[
  {
    "startMs": 1309,
    "text": "[music]"
  },
  {
    "startMs": 12639,
    "text": "I'm Diego Almeida and this talk is I"
  },
  {
    "startMs": 16000,
    "text": "love you so jeving much. Um, Glass"
  },
  {
    "startMs": 19279,
    "text": "Animals reference. We're not just weird."
  },
  {
    "startMs": 21279,
    "text": "Um, like I said, I'm Diego Almeida. I'm"
  },
  {
    "startMs": 24720,
    "text": "known as the Jev guy. Sometimes Jevy"
  },
  {
    "startMs": 27039,
    "text": "Pinkman. um also co-created chat GPT,"
  },
  {
    "startMs": 30480,
    "text": "RLHF and instruct GPT, but I assume that"
  },
  {
    "startMs": 33600,
    "text": "for the crowd today, you all know who I"
  },
  {
    "startMs": 35760,
    "text": "am, at least mostly. [snorts] Um I have"
  },
  {
    "startMs": 39120,
    "text": "like a bunch of AI specific stuff I want"
  },
  {
    "startMs": 41680,
    "text": "to get to, but I like going into my"
  },
  {
    "startMs": 43200,
    "text": "favorite topics in AI. First one is what"
  },
  {
    "startMs": 46719,
    "text": "the [&nbsp;__&nbsp;] is actually going on and where"
  },
  {
    "startMs": 48960,
    "text": "is all the automation. If you've heard"
  },
  {
    "startMs": 50719,
    "text": "me say this before, I'm sorry, but it's"
  },
  {
    "startMs": 52640,
    "text": "such an important thing to beat into the"
  },
  {
    "startMs": 54719,
    "text": "field because people still do not get"
  },
  {
    "startMs": 56559,
    "text": "it. And I need to yell this until"
  },
  {
    "startMs": 58399,
    "text": "everyone really understands what the"
  },
  {
    "startMs": 60480,
    "text": "[&nbsp;__&nbsp;] is going on. And if you're an AI,"
  },
  {
    "startMs": 62559,
    "text": "you should have a really good answer"
  },
  {
    "startMs": 63840,
    "text": "about why AI can do so many awesome"
  },
  {
    "startMs": 66400,
    "text": "things and yet still be so useless for a"
  },
  {
    "startMs": 68640,
    "text": "bunch of other ones. My relationship to"
  },
  {
    "startMs": 71200,
    "text": "chat GPT is somewhat unique. One of the"
  },
  {
    "startMs": 73760,
    "text": "few people who hates on chat GPT who"
  },
  {
    "startMs": 76560,
    "text": "made it. I don't hate chat GPT to be"
  },
  {
    "startMs": 79280,
    "text": "clear. I love using agents. I love using"
  },
  {
    "startMs": 81840,
    "text": "coding agents. I love chat bots. They're"
  },
  {
    "startMs": 83840,
    "text": "just quite limited. And in particular, I"
  },
  {
    "startMs": 85920,
    "text": "think it's a little tragic how much"
  },
  {
    "startMs": 87759,
    "text": "they've bifurcated the field into"
  },
  {
    "startMs": 89759,
    "text": "overpromise, underdeliver. Really,"
  },
  {
    "startMs": 92000,
    "text": "really terrible stuff for the world in"
  },
  {
    "startMs": 93280,
    "text": "my opinion. And I like to talk about"
  },
  {
    "startMs": 95920,
    "text": "like the two cults in AI and how we can"
  },
  {
    "startMs": 98799,
    "text": "resolve the cognitive dissonance that"
  },
  {
    "startMs": 100960,
    "text": "everyone I think should be having unless"
  },
  {
    "startMs": 103439,
    "text": "they have a very good answer. There's"
  },
  {
    "startMs": 105439,
    "text": "cult one, which is, you know, the cult"
  },
  {
    "startMs": 107200,
    "text": "of like EAS. They look at all the"
  },
  {
    "startMs": 109360,
    "text": "benchmarks and they say like not only"
  },
  {
    "startMs": 110960,
    "text": "are LMS surpassing human level on"
  },
  {
    "startMs": 113680,
    "text": "basically every benchmark they have been"
  },
  {
    "startMs": 115680,
    "text": "for two years we've been announcing AGI"
  },
  {
    "startMs": 118000,
    "text": "three months away for two years now even"
  },
  {
    "startMs": 120479,
    "text": "since like the 03 days do people"
  },
  {
    "startMs": 121920,
    "text": "remember that no one even care cares"
  },
  {
    "startMs": 123520,
    "text": "about that model anymore and it was that"
  },
  {
    "startMs": 125680,
    "text": "that was like peak internal to openai"
  },
  {
    "startMs": 127600,
    "text": "AGI hype and on the other hand you look"
  },
  {
    "startMs": 130000,
    "text": "at the nonbenchmarks and you look at the"
  },
  {
    "startMs": 132959,
    "text": "reality and you see the rest of the"
  },
  {
    "startMs": 134640,
    "text": "world especially if you guys go outside"
  },
  {
    "startMs": 136080,
    "text": "and leave San Francisco a lot of the"
  },
  {
    "startMs": 138000,
    "text": "world is like really unchanged despite"
  },
  {
    "startMs": 140560,
    "text": "AI being so [&nbsp;__&nbsp;] cool. And I think AI"
  },
  {
    "startMs": 143360,
    "text": "is transformative. It's it's a rare"
  },
  {
    "startMs": 145040,
    "text": "point of view to actually think AI is"
  },
  {
    "startMs": 146879,
    "text": "cool and transformative and also like"
  },
  {
    "startMs": 149920,
    "text": "underwhelming the world right now. And"
  },
  {
    "startMs": 152239,
    "text": "the tricky part is this is the same"
  },
  {
    "startMs": 153760,
    "text": "field and we are all seeing the same"
  },
  {
    "startMs": 155760,
    "text": "facts. So what gives you know why is AI"
  },
  {
    "startMs": 159200,
    "text": "jagged so to speak like why can it be so"
  },
  {
    "startMs": 161280,
    "text": "good at some things and so not good at"
  },
  {
    "startMs": 163920,
    "text": "other things and what is the aam's razor"
  },
  {
    "startMs": 166959,
    "text": "explanation for that. Um I will be"
  },
  {
    "startMs": 169120,
    "text": "blasting through my slides fast. I'm"
  },
  {
    "startMs": 171040,
    "text": "sorry maybe I could slow down for people"
  },
  {
    "startMs": 172879,
    "text": "taking pictures. Um but uh I'm noticing"
  },
  {
    "startMs": 176319,
    "text": "that uh this is going to go fast is all"
  },
  {
    "startMs": 178560,
    "text": "I'm saying. So I think that we as a"
  },
  {
    "startMs": 180720,
    "text": "field should have an aams razor"
  },
  {
    "startMs": 182480,
    "text": "explanation for why are there some"
  },
  {
    "startMs": 184800,
    "text": "things that are just not just good they"
  },
  {
    "startMs": 187360,
    "text": "are unbelievably good superhuman like"
  },
  {
    "startMs": 189440,
    "text": "millennium price problems in math but on"
  },
  {
    "startMs": 191680,
    "text": "the other hand they're not just bad"
  },
  {
    "startMs": 193599,
    "text": "they're so useless they can't even be"
  },
  {
    "startMs": 195599,
    "text": "used to touch accounting that sounds way"
  },
  {
    "startMs": 197599,
    "text": "easier than million prize problems in"
  },
  {
    "startMs": 198800,
    "text": "math right I could do the latter I"
  },
  {
    "startMs": 200480,
    "text": "definitely could not do the former and I"
  },
  {
    "startMs": 201760,
    "text": "was a mathlete so what gives here what"
  },
  {
    "startMs": 204560,
    "text": "is the reason for this and I normally"
  },
  {
    "startMs": 207200,
    "text": "love this as an interactive thing, but I"
  },
  {
    "startMs": 209120,
    "text": "will instead just tell you my answer."
