WEBVTT 00:00:02.080 --> 00:00:06.318 We're so excited for our very first 00:00:03.678 --> 00:00:10.320 special guest. He has helped build 00:00:06.318 --> 00:00:14.879 modern AI, then explain modern AI, and 00:00:10.320 --> 00:00:16.640 then occasionally rename modern AI. He 00:00:14.880 --> 00:00:18.640 actually helped co-ound open AAI right 00:00:16.640 --> 00:00:21.039 inside of this office. Was the one who 00:00:18.640 --> 00:00:23.760 actually got Autopilot working at Tesla 00:00:21.039 --> 00:00:26.480 back in the day, and he has a rare gift 00:00:23.760 --> 00:00:28.640 of making the most complex technical 00:00:26.480 --> 00:00:30.160 shifts feel both accessible and 00:00:28.640 --> 00:00:31.760 inevitable. 00:00:30.160 --> 00:00:35.039 You all know him for having coined the 00:00:31.760 --> 00:00:36.480 term vibe coding last year, but just in 00:00:35.039 --> 00:00:38.878 the last few months, he said something 00:00:36.479 --> 00:00:41.519 even more startling. That he's never 00:00:38.878 --> 00:00:43.039 felt more behind as a programmer. That's 00:00:41.520 --> 00:00:44.160 where we're starting today. Thank you, 00:00:43.039 --> 00:00:46.000 Andre, for joining us. 00:00:44.159 --> 00:00:47.119 >> Yeah. Hello. Excited to be here and to 00:00:46.000 --> 00:00:49.520 kick us off. 00:00:47.119 --> 00:00:51.119 >> Okay. So, just a couple months ago, you 00:00:49.520 --> 00:00:53.039 said that you've never felt more behind 00:00:51.119 --> 00:00:55.359 as a programmer. That's startling to 00:00:53.039 --> 00:00:57.198 hear from you of all people. Um, can you 00:00:55.359 --> 00:01:00.159 help us unpack that? Was that feeling 00:00:57.198 --> 00:01:02.479 exhilarating or unsettling? 00:01:00.159 --> 00:01:05.280 >> Uh yeah, a mixture of both for sure. Uh 00:01:02.479 --> 00:01:06.959 well, first of all, um 00:01:05.280 --> 00:01:08.400 I guess like as many of you, I've been 00:01:06.959 --> 00:01:10.000 using agentic tools like lot code, 00:01:08.400 --> 00:01:12.000 adjacent things, uh for a while, maybe 00:01:10.000 --> 00:01:13.680 over the last year as it came out and it 00:01:12.000 --> 00:01:15.200 was very good at you know chunks of code 00:01:13.680 --> 00:01:16.320 and sometimes it would mess up and you 00:01:15.200 --> 00:01:18.400 have to edit them and it was kind of 00:01:16.319 --> 00:01:21.438 helpful and then I would say December 00:01:18.400 --> 00:01:22.799 was this uh clear point where for me I 00:01:21.438 --> 00:01:24.959 was on a break so I had a bit more time. 00:01:22.799 --> 00:01:26.880 I think many other people were similar 00:01:24.959 --> 00:01:28.640 and uh I just started to notice that 00:01:26.879 --> 00:01:30.000 with the latest models uh the chunks 00:01:28.640 --> 00:01:31.280 just came out fine and then I kept 00:01:30.000 --> 00:01:32.640 asking for more and it just came out 00:01:31.280 --> 00:01:34.799 fine and then I can't remember the last 00:01:32.640 --> 00:01:36.478 time I corrected it and then I was I 00:01:34.799 --> 00:01:38.251 just you know trusted the system more 00:01:36.478 --> 00:01:39.039 and more and then I was vibe coding 00:01:38.251 --> 00:01:42.079 [laughter] 00:01:39.040 --> 00:01:43.600 and uh so it was kind of a I do think 00:01:42.078 --> 00:01:45.039 that it was a very stark transition. I 00:01:43.599 --> 00:01:47.280 think that a lot of people actually I 00:01:45.040 --> 00:01:49.680 tried to I tried to stress this on uh 00:01:47.280 --> 00:01:52.079 Twitter and or X because I think a lot 00:01:49.680 --> 00:01:54.320 of people experienced AI last year as 00:01:52.078 --> 00:01:55.919 ChachiPT adjacent thing. Uh but you 00:01:54.319 --> 00:01:58.000 really had to look again and you had to 00:01:55.920 --> 00:01:59.680 look as of December uh because things 00:01:58.000 --> 00:02:01.920 have changed fundamentally and uh 00:01:59.680 --> 00:02:04.079 especially on this like agentic coherent 00:02:01.920 --> 00:02:07.359 workflow uh that really started to 00:02:04.078 --> 00:02:09.359 actually work. Um, and so I would say 00:02:07.359 --> 00:02:12.560 that um, yeah, it was just that 00:02:09.360 --> 00:02:14.239 realization that really uh, uh, had me 00:02:12.560 --> 00:02:16.239 um, go down their whole rabbit hole of 00:02:14.239 --> 00:02:18.159 just, you know, infinity side projects. 00:02:16.239 --> 00:02:19.599 Uh, my side projects folder is like 00:02:18.159 --> 00:02:21.759 extremely full with lots of random 00:02:19.598 --> 00:02:23.759 things and, uh, just, uh, V coding all 00:02:21.759 --> 00:02:25.679 the time. Uh, so, uh, yeah, that kind of 00:02:23.759 --> 00:02:26.799 happened in December, I would say, and I 00:02:25.680 --> 00:02:28.080 was looking at the repercussions of that 00:02:26.800 --> 00:02:30.000 since. 