stanford Should Pull Budget And Procure That As A Utility For Everybody On Campus Whereas You Just Likened The Intelligence Part To Your Utilities Are Both Of These Things True Is One More Likely How Should People Reason About Compute As A Utilty Versus Tokens And By Compute I Mean Here Chips Vs. Tokens Does That

**[20:15]** Make Sense I Think As A Consumer As Like A Business Or An Individual You Will Think In Something Like The Hardware Is What Particular Chip It Is What's Powering And I Think That Stuff Will Be Abstracted Out And What You Care About When Your Interacting With This System Can Use A Lot As Cheap Doing Good Job So Right Now Its

**[20:46]** Like Tokens It May Get As We Move Into World Were We All Have Constant Agent Running For Us Being Useful To Us All Of The Time You May Think About It Is Even One Level Up But Yeah My Guess As New When You Like Pay For Your Cell Phone Bill You're Like All Right I'm Buying Access To Air Time And Some Number Of Gigabytes And You Know Its Going Do These Things And I'll Use All These Apps Whatever Else But What You Think About Pain For That Kind Internet Utility In This Case Um I Know I Could Nerd Out About

**[21:26]** Utility Infrastructure For A Long Time But I Want To Make Sure We Switch A Little Bit To Being Relevant For The Students Usually We Have Questions Where Were Not Hearing Those Today Well Unless You're Comfortable Oh Okay Great How About That Improv Ok Uh So One Final Question Start Getting The Creative Juices Flowing Is Um The Final Project For This Class Or Partly Private 183 Is The One Person Frontier Lab So Everybody Here Is Working On Projects Where They Are Simulating Being An Individual Uh As A Lab With Access To All Right Tools I've Got Hundreds Of Thousands Dollars Of Credits From

**[21:58]** Cloud Flare I Think We Have Some Open Ai Tokens Maybe But There's A Bunch Of Compute At Their Disposal Um What Would You If You Were In The Class What Would You Be Working On For Your One Person Frontier Lab Project First Of All I Think That's An Awesome Project Um I Think This Is Top Of Mind Because Uh You We Were Just Like Talking About Utility Frame Frameworks And There Was A Lot Very Smart People Working On Uh Training Ideas And We're Going To Have Incredible

**[22:29]** Models No Matter What You Do We Have Incredible Models I Promise Here Uh Like Pretty Quickly And I Think We Have Not Invested Enough In Being Able To Deliver At Scale Huge Amounts Of Cheap Intelligence So Maybe Go Work On The Influence Part Of This Stack And How Are Going To Get This Incredible Intelligence To Be Cheap And Abundant Uh I Think That's Under Invested In An N I Think All Of Frontier Labs Are Going To Have Become Inference Companies To A Senior Finger Degree Um Okay Might Be Too Late To

**[23:03]** Pivot Your Projects But Better Late Than Never Work On Whatever You Want To Work Out Uh Ok Let's Start Taking Questions And I'm Gonna Moderate And Try To Not You Know Please Try To Productive And Not Spicy Etc Remember It'S The CS Class But Up Do Sam If You First Of All In Terms Achieving Human Level Intelligence These Models Have Already Far

**[23:34]** Surpassed Human Intelligence And Some Ways And Then They're Wildly Worse Than Others Like For Exam Pl Seem Much Wors Th An People Are At Very Long Horizon Kind Of High Judgment Signal And Tasks Uh On The Other Hand Yesterday We Had One Of Our Models Discover Disprove A Conjecture One Of The Erdős Problem That Had Smart People Have Worked On For Long Time And A Lot Of People A Lot Of Smart Scientists I Don't Know If

**[24:05]** Lacuna Is One Or Not Even Quite Recently Said Something Like That Was Not Going To Happen And Then Like The Model Just Did It And You Know Now You Have All These Mathematicians Saying Like Is Math Over What Does This Mean For Our Field So Clearly LLMs Are Capable Of Figuring Out New Knowledge And Clearly They Are Capable Doing Some Things That Some Intelligence Tasks That Humans Just Can't Do There Going To Scale Much Further So How Much Better And What Distribution The Task They Can Do Better Than Humans Will Find Out But I Suspect