  },
  {
    "startMs": 210640,
    "text": "Sorry for the power dynamic. Um,"
  },
  {
    "startMs": 212640,
    "text": "everything on the left side is"
  },
  {
    "startMs": 214720,
    "text": "intrinsically a human in the loop task."
  },
  {
    "startMs": 216879,
    "text": "Like deep research, I could not do"
  },
  {
    "startMs": 218720,
    "text": "better than deep research. The models"
  },
  {
    "startMs": 220080,
    "text": "are so [&nbsp;__&nbsp;] good at that. But if gun"
  },
  {
    "startMs": 222959,
    "text": "to my head or gun to your head, you had"
  },
  {
    "startMs": 225040,
    "text": "to pick between me and deep research and"
  },
  {
    "startMs": 226879,
    "text": "the answer had to be correct, not well"
  },
  {
    "startMs": 228879,
    "text": "written, you'd probably pick me every"
  },
  {
    "startMs": 230319,
    "text": "time. And that's the distinction between"
  },
  {
    "startMs": 233280,
    "text": "like optimizing for automation which"
  },
  {
    "startMs": 235280,
    "text": "tends to be a lot easier and a lot"
  },
  {
    "startMs": 236879,
    "text": "simpler yet machine AI keeps failing at"
  },
  {
    "startMs": 239439,
    "text": "this every year there's like an infamous"
  },
  {
    "startMs": 241599,
    "text": "failure of AI failing to even automate"
  },
  {
    "startMs": 243680,
    "text": "drive-throughs and that should be kind"
  },
  {
    "startMs": 246000,
    "text": "of shocking to people and my claim is"
  },
  {
    "startMs": 249200,
    "text": "actually for these tasks pleasing a"
  },
  {
    "startMs": 251040,
    "text": "human is nothing to do with doing the"
  },
  {
    "startMs": 252879,
    "text": "task. What you want to do is basically"
  },
  {
    "startMs": 254799,
    "text": "be an abstraction away. You want the"
  },
  {
    "startMs": 257040,
    "text": "decision to be made and then you want to"
  },
  {
    "startMs": 258639,
    "text": "get back into doing automation and code"
  },
  {
    "startMs": 261199,
    "text": "is the language of automation."
  },
  {
    "startMs": 262400,
    "text": "Therefore, you need to integrate with"
  },
  {
    "startMs": 263600,
    "text": "code. This is a lot but I will summarize"
  },
  {
    "startMs": 265600,
    "text": "it again. I get the irony that uh you"
  },
  {
    "startMs": 268479,
    "text": "know I co-created RHF and I do believe"
  },
  {
    "startMs": 270720,
    "text": "RHF is part of the problem and RHF in"
  },
  {
    "startMs": 273680,
    "text": "particular is why all AI today does"
  },
  {
    "startMs": 277040,
    "text": "require a human in the loop with an"
  },
  {
    "startMs": 278960,
    "text": "asterisk unless the task is like very"
  },
  {
    "startMs": 280800,
    "text": "trivially verifiable and can run on its"
  },
  {
    "startMs": 282720,
    "text": "own but still requires a human in the"
  },
  {
    "startMs": 284320,
    "text": "loop to verify it afterwards. It is"
  },
  {
    "startMs": 286320,
    "text": "because we literally put it in. That is"
  },
  {
    "startMs": 288320,
    "text": "what the HF is in RLHF. And my claim is"
  },
  {
    "startMs": 291440,
    "text": "that there's a better way. You know, my"
  },
  {
    "startMs": 294080,
    "text": "summary of the state is 100% of LMS"
  },
  {
    "startMs": 296320,
    "text": "today are optimized for assistance. You"
  },
  {
    "startMs": 298400,
    "text": "might talk about RLVR, but the RLVR"
  },
  {
    "startMs": 300160,
    "text": "models are very bad to interact with if"
  },
  {
    "startMs": 302160,
    "text": "they've not been RHFed. The result is LM"
  },
  {
    "startMs": 305120,
    "text": "become too good to be true at"
  },
  {
    "startMs": 306880,
    "text": "assistance. They are just so [&nbsp;__&nbsp;]"
  },
  {
    "startMs": 308639,
    "text": "good. Like you should never pick me over"
  },
  {
    "startMs": 310160,
    "text": "chat GPT for your chat GPT queries. But"
  },
  {
    "startMs": 313039,
    "text": "for actually doing like you know run in"
  },
  {
    "startMs": 316400,
    "text": "the background cognitive work you should"
  },
  {
    "startMs": 318000,
    "text": "not trust LMS at all because they've not"
  },
  {
    "startMs": 320000,
    "text": "been optimized for that which is why"
  },
  {
    "startMs": 321600,
    "text": "they're so bad at the tasks of"
  },
  {
    "startMs": 323199,
    "text": "automation and I would also guess that"
  },
  {
    "startMs": 325199,
    "text": "automation is a lot more valuable than"
  },
  {
    "startMs": 326880,
    "text": "assistants because you can do way cooler"
  },
  {
    "startMs": 329039,
    "text": "deeper I'm going to get into this later"
  },
  {
    "startMs": 330400,
    "text": "but I think there's going to be so much"
  },
  {
    "startMs": 331680,
    "text": "stuff that we can do and we are just"
  },
  {
    "startMs": 333440,
    "text": "scratching the surface of that right now"
  },
  {
    "startMs": 335120,
    "text": "and that's the tragic thing about AI uh"
  },
  {
    "startMs": 337440,
    "text": "the state of AI for me"
  },
  {
    "startMs": 339759,
    "text": "let's talk about the evolution Um, I I"
  },
  {
    "startMs": 343440,
    "text": "can't tell if this is appropriate, but"
  },
  {
    "startMs": 344880,
    "text": "it's too late. Um, oh god, don't take"
  },
  {
    "startMs": 348000,
    "text": "pictures of this one. Um, cool. Oh, I'm"
  },
  {
    "startMs": 351840,
    "text": "getting a thumbs up. Yeah. Uh,"
  },
  {
    "startMs": 353600,
    "text": "excellent. So, we're we instead of doing"
  },
  {
    "startMs": 355680,
    "text": "RHF, we just have to do something else."
  },
  {
    "startMs": 357360,
    "text": "We need to think from first principles."