00:02:28.080 --> 00:02:33.120 >> Um, you've talked a lot about this idea 00:02:30.000 --> 00:02:35.360 of LLMs as a new computer. um that it 00:02:33.120 --> 00:02:38.080 isn't just better software, it's a whole 00:02:35.360 --> 00:02:41.040 new computing paradigm. And um software 00:02:38.080 --> 00:02:43.920 1.0 was explicit rules, software 2.0 was 00:02:41.039 --> 00:02:46.479 learned weights, software 3.0 is this. 00:02:43.919 --> 00:02:48.878 Um if that's actually true, what does a 00:02:46.479 --> 00:02:50.799 team build differently the day they 00:02:48.878 --> 00:02:53.280 actually believe this, 00:02:50.800 --> 00:02:56.000 >> right? So uh yeah, exactly. So software 00:02:53.280 --> 00:02:57.680 1.0, I'm writing code, software 2.0, I'm 00:02:56.000 --> 00:02:59.598 actually programming by creating data 00:02:57.680 --> 00:03:01.040 sets and training uh training neural 00:02:59.598 --> 00:03:02.479 networks. So the programming is kind of 00:03:01.039 --> 00:03:03.519 like arranging data sets and maybe some 00:03:02.479 --> 00:03:05.280 objectives and neural network 00:03:03.519 --> 00:03:07.439 architectures. And then what happened is 00:03:05.280 --> 00:03:09.759 that basically if you train one of these 00:03:07.439 --> 00:03:12.400 GPT models or LLMs on a sufficiently 00:03:09.759 --> 00:03:14.399 large set of tasks implicit basically um 00:03:12.400 --> 00:03:15.680 implicitly because by training on the 00:03:14.400 --> 00:03:17.280 internet you have to multitask all the 00:03:15.680 --> 00:03:18.800 things that are in the data set. Uh 00:03:17.280 --> 00:03:20.000 these actually become kind of like a 00:03:18.800 --> 00:03:21.760 programmable computer in a certain 00:03:20.000 --> 00:03:24.000 sense. So software 3.0 know is kind of 00:03:21.759 --> 00:03:25.840 about uh you know your programming now 00:03:24.000 --> 00:03:28.479 turns to prompting and what's in the 00:03:25.840 --> 00:03:30.479 context window is your lever over the 00:03:28.479 --> 00:03:32.399 interpreter that is the LLM that is kind 00:03:30.479 --> 00:03:34.158 of like interpreting your context and uh 00:03:32.400 --> 00:03:37.760 performing computation in the dig 00:03:34.158 --> 00:03:39.840 digital information space. So I guess um 00:03:37.759 --> 00:03:41.120 yeah that's kind of the transition and I 00:03:39.840 --> 00:03:42.560 think there's a few examples of that 00:03:41.120 --> 00:03:44.878 really drove it home for me and maybe 00:03:42.560 --> 00:03:48.000 that might be instructive. Uh so for 00:03:44.878 --> 00:03:49.759 example when you when openclaw came out 00:03:48.000 --> 00:03:50.878 when you want to install openclaw you 00:03:49.759 --> 00:03:52.719 would expect that normally this is a 00:03:50.878 --> 00:03:54.639 bash bash script like a shell script. So 00:03:52.719 --> 00:03:57.359 run the shell script to run to install 00:03:54.639 --> 00:03:58.559 open claw. Um but the thing is that in 00:03:57.360 --> 00:04:00.080 order to target lots of different 00:03:58.560 --> 00:04:01.680 platforms and lots of different types of 00:04:00.080 --> 00:04:03.200 computers you might run an open claw. 00:04:01.680 --> 00:04:05.120 This these shell scripts usually balloon 00:04:03.199 --> 00:04:06.158 up and become extremely complex. But the 00:04:05.120 --> 00:04:07.840 thing is you're still stuck in a 00:04:06.158 --> 00:04:09.840 software 1.0 universe of wanting to 00:04:07.840 --> 00:04:12.000 write the code. And actually the open 00:04:09.840 --> 00:04:13.920 claw installation is a is a copy paste 00:04:12.000 --> 00:04:15.759 of a b bunch of text that you're 00:04:13.919 --> 00:04:18.079 supposed to give to your agent. Uh so 00:04:15.759 --> 00:04:19.279 basically it's it's a little skill of uh 00:04:18.079 --> 00:04:20.639 you know copy paste this and give it to 00:04:19.279 --> 00:04:22.078 your agent and it will install open 00:04:20.639 --> 00:04:23.600 claw. And the reason this is a lot more 00:04:22.079 --> 00:04:25.199 powerful is you're working now in the 00:04:23.600 --> 00:04:27.759 software 3.0 paradigm where you don't 00:04:25.199 --> 00:04:29.439 have to precisely spell out you know all 00:04:27.759 --> 00:04:30.960 the individual details of that setup. 00:04:29.439 --> 00:04:32.560 The agent has its own intelligence that 00:04:30.959 --> 00:04:34.638 it packages up and then it kind of like 00:04:32.560 --> 00:04:36.399 follows the instructions and it looks at 00:04:34.639 --> 00:04:37.439 your environment, your computer and it 00:04:36.399 --> 00:04:38.638 kind of like performs intelligent 00:04:37.439 --> 00:04:40.399 actions to make things work and it 00:04:38.639 --> 00:04:42.960 debugs things in the loop and it's just 00:04:40.399 --> 00:04:44.478 like so much more powerful, right? So I 00:04:42.959 --> 00:04:46.000 think that's a very different kind of 00:04:44.478 --> 00:04:47.519 like way of thinking about it is just 00:04:46.000 --> 00:04:48.720 like what is the piece of text to copy 00:04:47.519 --> 00:04:50.560 paste to your agent? That's the 00:04:48.720 --> 00:04:52.560 programming paradigm. Now I think one 00:04:50.560 --> 00:04:54.160 more maybe uh example that comes to mind 00:04:52.560 --> 00:04:56.720 that is even more extreme than that is 00:04:54.160 --> 00:05:00.400 when I was building um menu genen. So, 00:04:56.720 --> 00:05:01.919 menu genen is this idea where you um you 00:05:00.399 --> 00:05:03.679 come to a restaurant, they give you a 00:05:01.918 --> 00:05:05.279 menu. There's no pictures usually. So, I 00:05:03.680 --> 00:05:07.680 don't know what any of these things are 00:05:05.279 --> 00:05:09.679 uh usually like 30% of the things I have 00:05:07.680 --> 00:05:12.000 no idea what they are, 50%. So, I wanted 00:05:09.680 --> 00:05:13.840 to take a photo of the restaurant menu 00:05:12.000 --> 00:05:16.240 and to get pictures of what those things 00:05:13.839 --> 00:05:18.239 might look like in a generic sense. And 00:05:16.240 --> 