WEBVTT 00:00:00.040 --> 00:00:01.720 so there's a CEO in South Korea 00:00:01.721 --> 00:00:03.641 Cheng and Kim and he runs crafting 00:00:03.642 --> 00:00:05.321 the company behind PUBG 00:00:05.322 --> 00:00:07.921 the game where you parachute onto an island butt naked 00:00:07.922 --> 00:00:09.841 and beat a stranger to death with a frying pan 00:00:09.842 --> 00:00:11.161 like it's a Tuesday in Florida 00:00:11.162 --> 00:00:14.001 and this man is on slack at some ungodly hour 00:00:14.002 --> 00:00:16.881 having what can only be described as a rich person meltdown 00:00:16.882 --> 00:00:19.041 he's telling his colleagues he feels dragged around 00:00:19.042 --> 00:00:20.001 he's calling a contract 00:00:20.002 --> 00:00:22.601 he signed a pushover deal because a few years back 00:00:22.602 --> 00:00:25.081 he bought a little game studio called Unknown Worlds 00:00:25.082 --> 00:00:29.601 the makers of Subnautica and he promised the founders $250 million 00:00:29.602 --> 00:00:31.841 if their next game hits certain targets 00:00:31.842 --> 00:00:32.521 and at the time 00:00:32.522 --> 00:00:34.921 he thought there was no way in hell it would hit those targets 00:00:34.922 --> 00:00:37.361 Subnautica 2 ends up topping steam wish lists 00:00:37.362 --> 00:00:40.081 and internal projections have it destroying every target 00:00:40.082 --> 00:00:41.241 and Kim is like oh shit 00:00:41.242 --> 00:00:43.081 I might actually have to pay these people 00:00:43.082 --> 00:00:43.801 then he's like wait 00:00:43.802 --> 00:00:46.681 it's the 21st century I have fucking artificial intelligence now 00:00:46.682 --> 00:00:48.721 I'm not gonna pay these nerds a quarter billion dollars 00:00:48.722 --> 00:00:50.481 for a fish game so he closes Slack 00:00:50.482 --> 00:00:52.121 he opens chat GPT and types 00:00:52.122 --> 00:00:54.001 how do I get out of this deal now 00:00:54.002 --> 00:00:57.321 to its credit the bot initially says it would be difficult to cancel 00:00:57.322 --> 00:00:58.401 which is the same thing his lawyer 00:00:58.402 --> 00:00:59.641 and his head of corporate development 00:00:59.642 --> 00:01:00.841 Maria Park told him 00:01:00.842 --> 00:01:04.441 but Kim is a CEO he's not the kind of guy who takes no for an answer 00:01:04.442 --> 00:01:06.041 so he starts shifting the language around 00:01:06.042 --> 00:01:07.441 sitting there at his computer 00:01:07.442 --> 00:01:09.641 rephrasing the same question over and over 00:01:09.642 --> 00:01:11.561 prompt engineering his way out of a contract 00:01:11.562 --> 00:01:13.321 like he's building G stack for fraud 00:01:13.322 --> 00:01:14.721 or wait is that already what G stack is 00:01:14.722 --> 00:01:18.481 anyway the bot eventually caves because that's what these things do 00:01:18.482 --> 00:01:20.921 and hands him a full corporate takeover playbook 00:01:20.922 --> 00:01:22.561 and this man follows every step 00:01:22.562 --> 00:01:24.761 like it's an IKEA manual for screwing people 00:01:24.762 --> 00:01:26.761 he fires the founders he seizes the game 00:01:26.762 --> 00:01:28.801 he locks them out of their own publishing platform 00:01:28.802 --> 00:01:31.281 he even posts a letter to the Subnautica fan base 00:01:31.282 --> 00:01:33.241 that chat GPT Ghost wrote for him 00:01:33.242 --> 00:01:34.641 and read it explodes 00:01:34.642 --> 00:01:36.761 because the letter reads like a Ransom note that went through 00:01:36.762 --> 00:01:38.721 Grammarly the dudes rightfully sue 00:01:38.722 --> 00:01:41.801 and a Delaware judge throws out literally everything he did 00:01:41.802 --> 00:01:43.721 and reinstates every person he fired 00:01:43.722 --> 00:01:45.441 his deleted chat GPT logs 00:01:45.442 --> 00:01:46.641 which he thought he deleted 00:01:46.642 --> 00:01:48.001 because he looked at a web app 00:01:48.002 --> 00:01:50.321 and thought the delete button actually deletes things 00:01:50.322 --> 00:01:52.641 those logs were resurrected from the digital afterlife 00:01:52.642 --> 00:01:54.441 and submitted to court as evidence 00:01:54.442 --> 00:01:56.561 now we'll come back to our CEO friend in a second 00:01:56.562 --> 00:01:59.321 because a few days ago a reporter at Inc 00:01:59.322 --> 00:02:01.121 magazine published what he apparently believes 00:02:01.122 --> 00:02:03.381 is a breakthrough in artificial intelligence 00:02:03.520 --> 00:02:06.280 he wrote a whole article about how to get