  },
  {
    "startMs": 358880,
    "text": "How do we optimize for machine native"
  },
  {
    "startMs": 360479,
    "text": "AI? You guys hopefully know the"
  },
  {
    "startMs": 361840,
    "text": "interface of Jev and that is what we are"
  },
  {
    "startMs": 363520,
    "text": "doing. more important than optimizing"
  },
  {
    "startMs": 365360,
    "text": "for its interface, we are uh rethinking"
  },
  {
    "startMs": 368720,
    "text": "how AI should be done because this is"
  },
  {
    "startMs": 370800,
    "text": "what we did again for Arlf for Sutton's"
  },
  {
    "startMs": 374000,
    "text": "bitter lesson. I'm going to loosely"
  },
  {
    "startMs": 376000,
    "text": "paraphrase to make my point easier to"
  },
  {
    "startMs": 377919,
    "text": "say is that scaling um and compute"
  },
  {
    "startMs": 381199,
    "text": "matters more than algorithms. This is"
  },
  {
    "startMs": 383039,
    "text": "why people become like scaling pill"
  },
  {
    "startMs": 384720,
    "text": "jironis and I would argue that there's"
  },
  {
    "startMs": 387199,
    "text": "an even bitter lesson than that which is"
  },
  {
    "startMs": 389759,
    "text": "data always matters more than compute."
  },
  {
    "startMs": 391600,
    "text": "We are learning that quite a bit in the"
  },
  {
    "startMs": 393280,
    "text": "field now. But it has been obvious since"
  },
  {
    "startMs": 395440,
    "text": "the beginning of time. I'll tell you"
  },
  {
    "startMs": 396960,
    "text": "when the this is the one you should be"
  },
  {
    "startMs": 398479,
    "text": "photographing. I would argue that the"
  },
  {
    "startMs": 400639,
    "text": "the bitterest lesson is doing the right"
  },
  {
    "startMs": 403120,
    "text": "task. And doing the right task is like a"
  },
  {
    "startMs": 405199,
    "text": "weird thing to think about especially"
  },
  {
    "startMs": 406720,
    "text": "for ML people because they like"
  },
  {
    "startMs": 408880,
    "text": "algorithms. They want number goes down"
  },
  {
    "startMs": 411120,
    "text": "or up or whichever is good direction."
  },
  {
    "startMs": 412560,
    "text": "They don't want to think about the real"
  },
  {
    "startMs": 413680,
    "text": "world. But this is always how you make"
  },
  {
    "startMs": 416000,
    "text": "gigantic gains in performance. This is"
  },
  {
    "startMs": 417919,
    "text": "how the big breakthroughs happen. 100%"
  },
  {
    "startMs": 420000,
    "text": "of the time it comes from this and this"
  },
  {
    "startMs": 422639,
    "text": "is true for classical RL games. This is"
  },
  {
    "startMs": 424560,
    "text": "why Sutton had his bitter lesson it"
  },
  {
    "startMs": 426800,
    "text": "which is data is derived by compute when"
  },
  {
    "startMs": 429280,
    "text": "you're doing rollouts like there's"
  },
  {
    "startMs": 430720,
    "text": "there's why would you need off policy"
  },
  {
    "startMs": 432400,
    "text": "data when playing chess or something"
  },
  {
    "startMs": 433840,
    "text": "like that and doing the right task is"
  },
  {
    "startMs": 436080,
    "text": "defined by the rules. So actually when"
  },
  {
    "startMs": 437599,
    "text": "you're doing RL you don't even need to"
  },
  {
    "startMs": 439199,
    "text": "think about the right task for"
  },
  {
    "startMs": 441039,
    "text": "pre-training we actually as a field had"
  },
  {
    "startMs": 443440,
    "text": "to do a lot of thinking about the right"
  },
  {
    "startMs": 445039,
    "text": "task. I don't know if people remember"
  },
  {
    "startMs": 446639,
    "text": "the birth days with um masked language"
  },
  {
    "startMs": 449120,
    "text": "modeling, but it took a while and it was"
  },
  {
    "startMs": 451759,
    "text": "very unclear to figure out that next pro"
  },
  {
    "startMs": 453759,
    "text": "token prediction was super [&nbsp;__&nbsp;] cool."
  },
  {
    "startMs": 456319,
    "text": "Not because it was better at masked"
  },
  {
    "startMs": 457840,
    "text": "language modeling. You know, the"
  },
  {
    "startMs": 459199,
    "text": "objectives on their own aren't special."
  },
  {
    "startMs": 461280,
    "text": "It's about what you can do with them"
  },
  {
    "startMs": 462479,
    "text": "afterwards. And it turns out next token"
  },
  {
    "startMs": 464400,
    "text": "prediction is incredible at generating"
  },
  {
    "startMs": 466000,
    "text": "strings, which is what these pre-trained"
  },
  {
    "startMs": 467680,
    "text": "models were so good at. So no level of"
  },
  {
    "startMs": 470000,
    "text": "mass language modeling even if you had"
  },
  {
    "startMs": 471440,
    "text": "today's compute and algorithms would"
  },
  {
    "startMs": 473280,
    "text": "make it as interesting as GPT3 was"
  },
  {
    "startMs": 475280,
    "text": "because it was just doing a different"
  },
  {
    "startMs": 476960,
    "text": "task and data you know the internet is"
  },
  {
    "startMs": 480000,
    "text": "obviously a giant bottleneck now we're"
  },
  {
    "startMs": 481599,
    "text": "like you know burning books or whatever"
  },
  {
    "startMs": 482960,
    "text": "in order to get more data but data"
  },
  {
    "startMs": 484720,
    "text": "obviously matters a lot more than"
  },
  {
    "startMs": 486160,
    "text": "compute and people keep like"
  },
  {
    "startMs": 487440,
    "text": "rediscovering this and figuring this"
  },
  {
    "startMs": 488800,
    "text": "out. It's extra true for making LLMs"
  },
  {
    "startMs": 491680,
    "text": "useful. um before RHF even existed the"
  },
  {
    "startMs": 495039,
    "text": "data for RHF didn't exist and you could"
  },
  {
    "startMs": 498720,
    "text": "like it was night and day I'll just show"
  },
  {
    "startMs": 500000,
    "text": "you the next slide this is the probably"
  },
  {
    "startMs": 501039,
    "text": "the relevant part this is what this was"
  },
  {
    "startMs": 502960,
    "text": "the work in the RHF paper and the high"
  },
  {
    "startMs": 505680,
    "text": "level was that even with the dumbest"
  },
  {
    "startMs": 507759,
    "text": "algorithms in the world even with a GPT2"
  },
  {
    "startMs": 510240,
    "text": "sized model you got something like GPT9"
  },
  {
    "startMs": 513519,
    "text": "level performance like real GPT9 not"
  },
  {
    "startMs": 515599,
    "text": "today's like fake GPT numbers like the"
  },
  {
    "startMs": 518000,
    "text": "numbers used to mean like a 100x scaling"
  },
  {
    "startMs": 520080,
    "text": "and effective compute for pre-training"
  },
  {
    "startMs": 521839,
    "text": "and they've given up on that because it"