**[24:36]** It's A Lot In The You Know In Terms Of This Like Lack Of A Belief And The Exponential We Were Talking About Earlier I Think The Field Was Honestly Held Back By A Generation Of Scientists Who Just Were Way Too Certain On What Wouldn't It What Scaling Is Not Going To Produce And Then Some People Looked At Graph S And Said Well Looks Like It's Continuing Beautifully Let'S Keep Going I Think World Models Are Clearly Important And To Will Need That For Things Like Robotics

**[25:07]** But Betting Against LLM Scaling At This Point Feels Quite Misguided To Me It Doesn't Get Annoying To Be The I Told You So Guy Not I Mean There Are These Light Twitter Trolls That You Know Four Years Have Just Been Like Is Not Going Work Its Not Gonna Work This Is So Dumb Like You Know This Is A Fraud This Company Is Gonna Fail This Research Approach Is Going To Fail And I Used To Get More Bothered By Them But I Don't Even Feel The I Told You At This Point It's Like You Were Just

**[25:38]** There She Was Never On Your Still Going About It Like The Data Is Quite Strong Our Side Only Be That Fun To Say I Told You So And Also The Fact That Your Like Still Saying We're Wrong Doesn't Really Bother Me I Think There's A Kind Of Move On This Is Saying That Like Insanity Is Doing The Same Thing Over Again When Presented With Data That It'S Not Working In A Sense Its Form Of Insane T E I Think I Think Something Happens Which If You Make Your Ident Ty About A Particular

**[26:10]** Thing Is Going To Work Or Not Work And You Associate Yourself With That Belief And Then The Science Are The Empirical Results Disprove You And You're Like Too Hung Up On Your Ident E Can't Let It Go You Can See The Truth Yeah And I Think This Is Like An Important Reminder In Both Directions How Do You See Education Um It Clearly Has Super Adapt And I Am Worried I Thought By Now It Would Have Um The The I Think If We

**[26:42]** Continue To Teach An Evaluate Students As In A Pre-AGI World Um Its Not Going To Work And Is Gonna Lead Into Like Atrophy Of Learning How To Think Or Whatever And I Thought That Was Obvious Enough That Wasn't That Worryed You Know When Chat GBT Launch I Was Like Yeah And We're Going To Teach People So Much Better You Know People Are Going Really Get Projects Where They Have To Use AI But

**[27:15]** Still Like Stretch Their Brain More And Think More And Figure Out New Things To Do And Honestly I Struggle To Point Any Significant Systemic Change That I've Seen In The Education System At Large Since Chat GBT Launch And I Thought Would Happened So I Have Done With Every Other Technological Leap Before Redesign How Education Works So That You Still Have To Learn How Think And There Will Be Some Things Like

**[27:46]** I Am A Person Who Thinks By Writing And Write Stuff That Is Never Showed Anyone Else But It's Important For Me To Figure Something Out And So I'm Grateful People Say The Same Thing About Programming So We Teach People To Do That Machines Can Do Better Just Because It's Helpful To Teach Them The Meta Skill Of Thinking And Learning And That Makes Sense But There Are A Lot Other Things Where We Should Totally Change How We Teach Or Learn Or Evaluate If

**[28:18]** You Don't Do That I Think There Will Be Like Significant Atrophy In People Critical Thinking Skills Question Is What Was Your Favorite Class And What Would Wish You Did When Stanford Does Stamford Still Do Intro Sims I Did Like All The I Did Like Three Interest Times A Quarter My Freshman Year Like And I Loved All Of Them They Were Super Different Eye Looking Back That Fact That I Was Able To Get Such Abroad Exposure To Stuff In Have Like A Very