00:05:20.079 so I built I've vcoded this app that 00:05:18.240 --> 00:05:21.439 basically lets you upload a photo and it 00:05:20.079 --> 00:05:24.879 does all this stuff and it runs on 00:05:21.439 --> 00:05:26.719 Verscell and uh it basically rerenders 00:05:24.879 --> 00:05:28.319 the menu and it gives you like all the 00:05:26.720 --> 00:05:31.039 items and it gives you a picture that it 00:05:28.319 --> 00:05:33.839 uses an image um you know generator uh 00:05:31.038 --> 00:05:35.918 for to basically OCR all the different 00:05:33.839 --> 00:05:37.038 titles uh use the image generator to get 00:05:35.918 --> 00:05:39.839 pictures of them and then shows it to 00:05:37.038 --> 00:05:41.680 you. And then I saw the software 3.0 00:05:39.839 --> 00:05:43.198 version of this which is which blew my 00:05:41.680 --> 00:05:46.000 mind which is literally just take your 00:05:43.199 --> 00:05:48.879 photo give it to Gemini and say use 00:05:46.000 --> 00:05:51.439 Nanobanana to overlay the the things 00:05:48.879 --> 00:05:52.959 onto the menu. Uh and Nanabanana 00:05:51.439 --> 00:05:54.319 basically returned an image that is 00:05:52.959 --> 00:05:56.638 exactly the picture of the menu that I 00:05:54.319 --> 00:05:58.478 took but it actually put into the pixels 00:05:56.639 --> 00:06:02.079 it rendered the different things in the 00:05:58.478 --> 00:06:04.318 menu and this blew my mind because 00:06:02.079 --> 00:06:06.159 actually all of my menu gen is spirious. 00:06:04.319 --> 00:06:09.039 It's working in the old paradigm that 00:06:06.160 --> 00:06:11.360 app shouldn't exist. uh and uh yeah the 00:06:09.038 --> 00:06:14.000 software 3.0 paradigm is a lot more kind 00:06:11.360 --> 00:06:15.840 of raw. It just um your neural network 00:06:14.000 --> 00:06:18.079 is doing more and more of the work and 00:06:15.839 --> 00:06:19.839 your prompt or context is just the image 00:06:18.079 --> 00:06:21.439 and the output is an image and there's 00:06:19.839 --> 00:06:24.959 no need to have any of the app in 00:06:21.439 --> 00:06:27.839 between. Um so I think that people have 00:06:24.959 --> 00:06:30.000 to kind of like reframe you know not to 00:06:27.839 --> 00:06:31.439 work in existing paradigm of what things 00:06:30.000 --> 00:06:33.839 existed and just think about it as a 00:06:31.439 --> 00:06:36.000 speed up of what exists. It's actually 00:06:33.839 --> 00:06:37.359 like new things are available now. And 00:06:36.000 --> 00:06:38.879 going back to your programming question, 00:06:37.360 --> 00:06:40.479 it's not even I think that's also an 00:06:38.879 --> 00:06:41.680 example of working in the in the old 00:06:40.478 --> 00:06:42.879 mindset because it's not just about 00:06:41.680 --> 00:06:44.959 programming and programming becoming 00:06:42.879 --> 00:06:47.360 faster. This is more general information 00:06:44.959 --> 00:06:49.439 processing that is automatable now. So 00:06:47.360 --> 00:06:51.600 um it's not just even about code. So 00:06:49.439 --> 00:06:53.439 previous code worked over kind of like 00:06:51.600 --> 00:06:55.199 structured data, right? And uh you write 00:06:53.439 --> 00:06:56.800 code over structured data. But like for 00:06:55.199 --> 00:06:59.759 example with my LLM knowledge basis 00:06:56.800 --> 00:07:01.439 project um basically you get LLMs to 00:06:59.759 --> 00:07:03.120 create wikis for your organization or 00:07:01.439 --> 00:07:04.319 for you in person etc. This is not even 00:07:03.120 --> 00:07:06.720 a program. This is not something that 00:07:04.319 --> 00:07:08.080 could exist before because there was no 00:07:06.720 --> 00:07:09.360 there was no code that would create a 00:07:08.079 --> 00:07:10.959 knowledge base based on a bunch of 00:07:09.360 --> 00:07:14.000 facts. But now you can just take these 00:07:10.959 --> 00:07:15.918 documents and uh basically uh recompile 00:07:14.000 --> 00:07:17.680 them in a different way and uh reorder 00:07:15.918 --> 00:07:19.680 them and create something that is uh new 00:07:17.680 --> 00:07:22.400 and interesting uh as a reframing of the 00:07:19.680 --> 00:07:24.639 data. And so these are new things that 00:07:22.399 --> 00:07:26.560 weren't possible. Uh and so I think this 00:07:24.639 --> 00:07:29.038 is uh something that I keep trying to 00:07:26.560 --> 00:07:31.038 get back to as to not only what can we 00:07:29.038 --> 00:07:33.519 do that existed that is faster now but I 00:07:31.038 --> 00:07:35.038 think there's new opportunities of just 00:07:33.519 --> 00:07:36.240 things that couldn't be possible before 00:07:35.038 --> 00:07:37.199 and I almost think that that's more 00:07:36.240 --> 00:07:40.000 exciting. 00:07:37.199 --> 00:07:41.840 >> I love the menu genen progression and 00:07:40.000 --> 00:07:43.839 dichotomy that you laid out and I think 00:07:41.839 --> 00:07:45.519 even I'm sure many folks here followed 00:07:43.839 --> 00:07:48.879 your own progression of programming from 00:07:45.519 --> 00:07:51.038 last October to early January February 00:07:48.879 --> 00:07:54.560 this year. Um, if you extrapolate that 00:07:51.038 --> 00:07:56.959 further, what is the 2026 equivalent um, 00:07:54.560 --> 00:07:59.519 for building websites in the '9s, 00:07:56.959 --> 00:08:02.560 building mobile apps in the 2010s, 00:07:59.519 --> 00:08:04.560 building SAS um, in the last cloud era, 00:08:02.560 --> 00:08:06.720 what will look completely obvious in 00:08:04.560 --> 00:08:08.000 hindsight that is still mostly unbuilt 00:08:06.720 --> 00:08:10.240 today? 