  },
  {
    "startMs": 523200,
    "text": "just can't scale anymore. And still even"
  },
  {
    "startMs": 525440,
    "text": "with that it would take something like"
  },
  {
    "startMs": 526880,
    "text": "GPT7 to9 to match GPT to the model from"
  },
  {
    "startMs": 530720,
    "text": "like 2018 or whatever it was at the task"
  },
  {
    "startMs": 533680,
    "text": "of instruction following. That is kind"
  },
  {
    "startMs": 535760,
    "text": "of wild because you think big LMS are so"
  },
  {
    "startMs": 538080,
    "text": "important for this and they are but that"
  },
  {
    "startMs": 540720,
    "text": "is not the most important thing compared"
  },
  {
    "startMs": 542399,
    "text": "to even the gigantic jump of doing the"
  },
  {
    "startMs": 545040,
    "text": "right task. And spoilers this is what we"
  },
  {
    "startMs": 547040,
    "text": "did for Jeff, right? like we made this"
  },
  {
    "startMs": 548800,
    "text": "gigantic leap in the field that no one"
  },
  {
    "startMs": 550720,
    "text": "realized could be coming because we were"
  },
  {
    "startMs": 552800,
    "text": "optimizing for a task and very few"
  },
  {
    "startMs": 554560,
    "text": "people think about the right task. Um"
  },
  {
    "startMs": 557839,
    "text": "this is why Jev has the Jebniqua. Um"
  },
  {
    "startMs": 561920,
    "text": "that's a really smart bun I think. Um"
  },
  {
    "startMs": 566800,
    "text": "little bit of a pity laugh, but thank"
  },
  {
    "startMs": 568080,
    "text": "you. Um you get what you optimize for. I"
  },
  {
    "startMs": 571040,
    "text": "I'm looking for a better word than"
  },
  {
    "startMs": 572560,
    "text": "reliable. So if you guys have"
  },
  {
    "startMs": 573760,
    "text": "suggestions, let me know. But there's"
  },
  {
    "startMs": 575760,
    "text": "there's this reliability of intelligence"
  },
  {
    "startMs": 577600,
    "text": "that's really missing in LLMs. I often"
  },
  {
    "startMs": 579920,
    "text": "call it you get Einstein 95% of the"
  },
  {
    "startMs": 582640,
    "text": "time, but Mr. Bean 5% of the time. And"
  },
  {
    "startMs": 585279,
    "text": "if you can't sell which is which, how"
  },
  {
    "startMs": 586959,
    "text": "the [&nbsp;__&nbsp;] are you going to automate"
  },
  {
    "startMs": 588000,
    "text": "anything? And instead instead of like"
  },
  {
    "startMs": 590000,
    "text": "focusing on like sometimes occasional"
  },
  {
    "startMs": 592000,
    "text": "smarts that you can then verify, what if"
  },
  {
    "startMs": 593839,
    "text": "you focused on extreme extreme narrow"
  },
  {
    "startMs": 596480,
    "text": "focused reliability so you can get that"
  },
  {
    "startMs": 598560,
    "text": "smarts every time? And this doesn't show"
  },
  {
    "startMs": 600480,
    "text": "up on benchmarks. It is the reason why"
  },
  {
    "startMs": 602399,
    "text": "people love Jev. And this is also I"
  },
  {
    "startMs": 604240,
    "text": "realized in hindsight we obviously"
  },
  {
    "startMs": 605440,
    "text": "didn't prepare for a chat GPT moment for"
  },
  {
    "startMs": 607839,
    "text": "developers but it happened because"
  },
  {
    "startMs": 609920,
    "text": "people in the AI space just overpromise"
  },
  {
    "startMs": 612640,
    "text": "underdel so [&nbsp;__&nbsp;] much that having a"
  },
  {
    "startMs": 615200,
    "text": "model that actually don't do that was a"
  },
  {
    "startMs": 617120,
    "text": "surprise and it is not the speed cost or"
  },
  {
    "startMs": 619519,
    "text": "even interface. This is something that"
  },
  {
    "startMs": 622000,
    "text": "you know my team tells me I shouldn't be"
  },
  {
    "startMs": 623600,
    "text": "telling people so that we can have a"
  },
  {
    "startMs": 625040,
    "text": "long-term advantage and I just care too"
  },
  {
    "startMs": 626560,
    "text": "much like it's reliable intelligence."
  },
  {
    "startMs": 628880,
    "text": "make sick models, please. Even if we,"
  },
  {
    "startMs": 630959,
    "text": "you know, if it doesn't involve our"
  },
  {
    "startMs": 632399,
    "text": "models. Um, we see that Jev as like a"
  },
  {
    "startMs": 635760,
    "text": "foundation for smart software. I'm going"
  },
  {
    "startMs": 637279,
    "text": "to skip all over this, but I think"
  },
  {
    "startMs": 638480,
    "text": "software is going to be so [&nbsp;__&nbsp;] back."
  },
  {
    "startMs": 640800,
    "text": "I think the narrative of SAS apocalypse"
  },
  {
    "startMs": 642480,
    "text": "is obviously not going to happen, but I"
  },
  {
    "startMs": 645440,
    "text": "think that even calling it a Sasa Palooa"
  },
  {
    "startMs": 648320,
    "text": "is like an understatement, and I will"
  },
  {
    "startMs": 649839,
    "text": "get into that a bit, and you guys"
  },
  {
    "startMs": 651120,
    "text": "ideally are the perfect crowd for this."
  },
  {
    "startMs": 652560,
    "text": "I'm so excited for this part because"
  },
  {
    "startMs": 653839,
    "text": "that's the part that's new to me. Um, I"
  },
  {
    "startMs": 656959,
    "text": "want to make the distinction between"
  },
  {
    "startMs": 658640,
    "text": "smart software and just in time"
  },
  {
    "startMs": 660320,
    "text": "software. The the crazy part is AI is so"
  },
  {
    "startMs": 663200,
    "text": "transformative and yet somehow despite"
  },
  {
    "startMs": 666480,
    "text": "AI being so awesome, software has"
  },
  {
    "startMs": 668720,
    "text": "essentially unchanged in the last decade"
  },
  {
    "startMs": 670959,
    "text": "except being cheaper to write. Isn't"
  },
  {
    "startMs": 673040,
    "text": "that weird? Isn't the it weird that the"
  },
  {
    "startMs": 675279,
    "text": "best we can do in software is having a"
  },
  {
    "startMs": 677040,
    "text": "chatbot to the side that sometimes"
  },
  {
    "startMs": 679200,
    "text": "interacts with a software but cannot"
  },
  {
    "startMs": 680800,
    "text": "take expensive decisions for you. That"
  },
  {
    "startMs": 682800,
    "text": "is kind of like a a weird thing if AI"
  },
  {
    "startMs": 684800,
    "text": "was really that transformative. And I"
  },
  {
    "startMs": 686560,
    "text": "think the problem is we're focusing too"
  },
  {
    "startMs": 688160,
    "text": "much on coding agents and not enough"
  },
  {
    "startMs": 690560,
    "text": "about putting the intelligence into the"
  },
  {
    "startMs": 692160,
    "text": "inner loop, which I'm going to get more"
  },
  {
    "startMs": 693519,
    "text": "into a bit. And people are jeving."