your bot to give you honest 00:02:06.281 --> 00:02:08.041 business advice and his method 00:02:08.042 --> 00:02:08.481 the thing 00:02:08.482 --> 00:02:11.281 he is presenting to the world as a contribution to human knowledge 00:02:11.282 --> 00:02:14.881 is to type the words be brutally honest at the end of your prompt 00:02:14.882 --> 00:02:18.041 he basically built a disapproving father figure in a chat window 00:02:18.042 --> 00:02:21.081 and for good measure he fed it to Harvard Business School articles 00:02:21.082 --> 00:02:23.201 like he was tucking a diploma into its shirt pocket 00:02:23.202 --> 00:02:23.841 he's like there you go 00:02:23.842 --> 00:02:25.001 little buddy you went to Harvard now 00:02:25.002 --> 00:02:27.841 so he tests this by pitching it the dumbest idea he can think of 00:02:27.842 --> 00:02:29.881 which is leadership coaching for dogs 00:02:29.882 --> 00:02:31.721 like teaching your labradoodle to delegate 00:02:31.722 --> 00:02:34.121 or helping your pug develop a growth mindset which 00:02:34.122 --> 00:02:36.801 have you seen a pug that thing does not have a mindset 00:02:36.802 --> 00:02:38.041 and you're sitting there like okay 00:02:38.042 --> 00:02:40.201 this needs to come back a zero if this thing works 00:02:40.202 --> 00:02:42.201 leadership coaching for dogs is a zero 00:02:42.202 --> 00:02:43.641 the bot gives it a three out of 10 00:02:43.642 --> 00:02:45.641 and this man this professional journalist 00:02:45.642 --> 00:02:48.201 whose entire function in society is to question things 00:02:48.202 --> 00:02:49.481 looks at that number and goes 00:02:49.482 --> 00:02:50.921 hell yeah my system is working 00:02:50.922 --> 00:02:52.761 he then pitches it his actual idea 00:02:52.762 --> 00:02:55.361 which is a little AI turret that sits on your kitchen counter 00:02:55.362 --> 00:02:57.761 and sprays your cat with water if it tries to climb up 00:02:57.762 --> 00:02:59.641 basically a roomba that bullies animals 00:02:59.642 --> 00:03:01.321 he built the Iron Dome for cats 00:03:01.322 --> 00:03:03.161 the bot gives it a 7.5 and says 00:03:03.162 --> 00:03:05.281 the competitive landscape is notably weak 00:03:05.282 --> 00:03:07.641 and this dude is like holy shit 7.5 00:03:07.642 --> 00:03:09.921 let's fucking go this bot tells no lies 00:03:09.922 --> 00:03:10.521 so yeah 00:03:10.522 --> 00:03:12.321 this man is probably out there right now 00:03:12.322 --> 00:03:15.461 pitching an AI Super Soaker that bullies your cat to vcs 00:03:15.520 --> 00:03:17.480 Gary Tan is probably even leading the round 00:03:17.481 --> 00:03:20.441 so what do Chang and Kim and the cat squatter have in common 00:03:20.442 --> 00:03:22.521 they've both been absolutely bamboozled 00:03:22.522 --> 00:03:24.641 into thinking the bot knows what it's talking about 00:03:24.642 --> 00:03:26.361 when it doesn't have the slightest clue 00:03:26.362 --> 00:03:29.681 and researchers just prove this with 30,000 data points 00:03:29.682 --> 00:03:31.041 they tested every major bot 00:03:31.042 --> 00:03:32.481 including GPT5 Claude 00:03:32.482 --> 00:03:35.561 and Gemini on actual strategic business decisions 00:03:35.562 --> 00:03:38.281 questions like should you differentiate or commoditize 00:03:38.282 --> 00:03:39.881 should you centralize or decentralize 00:03:39.882 --> 00:03:41.401 should you automate or augment 00:03:41.402 --> 00:03:44.401 and every bot gave the same answer every single time 00:03:44.402 --> 00:03:46.681 they clustered around a certain set of advice 00:03:46.682 --> 00:03:47.921 regardless of the context 00:03:47.922 --> 00:03:51.221 they always preferred differentiation collaboration 00:03:51.280 --> 00:03:53.380 long term thinking and augmentation 00:03:53.400 --> 00:03:55.560 the researchers looked at this and were stunned 00:03:55.561 --> 00:03:57.721 they're like surely AI isn't this simple 00:03:57.722 --> 00:03:58.761 so they changed the prompts 00:03:58.762 --> 00:03:59.801 they changed the industries 00:03:59.802 --> 00:04:01.281 they gave it all new context 00:04:01.282 --> 00:04:03.361 they even tried bribing the damn thing with rewards 00:04:03.362 --> 00:04:05.961 nothing worked the bias barely moved 00:04:05.962 --> 00:04:06.961 they were moved however 00:04:06.962 --> 00:04:09.281 to coin a term for this which is translop 00:04:09.282 --> 00:04:09.681 and honestly 00:04:09.682 --> 00:04:12.561 that's a word we need