00:08:08.000 --> 00:08:12.399 >> Um, [clears throat] well, going with the 00:08:10.240 --> 00:08:13.519 example of menu, I guess, uh, so a lot 00:08:12.399 --> 00:08:15.198 of this code shouldn't exist and it's 00:08:13.519 --> 00:08:17.120 just neural network doing most of the 00:08:15.199 --> 00:08:19.120 work. Um I do think that the 00:08:17.120 --> 00:08:21.439 extrapolation looks very weird because 00:08:19.120 --> 00:08:23.598 you could basically imagine 00:08:21.439 --> 00:08:25.439 I don't I yeah so you could imagine 00:08:23.598 --> 00:08:28.800 completely neural computers in a certain 00:08:25.439 --> 00:08:30.478 sense you feed raw videos like imagine a 00:08:28.800 --> 00:08:32.719 device you takes raw videos or audio 00:08:30.478 --> 00:08:35.199 into basically what's a neural net and 00:08:32.719 --> 00:08:37.440 uh uses diffusion to render a UI that is 00:08:35.200 --> 00:08:40.560 kind of like you know unique for that 00:08:37.440 --> 00:08:42.159 moment in a certain sense and um I kind 00:08:40.559 --> 00:08:43.359 of feel like in the early days of 00:08:42.158 --> 00:08:45.039 computing actually people were a little 00:08:43.360 --> 00:08:46.800 bit confused as to whether computers 00:08:45.039 --> 00:08:48.480 would look like calculators or computers 00:08:46.799 --> 00:08:50.319 would look like neural nets and in 50s 00:08:48.480 --> 00:08:52.159 and 60s it was not really obvious which 00:08:50.320 --> 00:08:53.360 way would go and of course we went down 00:08:52.159 --> 00:08:55.120 the calculator path and ended up 00:08:53.360 --> 00:08:56.320 building classical computing and then 00:08:55.120 --> 00:08:58.159 neural nets are currently running 00:08:56.320 --> 00:09:00.240 virtualized on existing computers but 00:08:58.159 --> 00:09:01.600 you could imagine I think that uh a lot 00:09:00.240 --> 00:09:02.959 of this will flip and that the neural 00:09:01.600 --> 00:09:05.600 net becomes kind of like the host 00:09:02.958 --> 00:09:07.599 process and uh the CPUs become kind of 00:09:05.600 --> 00:09:09.278 like the co-processor so we saw the 00:09:07.600 --> 00:09:10.800 diagram of you know intelligence compute 00:09:09.278 --> 00:09:12.958 is going to of neural networks is going 00:09:10.799 --> 00:09:14.879 to take over and become the dominant 00:09:12.958 --> 00:09:17.199 spend of flops so you could imagine 00:09:14.879 --> 00:09:18.799 something really weird and foreign when 00:09:17.200 --> 00:09:20.320 where neural nets are doing most of the 00:09:18.799 --> 00:09:22.879 heavy lifting. They're using tool use as 00:09:20.320 --> 00:09:24.000 this like you know um historical 00:09:22.879 --> 00:09:25.759 appendage for some kinds of like 00:09:24.000 --> 00:09:27.600 deterministic tasks. Uh but what's 00:09:25.759 --> 00:09:29.919 really running the show is these uh 00:09:27.600 --> 00:09:31.278 neural nets that are in a certain way. 00:09:29.919 --> 00:09:33.120 Um so you can imagine something 00:09:31.278 --> 00:09:34.639 extremely foreign as the extrapolation 00:09:33.120 --> 00:09:36.720 but I think we're going to probably get 00:09:34.639 --> 00:09:39.120 there uh sort of piece by piece. Um and 00:09:36.720 --> 00:09:40.990 I don't yeah that that progression is 00:09:39.120 --> 00:09:41.120 TBD I would say. 00:09:40.990 --> 00:09:43.278 >> [snorts] 00:09:41.120 --> 00:09:45.278 >> I'd like to talk a little bit about um 00:09:43.278 --> 00:09:47.519 uh this concept of verifiability, the 00:09:45.278 --> 00:09:49.838 fact that AI will automate faster and 00:09:47.519 --> 00:09:52.480 more easily domains where the output can 00:09:49.839 --> 00:09:54.320 be verified. Um if that framework is 00:09:52.480 --> 00:09:56.399 right, what work is about to move much 00:09:54.320 --> 00:09:58.640 faster than people realize and what 00:09:56.399 --> 00:10:00.480 professions do we have that people 00:09:58.639 --> 00:10:02.799 actually think are safe but that are 00:10:00.480 --> 00:10:05.360 actually highly verifiable? 00:10:02.799 --> 00:10:07.519 Uh yes. So I I spent uh some time 00:10:05.360 --> 00:10:09.680 writing about verifiability and um 00:10:07.519 --> 00:10:12.399 basically like traditional computers can 00:10:09.679 --> 00:10:14.958 easily automate what you can specify in 00:10:12.399 --> 00:10:16.958 code and uh kind of this latest round of 00:10:14.958 --> 00:10:19.518 LLMs can easily automate what you can uh 00:10:16.958 --> 00:10:20.958 verify in a certain in a certain sense 00:10:19.519 --> 00:10:22.720 because the way this works is that when 00:10:20.958 --> 00:10:24.078 frontier labs are training these LLMs 00:10:22.720 --> 00:10:25.519 these are giant reinforcement learning 00:10:24.078 --> 00:10:28.159 environments. So they are given 00:10:25.519 --> 00:10:29.679 verification rewards and then because of 00:10:28.159 --> 00:10:32.000 the way that these models are trained 00:10:29.679 --> 00:10:34.239 they end up basically uh progressing and 00:10:32.000 --> 00:10:36.320 creating these like jagged entities that 00:10:34.240 --> 00:10:37.759 really peak in capability in kind of 00:10:36.320 --> 00:10:39.440 like verifiable domains like math and 00:10:37.759 --> 00:10:41.600 code