  },
  {
    "startMs": 695920,
    "text": "There's been tons of successes. We don't"
  },
  {
    "startMs": 697920,
    "text": "need to flex this. There's so many cool"
  },
  {
    "startMs": 699519,
    "text": "things. But my AIE secret is going to be"
  },
  {
    "startMs": 703120,
    "text": "that the field isn't jebbing hard"
  },
  {
    "startMs": 705040,
    "text": "enough. And this is what Swix said that"
  },
  {
    "startMs": 706560,
    "text": "I can make a call to action for you guys"
  },
  {
    "startMs": 708240,
    "text": "that's unique. And hopefully I have"
  },
  {
    "startMs": 709360,
    "text": "enough time for this. I think the world"
  },
  {
    "startMs": 711440,
    "text": "still has LLM Stockholm syndrome. We are"
  },
  {
    "startMs": 713760,
    "text": "still thinking from a really LLM pill"
  },
  {
    "startMs": 715920,
    "text": "place which is I think tragic you know"
  },
  {
    "startMs": 719040,
    "text": "like I I I love the software saying the"
  },
  {
    "startMs": 721440,
    "text": "tools you use change the way you think"
  },
  {
    "startMs": 723279,
    "text": "and I think that it's changed the way we"
  },
  {
    "startMs": 724800,
    "text": "think way too much and we're not really"
  },
  {
    "startMs": 727279,
    "text": "rethinking software like I think the"
  },
  {
    "startMs": 729360,
    "text": "field needs. I think that we could just"
  },
  {
    "startMs": 730720,
    "text": "do be doing so much more. Hopefully I"
  },
  {
    "startMs": 732639,
    "text": "can try to do my best for my remaining"
  },
  {
    "startMs": 734399,
    "text": "time to convince you guys of that. Um so"
  },
  {
    "startMs": 737040,
    "text": "let's Jev to the future. Um"
  },
  {
    "startMs": 740639,
    "text": "I an analogy I love uh bear with me is"
  },
  {
    "startMs": 743600,
    "text": "about LEDs. Before there was the blue"
  },
  {
    "startMs": 746639,
    "text": "LED, there was red and green LEDs. These"
  },
  {
    "startMs": 748639,
    "text": "were really limited in capability and"
  },
  {
    "startMs": 750720,
    "text": "they were used for like simple"
  },
  {
    "startMs": 751920,
    "text": "indicators and you know like clocks and"
  },
  {
    "startMs": 753839,
    "text": "stuff like that. And when the blue LED"
  },
  {
    "startMs": 756560,
    "text": "was made discovered, I actually don't"
  },
  {
    "startMs": 758560,
    "text": "really know how things outside of"
  },
  {
    "startMs": 759839,
    "text": "computer science work. Um, but when it"
  },
  {
    "startMs": 762079,
    "text": "existed, I guess, um, it enabled not"
  },
  {
    "startMs": 765440,
    "text": "just like 50% more use cases, which what"
  },
  {
    "startMs": 767760,
    "text": "you'd expect about like changing colors"
  },
  {
    "startMs": 769120,
    "text": "from two to three, it enabled so many"
  },
  {
    "startMs": 771360,
    "text": "more. Like now LEDs could be used for"
  },
  {
    "startMs": 773200,
    "text": "white light. It could be used for like"
  },
  {
    "startMs": 774560,
    "text": "the displays that we all use right now."
  },
  {
    "startMs": 776320,
    "text": "And it created a revolution in lighting."
  },
  {
    "startMs": 778639,
    "text": "And it's it's it's unfathomable how much"
  },
  {
    "startMs": 781200,
    "text": "bigger the capabilities of LEDs were"
  },
  {
    "startMs": 784240,
    "text": "just ex, you know, adding a new"
  },
  {
    "startMs": 786000,
    "text": "primitive that composes together with a"
  },
  {
    "startMs": 788079,
    "text": "old primitive. So you can see where I'm"
  },
  {
    "startMs": 789600,
    "text": "going here maybe. And I think it's a"
  },
  {
    "startMs": 792000,
    "text": "similar story for software. Software is"
  },
  {
    "startMs": 794800,
    "text": "so [&nbsp;__&nbsp;] incredible. Ignore AI for the"
  },
  {
    "startMs": 796959,
    "text": "last, you know, half decade. But"
  },
  {
    "startMs": 798320,
    "text": "software has done so many insanely cool"
  },
  {
    "startMs": 801120,
    "text": "things. We've done a lot of stuff, you"
  },
  {
    "startMs": 803200,
    "text": "know, like do I love using SAS? Not"
  },
  {
    "startMs": 806720,
    "text": "really. But does it solve problems? Abso"
  },
  {
    "startMs": 809040,
    "text": "[&nbsp;__&nbsp;] lutely. And it's incredible how"
  },
  {
    "startMs": 811040,
    "text": "much we've done with three"
  },
  {
    "startMs": 812209,
    "text": "[clears throat] fundamental logic gates."
  },
  {
    "startMs": 813760,
    "text": "You know like we are doing this with"
  },
  {
    "startMs": 815200,
    "text": "like simple binary operations that can"
  },
  {
    "startMs": 817519,
    "text": "do things like greater than less than"
  },
  {
    "startMs": 819200,
    "text": "equal. It's kind of nuts how much"
  },
  {
    "startMs": 821519,
    "text": "software has been built from such simple"
  },
  {
    "startMs": 823760,
    "text": "primitives. And my claim is that what we"
  },
  {
    "startMs": 827200,
    "text": "are doing with AI now when we're making"
  },
  {
    "startMs": 829440,
    "text": "our intelligence machine native this is"
  },
  {
    "startMs": 831600,
    "text": "kind of conceptually adding a fourth"
  },
  {
    "startMs": 833920,
    "text": "semantic logic aid. What if we can"
  },
  {
    "startMs": 836480,
    "text": "program with common sense with intention"
  },
  {
    "startMs": 838959,
    "text": "with human understanding? Like do we"
  },
  {
    "startMs": 840880,
    "text": "still need reexes with like crazy things"
  },
  {
    "startMs": 842800,
    "text": "or can we just say like we want a Gmail"
  },
  {
    "startMs": 844560,
    "text": "email address or a real looking one?"
  },
  {
    "startMs": 846240,
    "text": "There's so much that we can do that"
  },
  {
    "startMs": 847839,
    "text": "we're going to need to push the you know"
  },
  {
    "startMs": 849360,
    "text": "push the frontiers of cheap, smart, fast"
  },
  {
    "startMs": 852480,
    "text": "in order to get done. But I think that"
  },
  {
    "startMs": 854000,
    "text": "it's going to be like so many things"
  },
  {
    "startMs": 855360,
    "text": "here and I think just like the blue LED"
  },
  {
    "startMs": 857680,
    "text": "it's going to be way bigger than just"
  },
  {
    "startMs": 859519,
    "text": "increasing the representational ability"
  },
  {
    "startMs": 861839,
    "text": "of software by 33% because it's adding"
  },
  {
    "startMs": 864560,
    "text": "effectively another kind of sort of"
  },
  {
    "startMs": 866560,
    "text": "logic gate. And this sci-fi is the stuff"
  },
  {
    "startMs": 868800,
    "text": "that I think will be impossible to"
  },
  {
    "startMs": 870560,
    "text": "predict just like who could predict how"
  },
  {
    "startMs": 872560,
    "text": "powerful the internet would have been"
  },
  {
    "startMs": 874160,
    "text": "because it requires it's like much more"
  },
  {
    "startMs": 875760,
    "text": "grassroots and decentralized and it"
  },
  {
    "startMs": 877519,
    "text": "requires people to be empowered to do"
  },
  {
    "startMs": 879120,
    "text": "insane stuff. Um I call this software"
  },
  {
    "startMs": 882880,
    "text": "research. Um that's not a joke. Um but"
  },
  {
    "startMs": 886800,
    "text": "uh I I think that there's going to be"
  },
  {
    "startMs": 888639,
    "text": "like an explosion of what software can"
  },
  {
    "startMs": 891199,
    "text": "do and we will want people to be"
  },
  {
    "startMs": 893440,
    "text": "exploring those boundaries of software."