**[28:49]** Shall Understanding Of Lots Of Different Fields Was An Incredible Thing If It Not Been For That I Would Have Taken Cs And Physics Classes Which Still Would Be Great But Um I Think More About The Stuff The Classes I Took That Were Like Totally Random An Unrelated To What I Do Now In Some Important Way Gave Me A Perspective Than I Did I Think I Would Of Learned Program No Matter What So I Didn't Think At Time Was Like Kind Of You Know It's

**[29:22]** This Stuff Is All Cool But Mostly Going To Be About Learning Cs I Only Did Two Years Of School So There Was A Lot Of Stuff That Wanted Take That Didn't Get Too The Question What Is Your Spiciest Steak Of All I Think With More Time To Think I Could Come Up With Much Spicer One But AI Is Just Gonna Keep Going And

**[29:55]** I This Is Considered I Don't Think This Like Widely Believed Yet If It Were Widly Believd There Would Be Significantly More Reverberations That Are Happening Through Society Right Now Maybe I Don't Have A Spicer Take Actually Maybe This Is The High Order Bit But If AI Progress Continues On The Exponential That Its One For Another Three Half Years Since Tragic Metaphory Even The That Society What Is Capable

**[30:31]** Of Are Just Completely Different Well Let Me Try To Prompt With More Thinking Tokens On That One You Have If We Treated As A Model Like Frontier Model And Do Some Inherent Capabilities I Mean It Will Try To Elicit People Don't Know About For Next Two Minutes One Of Them Is That Even Post Train Now On Even Continuously Rled On Opening Eye As Well Is The External Feedback Loop Of World On What Doesn't Work And Works In Does Not Work So Now If We're Going To Treat You As A Prediction Engine For Sec The Prompt Is One Are Three

**[31:04]** Most Likely Forks Of The Universe You See Over Next Ten Years And What Is Your Probability Assessment On Each Of Those That Make Sense One It Feels Very Important Is Like How Much Of This Technology Going To Be Very Widely Democratized Versus How Much It Will Sit In A Few Companies I Think Our World There Are All These Reasons Why You Could Imagine The Default Is That This Gets Concentrated To A Few Companies And They Become Like You Know Significant

**[31:35]** Fraction Of Wealth On Earth Now Would Obviously Terrible And Work Super Hard To Push To Kind Of Utility Model Of The World Is That A It's Quite Unstable And Bad And Will Feel Unfair If A Few Companies Have All This But B I Think There Is Real Alignment Failure In A Very Fragile World And The Best Way To Get To A World We Want That Represents Like

**[32:06]** Everybody Winning And Every Values Being Represented Having An Agency Is Just Push This Technology Out Into The World But There Will Be A Very Strong Argument Against That Around Sort Of Safety Instability And I Think That Would Be Big Fork It's Very Important Encourage All You In Your Careers To Push Hard That This Is A Technology Can Bring Us An Incredible Sci-Fi Future Life Can Be Unbelievably Much Better We Are Going To Incur Some Risk To Get There But The Risk Of Keeping Us Concentrated In A Handful Of Companies Even

**[32:37]** Though We Would Be One These Companies Is Not Something You Should Tolerate So I Think That Will Be Big Fork In Terms Of Probability I Think It's The World Should Have Such An Interest And Happening This Way But I Think 80% End Up On The Democratic Path There Will Be Very Strong Safety Message And You Have An Direction But If That

**[33:11]** Fork Then You Have An Agency To Affect The Forecasts Well I Mean We Are Clear What Is Our Agency For Like This Is What We Believe In We Think That Uh You Know We Can Do Everything We Can Push It In This Direction Maybe Related There's A Lot Of Talk About Me Like Are We Gonna Capitalism With No Change Is It Like Full

**[33:42]** On Communism There's A Lot Of Talk About This One Thing That I Think Has Not Talked About Much Is How Specifically How We Distribute Computes So Maybe Alot Of The Economy Can Work In A Way That Its Going To Work And Actually I've Become Much Less Even Short-Term Jobs Doomer Have Always Been Optimistic We Find New Things To Do But This May Not Be As Disrupted Is I Originally Thought In The Short-Term But We Are Seeing Compute Shortages Now I Can Imagine Them Getting Much Worse And I Can Imagine Compete