and adjacent and kind of like 00:10:39.440 --> 00:10:43.279 stagnate and are a little bit um you 00:10:41.600 --> 00:10:44.959 know rough around the edges when uh 00:10:43.278 --> 00:10:46.480 things are not kind of like in that in 00:10:44.958 --> 00:10:47.759 that space. So I think the reason I 00:10:46.480 --> 00:10:49.519 wrote about verifiability is I'm trying 00:10:47.759 --> 00:10:52.159 to understand why these things are so 00:10:49.519 --> 00:10:54.000 jagged. Um and some of it has to do with 00:10:52.159 --> 00:10:55.838 how the labs train the models but I 00:10:54.000 --> 00:10:57.519 think some of it also has to do with um 00:10:55.839 --> 00:10:58.880 the focus of the labs and what they 00:10:57.519 --> 00:11:00.799 happen to put into the data 00:10:58.879 --> 00:11:01.919 distribution. Uh because some things 00:11:00.799 --> 00:11:03.759 basically are significantly more 00:11:01.919 --> 00:11:05.039 valuable in economy and end up creating 00:11:03.759 --> 00:11:06.720 more environments because the labs 00:11:05.039 --> 00:11:08.159 wanted to work in those settings. So I 00:11:06.720 --> 00:11:09.440 think code is a good example of that. 00:11:08.159 --> 00:11:10.879 There's probably lots of verifiable 00:11:09.440 --> 00:11:12.079 environments they could think about that 00:11:10.879 --> 00:11:13.278 happen not to make it into the mix 00:11:12.078 --> 00:11:15.919 because they're just not that useful to 00:11:13.278 --> 00:11:18.480 have the capability around. Um, but I 00:11:15.919 --> 00:11:21.120 think to me the big um I guess like the 00:11:18.480 --> 00:11:22.959 big mystery is uh the favorite example 00:11:21.120 --> 00:11:24.560 for a while was that how many letters 00:11:22.958 --> 00:11:26.000 are are in a strawberry and the models 00:11:24.559 --> 00:11:27.919 would famously get this wrong and it's 00:11:26.000 --> 00:11:29.759 an example of jaggedness. Uh the models 00:11:27.919 --> 00:11:32.159 now patch this I think but the new one 00:11:29.759 --> 00:11:34.480 is I want to go to a car wash to wash my 00:11:32.159 --> 00:11:36.799 car and it's 50 meters away. Should I 00:11:34.480 --> 00:11:38.399 drive or should I walk? And 00:11:36.799 --> 00:11:40.958 state-of-the-art models today will tell 00:11:38.399 --> 00:11:42.958 you to walk because it's so close. How 00:11:40.958 --> 00:11:46.078 is it possible that state-of-the-art 00:11:42.958 --> 00:11:48.799 Opus 4.7 will simultaneously refactor a 00:11:46.078 --> 00:11:50.479 100,000 like [laughter] codebase line 00:11:48.799 --> 00:11:52.559 codebase or find zero day 00:11:50.480 --> 00:11:56.480 vulnerabilities and yet tells me to walk 00:11:52.559 --> 00:11:58.958 to this car wash? This is insane. And to 00:11:56.480 --> 00:12:01.278 whatever extent these uh models are 00:11:58.958 --> 00:12:02.479 remain jagged, it's an indication that 00:12:01.278 --> 00:12:05.600 number one maybe something's slightly 00:12:02.480 --> 00:12:07.759 off or um number two you need to 00:12:05.600 --> 00:12:09.360 actually be in the loop a little bit and 00:12:07.759 --> 00:12:11.200 you need to treat them as tools and you 00:12:09.360 --> 00:12:12.879 do have to kind of stay in touch with 00:12:11.200 --> 00:12:14.480 what they're doing. And so I think all 00:12:12.879 --> 00:12:16.078 of my writing long story short about 00:12:14.480 --> 00:12:18.480 verifiability is just trying to 00:12:16.078 --> 00:12:20.399 understand um why these things are 00:12:18.480 --> 00:12:22.079 jacked. Is there any pattern to it? And 00:12:20.399 --> 00:12:25.200 I think it's some kind of a combination 00:12:22.078 --> 00:12:28.078 of verifiable plus labs care. Maybe one 00:12:25.200 --> 00:12:31.040 more anecdote that is instructive is uh 00:12:28.078 --> 00:12:33.199 from GPT 3.5 to GPT4 people noticed that 00:12:31.039 --> 00:12:34.480 chess improved a lot and I think a lot 00:12:33.200 --> 00:12:36.240 of people thought oh well it's just a 00:12:34.480 --> 00:12:38.000 progression of the capabilities but 00:12:36.240 --> 00:12:39.360 actually it's it's more that uh I think 00:12:38.000 --> 00:12:41.120 this is public information I think I saw 00:12:39.360 --> 00:12:43.759 it on the internet um a huge amount of 00:12:41.120 --> 00:12:46.000 like um data of chess made it into the 00:12:43.759 --> 00:12:48.319 pre-training set and just because it's 00:12:46.000 --> 00:12:50.159 in a data distribution uh basically the 00:12:48.320 --> 00:12:53.120 model improved a lot more than it would 00:12:50.159 --> 00:12:55.120 just by default. So someone at OpenAI 00:12:53.120 --> 00:12:56.799 decided to add this data and now you 00:12:55.120 --> 00:12:58.320 have a capability that just peaked a lot 00:12:56.799 --> 00:13:01.519 more. And so that's why I think I'm 00:12:58.320 --> 00:13:03.040 stressing this um dimension of it as we 00:13:01.519 --> 00:13:04.639 are slightly at the mercy of whatever 00:13:03.039 --> 00:13:06.240 the labs are doing, whatever they happen 00:13:04.639 --> 00:13:08.000 to put into the mix. And you have to 00:13:06.240 --> 00:13:10.159 actually explore this thing that they 00:13:08.000 --> 00:13:11.679 give you that has no manual. And it 00:13:10.159 --> 00:13:13.679 works in certain settings, but maybe not 00:13:11.679 --> 00:13:16.559 in some settings. And you have to kind 00:13:13.679 --> 00:13:17.838 of um explore it