  },
  {
    "startMs": 895680,
    "text": "Not necessarily because it's valuable,"
  },
  {
    "startMs": 897680,
    "text": "but because they can because we have"
  },
  {
    "startMs": 899600,
    "text": "like new frontiers to explore and this"
  },
  {
    "startMs": 901680,
    "text": "is what the early internet was like."
  },
  {
    "startMs": 903199,
    "text": "Should all of those websites existed?"
  },
  {
    "startMs": 905760,
    "text": "Probably not. But like it was very cool"
  },
  {
    "startMs": 908160,
    "text": "and we found a lot of awesome stuff from"
  },
  {
    "startMs": 909760,
    "text": "that. And I think we we, you know, we"
  },
  {
    "startMs": 911519,
    "text": "have another opportunity for that. So we"
  },
  {
    "startMs": 913199,
    "text": "should not just keep living in this"
  },
  {
    "startMs": 915120,
    "text": "little, you know, fish tank that we are"
  },
  {
    "startMs": 917120,
    "text": "the victims of open anthropic on"
  },
  {
    "startMs": 919040,
    "text": "whatever they ship and we just have to"
  },
  {
    "startMs": 920560,
    "text": "be like LM rappers. So I'm going to"
  },
  {
    "startMs": 923040,
    "text": "pitch software research as a direction"
  },
  {
    "startMs": 924800,
    "text": "here. The theme is going to be real"
  },
  {
    "startMs": 927279,
    "text": "software engineering, not just vibe"
  },
  {
    "startMs": 929600,
    "text": "coding. There's it's going to be like"
  },
  {
    "startMs": 931199,
    "text": "the kind of stuff that everyone has"
  },
  {
    "startMs": 932480,
    "text": "known if you've been a software"
  },
  {
    "startMs": 933680,
    "text": "engineer. Decomposition, breaking tasks"
  },
  {
    "startMs": 936079,
    "text": "down. We need to think about stages,"
  },
  {
    "startMs": 938000,
    "text": "ideally ones that can be verified so we"
  },
  {
    "startMs": 939839,
    "text": "have real encapsulation. And we need to"
  },
  {
    "startMs": 942320,
    "text": "actually do state management instead of"
  },
  {
    "startMs": 944240,
    "text": "assuming everything is global state"
  },
  {
    "startMs": 945920,
    "text": "within a KV cache. And the theme is just"
  },
  {
    "startMs": 947760,
    "text": "going to be like real [&nbsp;__&nbsp;]"
  },
  {
    "startMs": 949279,
    "text": "engineering by people who care a lot"
  },
  {
    "startMs": 951440,
    "text": "about the end result. It doesn't have to"
  },
  {
    "startMs": 953279,
    "text": "be valuable. Ideally, it's sick as [&nbsp;__&nbsp;]"
  },
  {
    "startMs": 955440,
    "text": "Um, and you know, we should be pushing"
  },
  {
    "startMs": 957040,
    "text": "the limits."
  },
  {
    "startMs": 958720,
    "text": "I'm going to go into a couple of like my"
  },
  {
    "startMs": 960480,
    "text": "favorite ideas and I want to figure out"
  },
  {
    "startMs": 962399,
    "text": "how to encourage these in the world. So,"
  },
  {
    "startMs": 964959,
    "text": "this is my first step at this. I'm just"
  },
  {
    "startMs": 966880,
    "text": "going to talk about it. We have raised a"
  },
  {
    "startMs": 968880,
    "text": "bunch of money. So, I would like to"
  },
  {
    "startMs": 969839,
    "text": "figure out how to fund this, but we"
  },
  {
    "startMs": 971040,
    "text": "don't have the operations help to do"
  },
  {
    "startMs": 972480,
    "text": "that. But here are just some ideas out"
  },
  {
    "startMs": 974399,
    "text": "there. Um, I think agents are they're"
  },
  {
    "startMs": 978800,
    "text": "very cool, but like they're they're so"
  },
  {
    "startMs": 981279,
    "text": "inflexible in their design because of"
  },
  {
    "startMs": 983040,
    "text": "the tyranny of the KV cache. I don't"
  },
  {
    "startMs": 984880,
    "text": "have the time for this if this timer is"
  },
  {
    "startMs": 986399,
    "text": "to be believed. But believe me that"
  },
  {
    "startMs": 988320,
    "text": "there's so much that people have tried"
  },
  {
    "startMs": 990079,
    "text": "to do but failed at because it ends up"
  },
  {
    "startMs": 991920,
    "text": "becoming inefficient because it fights"
  },
  {
    "startMs": 993680,
    "text": "against the KV cache. But I think"
  },
  {
    "startMs": 995519,
    "text": "there's a lot more to do. Why is memory"
  },
  {
    "startMs": 997600,
    "text": "o, right? Like why can't we have all of"
  },
  {
    "startMs": 1000240,
    "text": "our history of coding agents always"
  },
  {
    "startMs": 1002240,
    "text": "included in every new call? That is how"
  },
  {
    "startMs": 1005040,
    "text": "continual learning works, right? And we"
  },
  {
    "startMs": 1007600,
    "text": "have the memory available. It's just"
  },
  {
    "startMs": 1008959,
    "text": "that we don't have the smarts to map"
  },
  {
    "startMs": 1010800,
    "text": "produce over the memory to find out like"
  },
  {
    "startMs": 1012880,
    "text": "what you should what you should"
  },
  {
    "startMs": 1014160,
    "text": "remember, what you shouldn't. Um, I"
  },
  {
    "startMs": 1016399,
    "text": "think there's many ways to do login or"
  },
  {
    "startMs": 1018000,
    "text": "even better lookups. Why when you're"
  },
  {
    "startMs": 1020079,
    "text": "making many agents at once, why can't"
  },
  {
    "startMs": 1021920,
    "text": "they not share state? Why do they need"
  },
  {
    "startMs": 1023680,
    "text": "to rediscover the same patterns of the"
  },
  {
    "startMs": 1025360,
    "text": "codebase over and over and over again?"