**[34:14]** Being Like The Most Important Utility That People Need So If The Price Of Computes From A Supply Demand Perspective Gets Way Out Of Whack Then There Will Be An Interesting Fork About What It Means To Equitably Distribute Compete So You Did Do Very Interesting Things There Which He Said On The Economic Side We Might Have Need Universal Basic Income Everybody Wants A Piece Of Shares You Know One Of Speakers In This Class Is Nicolai Tangent Who Runs The Norwegian Sovereign Well Fund These Awesome His Awesome Yeah That's No Region Sovereig

**[34:46]** And Weld Fun Owns 1.5% Of All Public Trade Companies On The Planet They Also Have Effectively Universal Basic Income You Could Argue There's Flavors Of This Already Today Because You Know The Largest Employer Now In United States Is The Government And You Can Argo Like Large Sections Of That Or A Way For The Government To Redistribute Income From Taxpayers So Are These Solutions That Actually Need Be Novel Or Just Reimplemented For This Era How Do You Think About The Novelty Of Those Solutions Where We Often In Silicon Valley Have A Tendency To Be Like Reinvent Old Things From First Principles And

**[35:18]** So Should Look At Existing Systems And Tweak Them I Don't Think That These Require Deeply New Ideas Although I Will Say I Am Much More Excited About People Having Some Sort Of Funded Like A Big Universal Basic Income Study A While Ago I Have Also Watch What Happens When People Invest In Startups And Know Which Model Hits Human Psychology Better So

**[35:50]** What I Would Love To See Is As Leverage In The World Shifts From Labor To Capital That We Find A Way To Have Something Like Citizens Well Fun Than The Country Or In The World Eventually Where You Like Basically Own A Slice Of Capitalism Right On The Slices These Companies And Then In The Second Fork There Compute Bottlenecks At Some Point When Computes Prices Get Out Of Whack Between January This Year My Current Understanding Is Based

**[36:21]** On Data We Have Seen That H100 Price And Blackwell Spreads Between Long-Term Reservations And Spot Is 5X I Don't Know If It's That High Anymore I Think Got A Little Better But Yeah Time Or If You Can Even Find H100 Because They're Pretty Much All Gone For This Year Does Sound Right No Argument There Is A Gigantic Computer Shortage Out So The That Is Good Example Of An Other Systems Problem Right Now That Live At Least To Some Folks It Feels Like Covid You Know For The Compute Era Like All The

**[36:52]** Toilet Papers Gone Yeah Why Are People Not Freaking Out About This Well I Think People Assume We Will Make Big Inference Gains On Hardware We Have I Also Think There Is A Tsunami Of Hardware Coming But Maybe That Demand Tsunamis Even Bigger And Pete I Think People Should Be Freakin Out Somewhat And Would You Say It's Fair How Long Exist In A Compute Shortage At Least You Know Based On Current Data You Have I Think Like Other You Can't Talk Really About Like

**[37:23]** Worldwide Demand For Electricity Without Talking About The Price Like It's There'S An Extremely Different Demand About How Much Energy People Use In The World If The Price Comes Down By Factor Of 10 Or Goes Up By Factor Of Ten And Ai Is Like That The If We Can Make Models Sufficiently Smart And It'S A Sufficiency Low Cost I Think Demand Is Like Kind Of Uncapped And So In Some Sense As Long As We

**[37:54]** Continue To Make Progress On This There Will Be A Shortage Forever And Things Will Be Bid Among Above What The Price Should Be Even Though People Are Getting Better Smarter More Whatever Intelligence Just Because You Can Use A You Can Have 10 Of Them Running Or Working For 100 All The Think It'S A Lot Of Inference That Will Conclude Awesome With That I'm Give You The Swag Which Is Thank You For Coming Thank You