a little bit. And uh if 00:13:16.559 --> 00:13:19.919 you're in the circuits that were part of 00:13:17.839 --> 00:13:21.519 the RL, you fly. And if you're in the 00:13:19.919 --> 00:13:22.879 circuits that are out of the data 00:13:21.519 --> 00:13:24.240 distribution, uh you're going to 00:13:22.879 --> 00:13:26.159 struggle and you have to kind of figure 00:13:24.240 --> 00:13:28.159 out which which circuits you're in in 00:13:26.159 --> 00:13:29.519 your application. And if you and if 00:13:28.159 --> 00:13:30.958 you're not in the circuits, then you 00:13:29.519 --> 00:13:32.879 have to really look at fine-tuning and 00:13:30.958 --> 00:13:34.078 doing some of your own work because it's 00:13:32.879 --> 00:13:36.639 not going to necessarily come out of the 00:13:34.078 --> 00:13:38.078 LLM out of the box. 00:13:36.639 --> 00:13:40.240 >> I'd love to come back to the concept of 00:13:38.078 --> 00:13:42.479 jagged intelligence in a little bit. Um, 00:13:40.240 --> 00:13:44.799 if you are a founder today and thinking 00:13:42.480 --> 00:13:46.800 about building a company, you are trying 00:13:44.799 --> 00:13:49.039 to solve a problem that you think is 00:13:46.799 --> 00:13:51.359 tractable, something that uh is a domain 00:13:49.039 --> 00:13:53.120 that is verifiable, but you look around 00:13:51.360 --> 00:13:56.560 and you think, "Oh my gosh, well, the 00:13:53.120 --> 00:13:58.560 labs have really really started uh 00:13:56.559 --> 00:14:00.799 getting to escape velocity in the ones 00:13:58.559 --> 00:14:02.638 that seem most obvious, math, coding, 00:14:00.799 --> 00:14:05.679 and others." What would your advice be 00:14:02.639 --> 00:14:08.639 to to the founders in the audience? 00:14:05.679 --> 00:14:10.479 Um 00:14:08.639 --> 00:14:12.079 so I think maybe that comes to the 00:14:10.480 --> 00:14:14.800 previous question of I do think that 00:14:12.078 --> 00:14:17.039 verifiability because it um let me 00:14:14.799 --> 00:14:18.639 think. So verifiability makes something 00:14:17.039 --> 00:14:20.319 tractable in the current paradigm 00:14:18.639 --> 00:14:24.560 because you can throw a huge amount of 00:14:20.320 --> 00:14:26.800 RL at it. Um so maybe one way to see it 00:14:24.559 --> 00:14:28.638 is that uh that remains true even if the 00:14:26.799 --> 00:14:30.559 labs are not focusing on it directly. So 00:14:28.639 --> 00:14:31.839 if you are in a verifiable setting where 00:14:30.559 --> 00:14:34.078 you could create these RL environments 00:14:31.839 --> 00:14:35.279 or examples then that actually sets you 00:14:34.078 --> 00:14:36.719 up to potentially do your own fine 00:14:35.278 --> 00:14:38.078 tuning and you might benefit from that. 00:14:36.720 --> 00:14:39.759 But that is fundamentally technology 00:14:38.078 --> 00:14:41.198 that just works. You can pull a lever if 00:14:39.759 --> 00:14:43.439 you have huge amount of diverse data 00:14:41.198 --> 00:14:44.958 sets of RL environments etc. Uh you can 00:14:43.440 --> 00:14:47.920 use your favorite fine-tuning framework 00:14:44.958 --> 00:14:49.198 and um and uh pull the lever and get 00:14:47.919 --> 00:14:51.919 something that actually uh works pretty 00:14:49.198 --> 00:14:54.958 well. So um I don't know what the 00:14:51.919 --> 00:14:56.479 examples of this might be. Um, but I do 00:14:54.958 --> 00:14:58.159 think there are some very valuable uh 00:14:56.480 --> 00:14:59.519 reinforcement learning environments that 00:14:58.159 --> 00:15:01.278 people could think of that I think are 00:14:59.519 --> 00:15:02.720 not part of the Yeah, I don't want to 00:15:01.278 --> 00:15:04.799 give away the answer, but there is one 00:15:02.720 --> 00:15:06.480 domain that I think is very uh Oh, okay. 00:15:04.799 --> 00:15:08.639 Sorry, I don't mean to vape post on on 00:15:06.480 --> 00:15:09.360 the stage, but there are some examples 00:15:08.639 --> 00:15:11.039 of this. 00:15:09.360 --> 00:15:13.039 >> On the flip side, what do you think 00:15:11.039 --> 00:15:14.958 still feels automatable only from a 00:15:13.039 --> 00:15:17.278 distance? 00:15:14.958 --> 00:15:19.439 >> I do think that ultimately almost 00:15:17.278 --> 00:15:21.039 everything can be made uh verifiable to 00:15:19.440 --> 00:15:23.839 some extent. some things easier than 00:15:21.039 --> 00:15:25.679 others. Um because even for like things 00:15:23.839 --> 00:15:27.839 like writing or so on, you can imagine 00:15:25.679 --> 00:15:29.759 having a council of LLM judges and 00:15:27.839 --> 00:15:31.760 probably get get to some get get 00:15:29.759 --> 00:15:33.360 something uh reasonable out of the um 00:15:31.759 --> 00:15:36.639 from from this kind of an approach. So 00:15:33.360 --> 00:15:40.320 it's more about what's easy or hard. Um 00:15:36.639 --> 00:15:42.000 so I I do think that ultimately um uh 00:15:40.320 --> 00:15:43.199 yeah, I think uh 00:15:42.000 --> 00:15:45.679 >> everything [laughter] 00:15:43.198 --> 00:15:47.599 >> everything is automatable. 00:15:45.679 --> 00:15:49.278 >> Amazing. Okay. Um, so last year you 00:15:47.600 --> 00:15:50.800 coined the term vibe coding and today 00:15:49.278 --> 00:15:52.958 we're in a world that feels a little bit 00:15:50.799 --> 00:15:54.159 more serious, more regent engineering. 00:15:52.958 --> 00:15:55.359 What do you think is the difference 00:15:54.159 --> 00:15:57.360 between the two and what would you 00:15:55.360 --> 00:15:59.120 actually call what we're in today? 