  },
  {
    "startMs": 1027280,
    "text": "And why can't non-naive routing and sub"
  },
  {
    "startMs": 1029520,
    "text": "aents work? Like even naive routing"
  },
  {
    "startMs": 1030959,
    "text": "doesn't work, but like why can't we do"
  },
  {
    "startMs": 1032799,
    "text": "these things? And why does like the you"
  },
  {
    "startMs": 1034240,
    "text": "know the the big labs keep going back"
  },
  {
    "startMs": 1035520,
    "text": "and forth on whether or not sub aents"
  },
  {
    "startMs": 1037120,
    "text": "work. Um I think that like from an"
  },
  {
    "startMs": 1039839,
    "text": "engineering perspective there's elegance"
  },
  {
    "startMs": 1041600,
    "text": "to it but they fight against the KV"
  },
  {
    "startMs": 1043280,
    "text": "cache and I think that if we design"
  },
  {
    "startMs": 1044880,
    "text": "agents around that we could do so much"
  },
  {
    "startMs": 1046880,
    "text": "more. This one's the the probably the"
  },
  {
    "startMs": 1049760,
    "text": "most vague of my directions but I feel"
  },
  {
    "startMs": 1051760,
    "text": "very strongly about it. Um the world is"
  },
  {
    "startMs": 1054240,
    "text": "full of principal agent problems. This"
  },
  {
    "startMs": 1056240,
    "text": "is when you have like you know ulterior"
  },
  {
    "startMs": 1057760,
    "text": "motives between this is not a coding"
  },
  {
    "startMs": 1059120,
    "text": "agent but like any agent. The the"
  },
  {
    "startMs": 1061120,
    "text": "classic example is citizens versus"
  },
  {
    "startMs": 1063039,
    "text": "elected officials. Now, I don't want to"
  },
  {
    "startMs": 1064880,
    "text": "make a too political claim here, but"
  },
  {
    "startMs": 1066720,
    "text": "like there's misalignment in all levels"
  },
  {
    "startMs": 1069039,
    "text": "of society. And this results in like"
  },
  {
    "startMs": 1071120,
    "text": "waste, fraud, bad things, and it it"
  },
  {
    "startMs": 1074480,
    "text": "results in for a society that wants"
  },
  {
    "startMs": 1076400,
    "text": "fairness. You need to overemphasize the"
  },
  {
    "startMs": 1078480,
    "text": "letter over the spirit, which then"
  },
  {
    "startMs": 1080160,
    "text": "results in loopholes and play uh"
  },
  {
    "startMs": 1082480,
    "text": "payto-in dynamics in the world. And I"
  },
  {
    "startMs": 1084799,
    "text": "think the root of that problem is we"
  },
  {
    "startMs": 1086080,
    "text": "can't encode the spirit. Just like with"
  },
  {
    "startMs": 1088080,
    "text": "software, we're constrained to encoding"
  },
  {
    "startMs": 1090240,
    "text": "bits instead of what we actually meant"
  },
  {
    "startMs": 1092000,
    "text": "in that situation. And I think so many"
  },
  {
    "startMs": 1094640,
    "text": "organizations, society could be"
  },
  {
    "startMs": 1096160,
    "text": "transformed by really encoding the"
  },
  {
    "startMs": 1097679,
    "text": "spirit all over the place. Obviously,"
  },
  {
    "startMs": 1100320,
    "text": "real-time intelligence security, can we"
  },
  {
    "startMs": 1103039,
    "text": "prevent the OpenAI hugging face sandbox,"
  },
  {
    "startMs": 1105120,
    "text": "blah blah blah. Um, why did why couldn't"
  },
  {
    "startMs": 1107679,
    "text": "they prevent it? It is not because they"
  },
  {
    "startMs": 1109679,
    "text": "didn't have the intelligence. I would"
  },
  {
    "startMs": 1111039,
    "text": "argue it's a lot easier to figure out"
  },
  {
    "startMs": 1112559,
    "text": "they're doing a naughty thing than to"
  },
  {
    "startMs": 1114160,
    "text": "actually do the naughty thing. It is"
  },
  {
    "startMs": 1115840,
    "text": "because it was not fast, cheap, and"
  },
  {
    "startMs": 1117760,
    "text": "reliable enough to actually run real"
  },
  {
    "startMs": 1120160,
    "text": "time. And what happens when every"
  },
  {
    "startMs": 1122640,
    "text": "corporation has their own real-time"
  },
  {
    "startMs": 1124640,
    "text": "security that's actually smart? How cool"
  },
  {
    "startMs": 1126640,
    "text": "would that [&nbsp;__&nbsp;] be? I can't talk"
  },
  {
    "startMs": 1128960,
    "text": "about generative UIs because people are"
  },
  {
    "startMs": 1130799,
    "text": "going to be talking about it afterwards."
  },
  {
    "startMs": 1132320,
    "text": "I don't want to steal their thunder, but"
  },
  {
    "startMs": 1133679,
    "text": "it's cool as [&nbsp;__&nbsp;] Trust me. Um,"
  },
  {
    "startMs": 1136799,
    "text": "like semantic dependent typing. Why"
  },
  {
    "startMs": 1139440,
    "text": "can't we have like a type system that"
  },
  {
    "startMs": 1142480,
    "text": "actually is semantic, right? like why"
  },
  {
    "startMs": 1144720,
    "text": "can't we like declare what local"
  },
  {
    "startMs": 1147679,
    "text": "properties we want so we don't have to"
  },
  {
    "startMs": 1149600,
    "text": "read a chain of dependencies deep down"
  },
  {
    "startMs": 1151919,
    "text": "these all should be possible and imagine"
  },
  {
    "startMs": 1153840,
    "text": "if you could like run a jevlike model in"
  },
  {
    "startMs": 1156320,
    "text": "your lint checker to check if you're"
  },
  {
    "startMs": 1157840,
    "text": "it's all type safe and it could you know"
  },
  {
    "startMs": 1160160,
    "text": "uphold semantic properties it would be"
  },
  {
    "startMs": 1162400,
    "text": "so [&nbsp;__&nbsp;] cool there's there was a"
  },
  {
    "startMs": 1165600,
    "text": "great Twitter video on um like a voice"
  },
  {
    "startMs": 1168240,
    "text": "interface in general I think new"
  },
  {
    "startMs": 1170320,
    "text": "interfaces are so exciting this is to me"
  },
  {
    "startMs": 1172799,
    "text": "the return of like this going to show by"
  },
  {
    "startMs": 1174400,
    "text": "agent software like do what I mean. This"
  },
  {
    "startMs": 1176559,
    "text": "used to be like a popular thing but it"
  },
  {
    "startMs": 1178160,
    "text": "never really worked because you it was"
  },
  {
    "startMs": 1180080,
    "text": "it was too many hard-coded rules to"
  },
  {
    "startMs": 1181760,
    "text": "really figure out what I mean. Um"
  },
  {
    "startMs": 1184400,
    "text": "there's stopping slop obviously a big"
  },
  {
    "startMs": 1186720,
    "text": "theme of here but like we definitely"
  },
  {
    "startMs": 1188640,
    "text": "believe this. This is my props. I have"
  },
  {
    "startMs": 1190080,
    "text": "to we have a stop geni baseball bat for"
  },
  {
    "startMs": 1193120,
    "text": "type safe. It is very very awesome. I"
  },
  {
    "startMs": 1196400,
    "text": "want to talk about the root of slop. Um"
  },
  {
    "startMs": 1198480,
    "text": "the the root of slop is that AI is"
  },
  {
    "startMs": 1200240,
    "text": "unreliable. Therefore, you as a company"
  },
  {
    "startMs": 1202880,
    "text": "or you as you know like someone who's"
  },
  {
    "startMs": 1204559,
    "text": "doing stuff, it can't be used for"
  },
  {
    "startMs": 1206080,
    "text": "decisions that costs you stuff. But it's"
  },
  {
    "startMs": 1208559,
    "text": "actually totally fine to push the costs"
  },
  {
    "startMs": 1210160,
    "text": "into other people, which is horrible."