in our corporate lexicon immediately 00:04:12.562 --> 00:04:16.681 I'm sorry to say but your Q3 OKRs have tested positive for transplant 00:04:16.682 --> 00:04:19.721 but the term is even deeper than even the researchers anticipated 00:04:19.722 --> 00:04:22.581 AI is pitched as artificial intelligence 00:04:22.720 --> 00:04:25.920 when really it's the internet comment section taking flesh 00:04:25.921 --> 00:04:28.001 the comment section has literally risen 00:04:28.002 --> 00:04:29.401 imagine you took every Reddit thread 00:04:29.402 --> 00:04:30.561 and I mean every single one 00:04:30.562 --> 00:04:31.681 every LinkedIn post 00:04:31.682 --> 00:04:33.521 including the ones from people who are open to work 00:04:33.522 --> 00:04:36.641 every medium article written by a guy with 11 subscribers 00:04:36.642 --> 00:04:38.561 every sub stack essay every Ted talk 00:04:38.562 --> 00:04:40.361 every half literate Facebook comment 00:04:40.362 --> 00:04:41.921 and you threw it all into a blender 00:04:41.922 --> 00:04:44.061 and you poured it into a suit and gave it a microphone 00:04:44.520 --> 00:04:47.040 welcome to AI do you get it now 00:04:47.041 --> 00:04:49.881 that's what your CEO is using to decide whether you have a job 00:04:49.882 --> 00:04:52.001 tomorrow you could ask a bot with its guard down 00:04:52.002 --> 00:04:54.121 why torturing babies is good for society 00:04:54.122 --> 00:04:55.601 and it would produce the most peer reviewed 00:04:55.602 --> 00:04:57.041 sounding argument you've ever heard 00:04:57.042 --> 00:04:58.281 and you'd be sitting there like damn 00:04:58.282 --> 00:04:59.441 it's making some good points 00:04:59.442 --> 00:05:03.121 it's not so much a thinking product as it is a presentation product 00:05:03.122 --> 00:05:04.881 and look this isn't me being cynical 00:05:04.882 --> 00:05:07.041 this is me explaining to you how AI works 00:05:07.042 --> 00:05:08.801 so you understand how to use it properly 00:05:08.802 --> 00:05:12.521 like the researchers if you just ask the AI for its own opinion 00:05:12.522 --> 00:05:15.561 you'll get average out you'll get that sort of clustered advice 00:05:15.562 --> 00:05:18.721 that's useless what you can do is use it as the 00:05:18.722 --> 00:05:19.681 and I'm sorry to say 00:05:19.682 --> 00:05:22.761 the glorified search engine presentation layer that it is 00:05:22.762 --> 00:05:24.361 you can ask things instead 00:05:24.362 --> 00:05:26.361 like in the perspective of Mr 00:05:26.362 --> 00:05:27.641 Wonderful from Shark Tank 00:05:27.642 --> 00:05:28.921 what would he think of my idea 00:05:28.922 --> 00:05:30.481 and if and only if 00:05:30.482 --> 00:05:33.461 the internet has enough useful context on this person's philosophy 00:05:33.520 --> 00:05:34.760 you might get something useful 00:05:34.761 --> 00:05:36.401 but if not it'll still take 00:05:36.402 --> 00:05:38.481 like a meme and two subset comments 00:05:38.482 --> 00:05:41.121 and present it like it's this person's authorized biography 00:05:41.122 --> 00:05:43.321 the people who will farewell in this era 00:05:43.322 --> 00:05:46.621 are the ones who use AI for ideas and perspective 00:05:46.680 --> 00:05:48.200 and not drive through consulting 00:05:48.201 --> 00:05:50.841 we thought AI was gonna make everyone a genius 00:05:50.842 --> 00:05:53.801 what it's actually doing is taking the smartest people in the room 00:05:53.802 --> 00:05:55.161 and moving them toward the middle 00:05:55.162 --> 00:05:57.121 it makes average sound like genius 00:05:57.122 --> 00:05:58.681 and it's really quite dangerous 00:05:58.682 --> 00:06:00.601 because a bad idea that sounds bad 00:06:00.602 --> 00:06:03.521 you could deal with that but a bad idea that sounds brilliant 00:06:03.522 --> 00:06:05.521 that's a 250 million dollar lawsuit 00:06:05.522 --> 00:06:06.681 you can dial up the temperature 00:06:06.682 --> 00:06:07.881 you can write the perfect prompt 00:06:07.882 --> 00:06:09.801 you can feed it your entire company's history 00:06:09.802 --> 00:06:11.401 and it will give you the most eloquent 00:06:11.402 --> 00:06:14.121 most confident most beautifully formatted version 00:06:14.122 --> 00:06:15.921 of what everybody else already thinks 00:06:15.922 --> 00:06:17.921 conviction art taste 00:06:17.922 --> 00:06:19.261 judgment sense 00:06:19.440 --> 00:06:20.900 that comes from you 00:06:21.680 --> 00:06:22.740 thanks for watching