  },
  {
    "startMs": 1211840,
    "text": "You know, externalized costs and then"
  },
  {
    "startMs": 1213440,
    "text": "this is what ends up becoming slop. You"
  },
  {
    "startMs": 1215520,
    "text": "know, if people actually read the stuff"
  },
  {
    "startMs": 1216960,
    "text": "and filter it and like care about the"
  },
  {
    "startMs": 1218640,
    "text": "quality, it becomes not slop. But now"
  },
  {
    "startMs": 1221039,
    "text": "the world is full of this because"
  },
  {
    "startMs": 1222559,
    "text": "there's an asymmetric advantage for"
  },
  {
    "startMs": 1224320,
    "text": "people who are like bad actors and don't"
  },
  {
    "startMs": 1226000,
    "text": "give a [&nbsp;__&nbsp;] So, I think that we can"
  },
  {
    "startMs": 1228000,
    "text": "save the world from at least inbound"
  },
  {
    "startMs": 1229760,
    "text": "hell among many other things and that"
  },
  {
    "startMs": 1231440,
    "text": "would be very cool. Um, and even"
  },
  {
    "startMs": 1234400,
    "text": "something I think is super sick is"
  },
  {
    "startMs": 1236320,
    "text": "cursed software. Like why if you guys"
  },
  {
    "startMs": 1239039,
    "text": "ever seen the subreddit bad UI battles,"
  },
  {
    "startMs": 1241120,
    "text": "it is so [&nbsp;__&nbsp;] cool. Even though they"
  },
  {
    "startMs": 1243520,
    "text": "make the horrible interfaces like this"
  },
  {
    "startMs": 1245440,
    "text": "whole game of like would you like to"
  },
  {
    "startMs": 1246559,
    "text": "delete your account becomes you know a"
  },
  {
    "startMs": 1248559,
    "text": "random thing. And and I actually think"
  },
  {
    "startMs": 1251200,
    "text": "doing things because we can is what"
  },
  {
    "startMs": 1253360,
    "text": "software engineering is about because"
  },
  {
    "startMs": 1255039,
    "text": "who knows maybe 1% of these ideas"
  },
  {
    "startMs": 1257520,
    "text": "actually lead to something super"
  },
  {
    "startMs": 1259440,
    "text": "awesome. And I think like really"
  },
  {
    "startMs": 1261360,
    "text": "cultivating even bad ideas is kind of"
  },
  {
    "startMs": 1263520,
    "text": "what being like this you know playful"
  },
  {
    "startMs": 1265280,
    "text": "hacker is all about and generally just"
  },
  {
    "startMs": 1269120,
    "text": "automating everything. This is what the"
  },
  {
    "startMs": 1271520,
    "text": "followup this is this is a slop image"
  },
  {
    "startMs": 1273360,
    "text": "but uh sorry logo is a little bit wrong"
  },
  {
    "startMs": 1275039,
    "text": "but don't stare at that. That's a little"
  },
  {
    "startMs": 1276480,
    "text": "weird. Um, this is what a lot of the"
  },
  {
    "startMs": 1279280,
    "text": "follow-up talks are going to be about."
  },
  {
    "startMs": 1281120,
    "text": "Thank you. Um, and if you're a software"
  },
  {
    "startMs": 1284240,
    "text": "engineer, you almost certainly think in"
  },
  {
    "startMs": 1286640,
    "text": "the world of bits. If you know anything"
  },
  {
    "startMs": 1288000,
    "text": "about software, uh, and you've"
  },
  {
    "startMs": 1289919,
    "text": "interacted with software, you know that"
  },
  {
    "startMs": 1292400,
    "text": "there are things software can do and"
  },
  {
    "startMs": 1294400,
    "text": "software cannot do because that is just"
  },
  {
    "startMs": 1296400,
    "text": "only obvious, right? Like there's some"
  },
  {
    "startMs": 1298880,
    "text": "things that you know you need to do"
  },
  {
    "startMs": 1300559,
    "text": "because that is not compatible with"
  },
  {
    "startMs": 1302080,
    "text": "software. But what if those constraints"
  },
  {
    "startMs": 1304400,
    "text": "were no longer true? Maybe not today,"
  },
  {
    "startMs": 1306240,
    "text": "but maybe in like six months or a year."
  },
  {
    "startMs": 1308320,
    "text": "What couldn't we automate? And that's"
  },
  {
    "startMs": 1310320,
    "text": "kind of where I want to leave this whole"
  },
  {
    "startMs": 1311679,
    "text": "thing because I think there is very"
  },
  {
    "startMs": 1314000,
    "text": "likely to be a future where it's not a"
  },
  {
    "startMs": 1316080,
    "text": "question of whether or not we can"
  },
  {
    "startMs": 1317360,
    "text": "automate something. It's going to be a"
  },
  {
    "startMs": 1318799,
    "text": "question of like how much work is it,"
  },
  {
    "startMs": 1320159,
    "text": "which is how it is right now for the"
  },
  {
    "startMs": 1321600,
    "text": "stuff that can be automated. And that"
  },
  {
    "startMs": 1323280,
    "text": "would be cool as [&nbsp;__&nbsp;] So, hell yeah."
  },
  {
    "startMs": 1326000,
    "text": "Thank you. Um,"
  },
  {
    "startMs": 1329440,
    "text": "I'm super overtime, but thank you to the"
  },
  {
    "startMs": 1332799,
    "text": "early believers. You know, we had no"
  },
  {
    "startMs": 1334799,
    "text": "idea so many people would resonate with"
  },
  {
    "startMs": 1336480,
    "text": "this. We thought we were the insane"
  },
  {
    "startMs": 1338080,
    "text": "ones, but it turns out like we were"
  },
  {
    "startMs": 1339760,
    "text": "actually the majority of the world who"
  },
  {
    "startMs": 1341280,
    "text": "realized AI could be so much more. Thank"
  },
  {
    "startMs": 1344000,
    "text": "you to like the people who are like"
  },
  {
    "startMs": 1345280,
    "text": "really realistic about reliability."
  },
  {
    "startMs": 1347200,
    "text": "Everyone knows I hate benchmarks and a"
  },
  {
    "startMs": 1348960,
    "text": "lot of other people are really screaming"
  },
  {
    "startMs": 1350320,
    "text": "out like benchmarks [&nbsp;__&nbsp;] suck."
  },
  {
    "startMs": 1352080,
    "text": "They're actually they hurt the people"
  },
  {
    "startMs": 1354480,
    "text": "who do not have time for anything other"
  },
  {
    "startMs": 1356000,
    "text": "than benchmarks most. It's tragic. And"
  },
  {
    "startMs": 1358000,
    "text": "for the people who are pushing the"
  },
  {
    "startMs": 1359280,
    "text": "limits of software, like [&nbsp;__&nbsp;] yes. and"
  },
  {
    "startMs": 1362159,
    "text": "we want to be there with you helping you"
  },
  {
    "startMs": 1363919,
    "text": "all do that. The the dreamers and the"
  },
  {
    "startMs": 1365919,
    "text": "people who don't want to be limited by"
  },
  {
    "startMs": 1368240,
    "text": "everything that has happened so far, we"
  },
  {
    "startMs": 1371120,
    "text": "want to really fight for you and we want"
  },
  {
    "startMs": 1372960,
    "text": "this to be as reliable as a fourth logic"
  },
  {
    "startMs": 1376080,
    "text": "age and that is a very high bar and we"
  },
  {
    "startMs": 1378559,
    "text": "have like a lot of work to do. So, thank"
  },
  {
    "startMs": 1380720,
    "text": "you. Thank you."
  }
]