00:15:57.360 --> 00:16:01.120 >> Uh, yeah. So I would say vibe coding is 00:15:59.120 --> 00:16:03.039 about raising the floor for everyone in 00:16:01.120 --> 00:16:05.120 terms of what they can do in software. 00:16:03.039 --> 00:16:06.639 So the floor rises, everyone can vibe 00:16:05.120 --> 00:16:08.639 code anything and that's amazing, 00:16:06.639 --> 00:16:10.079 incredible. But then I would say agentic 00:16:08.639 --> 00:16:11.919 engineering is about preserving the 00:16:10.078 --> 00:16:13.838 quality bar of what existed before in 00:16:11.919 --> 00:16:15.919 professional software. So you're not 00:16:13.839 --> 00:16:18.880 allowed to introduce vulnerabilities due 00:16:15.919 --> 00:16:20.319 to VIP coding. Um you are um you're 00:16:18.879 --> 00:16:22.639 still responsible for your software just 00:16:20.320 --> 00:16:24.800 as before, but can you go faster? And 00:16:22.639 --> 00:16:26.240 spoiler is you can but how do you how do 00:16:24.799 --> 00:16:28.240 you do that properly? And so to me 00:16:26.240 --> 00:16:29.600 agentic engineering when I call it that 00:16:28.240 --> 00:16:31.198 because I do think it's kind of like an 00:16:29.600 --> 00:16:32.480 engineering discipline. You have these 00:16:31.198 --> 00:16:33.838 agents which are these like spiky 00:16:32.480 --> 00:16:35.759 entities. They're a bit fable, a little 00:16:33.839 --> 00:16:37.680 bit stocastic, but they are extremely 00:16:35.759 --> 00:16:39.839 powerful. is how do you how do you 00:16:37.679 --> 00:16:42.479 coordinate them to go faster without 00:16:39.839 --> 00:16:46.000 sacrificing your quality bar and doing 00:16:42.480 --> 00:16:48.879 that well and correctly um is the the 00:16:46.000 --> 00:16:50.078 realm of agentic engineering um so I 00:16:48.879 --> 00:16:51.759 kind of see them as as different like 00:16:50.078 --> 00:16:53.599 one is about maybe raising the raising 00:16:51.759 --> 00:16:55.360 the floor and the other is about um you 00:16:53.600 --> 00:16:58.159 know extrapolating and what I'm seeing I 00:16:55.360 --> 00:17:01.199 think is there is a very high ceiling on 00:16:58.159 --> 00:17:02.719 agentic engineer uh capability and you 00:17:01.198 --> 00:17:04.720 know people used to talk about the 10x 00:17:02.720 --> 00:17:08.558 engineer previously I think that this is 00:17:04.720 --> 00:17:11.759 magnified a lot more 10x is uh is not uh 00:17:08.558 --> 00:17:13.519 the speed up you gain. Um and I think uh 00:17:11.759 --> 00:17:16.160 it does seem to me like people who are 00:17:13.519 --> 00:17:18.078 very good at this um peak a lot more 00:17:16.160 --> 00:17:18.558 than 10x uh from from my perspective 00:17:18.078 --> 00:17:21.279 right now. 00:17:18.558 --> 00:17:23.519 >> I really like that framing. Um one thing 00:17:21.279 --> 00:17:25.199 that when Sam Alman came to AIN last 00:17:23.519 --> 00:17:27.199 year, one memorable thing he said was 00:17:25.199 --> 00:17:29.200 that people of different generations use 00:17:27.199 --> 00:17:31.200 chatpt differently. So if you're in your 00:17:29.200 --> 00:17:32.798 30s, you use it as a Google search 00:17:31.200 --> 00:17:35.200 replacement. But if you're in your 00:17:32.798 --> 00:17:37.440 teens, tragic is your gateway to the 00:17:35.200 --> 00:17:39.279 internet. What is the parallel here in 00:17:37.440 --> 00:17:42.640 coding today? If we were to watch two 00:17:39.279 --> 00:17:45.359 people code using OpenClaw, Claude Code, 00:17:42.640 --> 00:17:47.840 Codeex, one you'd consider mediocre at 00:17:45.359 --> 00:17:49.599 it and one you would consider fully AI 00:17:47.839 --> 00:17:51.591 native. How would you describe the 00:17:49.599 --> 00:17:51.678 difference? 00:17:51.592 --> 00:17:53.600 >> [clears throat] 00:17:51.679 --> 00:17:55.038 >> I mean I think it's a just trying to get 00:17:53.599 --> 00:17:56.798 the most out of the tools that are 00:17:55.038 --> 00:17:59.679 available utilizing all of their 00:17:56.798 --> 00:18:02.160 features investing into your own um kind 00:17:59.679 --> 00:18:03.440 of setup. Uh so just like previously all 00:18:02.160 --> 00:18:04.480 the engineers are used to basically 00:18:03.440 --> 00:18:06.558 getting the most out of the tools you 00:18:04.480 --> 00:18:09.519 use either it's vim or v code or now 00:18:06.558 --> 00:18:13.038 it's you know cloth code or codec or so 00:18:09.519 --> 00:18:16.400 on. So um just investing into your setup 00:18:13.038 --> 00:18:18.558 um and um utilizing a lot of the you 00:18:16.400 --> 00:18:20.798 know uh tools that are available to you. 00:18:18.558 --> 00:18:23.119 Um and I think it just kind of looks 00:18:20.798 --> 00:18:26.798 like that. I do think that um maybe 00:18:23.119 --> 00:18:29.839 related thought is um a lot of people 00:18:26.798 --> 00:18:31.918 are maybe hiring um for this right 00:18:29.839 --> 00:18:34.639 because they want to hire strong agentic 00:18:31.919 --> 00:18:37.280 engineers. I do think that um what I'm 00:18:34.640 --> 00:18:39.440 seeing is that uh the you know most 00:18:37.279 --> 00:18:41.918 people have still not refactored their 00:18:39.440 --> 00:18:44.240 um their hiring process for a gentic 00:18:41.919 --> 00:18:46.400 engineer capability right like if you're 00:18:44.240 --> 00:18:48.240 giving out puzzles to solve and this is 00:18:46.400 --> 00:18:50.000 still the old paradigm I would say that 00:18:48.240 --> 00:18:52.400 hiring have to has to look like give me 00:18:50.000 --> 00:18:53.839 a really big project and see someone 00:18:52.400 --> 00:18:57.280 implement that big project like let's 00:18:53.839 --> 00:18:59.038 write say a Twitter clone uh for agents 00:18:57.279 --> 00:19:01.519 and then uh make it really good make it 00:18:59.038 --> 00:19:03.839 really secure and then have some agents 00:19:01.519 --> 00:19:06.639 uh simulate some activity uh on this 00:19:03.839 --> 00:19:09.038 Twitter and then I'm going to use 10 00:19:06.640 --> 00:19:12.960 codecs 5.4x for X high to try to break 00:19:09.038 --> 00:19:15.440 your break your um uh this website that 00:19:12.960 --> 00:19:16.640 you deployed and they're going to try to 00:19:15.440 --> 00:19:18.320 basically break it and they should not 00:19:16.640 --> 00:19:20.000 be able to break it. And so maybe it 00:19:18.319 --> 00:19:21.678 looks like that, right? And so yeah, 00:19:20.000 --> 00:19:25.038 watching people in that that setting and 00:19:21.679 --> 00:19:26.559 building bigger uh projects and uh 00:19:25.038 --> 00:19:28.400 utilize utilizing the tooling is maybe 00:19:26.558 --> 00:19:29.038 what I would uh look at for the most 00:19:28.400 --> 00:19:31.280 part. 00:19:29.038 --> 00:19:33.679 >> And as agents do more, what human skill 00:19:31.279 --> 00:19:34.879 do you think becomes more valuable, not 00:19:33.679 --> 00:19:37.038 less? 00:19:34.880 --> 00:19:39.440 >> Uh so um yeah, it's a good question. I 00:19:37.038 --> 00:19:40.558 think um well right now the answer is 00:19:39.440 --> 00:19:44.480 that the agents are kind of like these 00:19:40.558 --> 00:19:46.960 intern entities right so it's remarkable 00:19:44.480 --> 00:19:48.558 um you basically still have to be in 00:19:46.960 --> 00:19:50.400 charge of the aesthetics the the 00:19:48.558 --> 00:19:52.480 judgment the taste and a little bit of 00:19:50.400 --> 00:19:54.559 oversight maybe one one of my favorite 00:19:52.480 --> 00:19:57.679 examples of like the the weirdness of 00:19:54.558 --> 00:20:00.558 agents is um for menu genen uh you sign 00:19:57.679 --> 00:20:02.559 up with a Google Google account but you 00:20:00.558 --> 00:20:04.160 um purchase credits using a stripe 00:20:02.558 --> 00:20:06.319 account and both of them have email 00:20:04.160 --> 00:20:08.400 addresses and my agent actually tried to 00:20:06.319 --> 00:20:10.879 basically 00:20:08.400 --> 00:20:13.038 um like when you purchase credits, it 00:20:10.880 --> 00:20:15.760 assigned it using the email address from 00:20:13.038 --> 00:20:18.000 Stripe to the Google email address like 00:20:15.759 --> 00:20:20.319 there wasn't a persistent user ID that 00:20:18.000 --> 00:20:21.599 that uh for people it was trying to 00:20:20.319 --> 00:20:22.720 match up the email addresses, but you 00:20:21.599 --> 00:20:24.480 could use different email address for 00:20:22.720 --> 00:20:26.798 your Stripe and your Google and 00:20:24.480 --> 00:20:28.240 basically would not associate the funds. 00:20:26.798 --> 00:20:29.918 And so this is the kind of thing that 00:20:28.240 --> 00:20:31.519 these agents still will make mistakes 00:20:29.919 --> 00:20:33.038 about is like why would you use email 00:20:31.519 --> 00:20:34.558 addresses to try to crossorrelate the 00:20:33.038 --> 00:20:36.720 funds? They can be arbitrary. You can 00:20:34.558 --> 00:20:39.038 use different emails, etc. Like this is 00:20:36.720 --> 00:20:40.480 such a weird thing to do. So I think 00:20:39.038 --> 00:20:43.519 people have to be in charge of this 00:20:40.480 --> 00:20:46.000 spec, this plan. And um I actually don't 00:20:43.519 --> 00:20:47.359 even like the plan mode. I I would I 00:20:46.000 --> 00:20:48.240 mean obviously it's very useful, but I 00:20:47.359 --> 00:20:49.599 think there's something more general 00:20:48.240 --> 00:20:51.440 here where you have to work with your 00:20:49.599 --> 00:20:53.599 agent to design a spec that is very 00:20:51.440 --> 00:20:55.360 detailed and maybe it's uh maybe 00:20:53.599 --> 00:20:56.959 basically the docs and then get the 00:20:55.359 --> 00:20:58.719 agents to write them and you're in 00:20:56.960 --> 00:21:00.480 charge of the oversight and the top 00:20:58.720 --> 00:21:02.480 level categories, but the agents are 00:21:00.480 --> 00:21:04.000 doing a lot of the under the hood. And 00:21:02.480 --> 00:21:05.839 um so I think you're not caring about 00:21:04.000 --> 00:21:09.200 some of the details. So as an example 00:21:05.839 --> 00:21:11.519 also with um arrays or tensors in neural 00:21:09.200 --> 00:21:13.279 networks. Um there's a ton of details 00:21:11.519 --> 00:21:14.960 between PyTorch and NumPy and all the 00:21:13.279 --> 00:21:17.279 different like pandas and so on for all 00:21:14.960 --> 00:21:18.960 the different little API details. And I 00:21:17.279 --> 00:21:20.639 I already forgot about the keep dims 00:21:18.960 --> 00:21:22.798 versus keep dim or whether it's dim or 00:21:20.640 --> 00:21:24.000 axis or reshape or permute or transpose. 00:21:22.798 --> 00:21:25.440 I don't remember this stuff anymore, 00:21:24.000 --> 00:21:26.640 right? Because you don't have to. This 00:21:25.440 --> 00:21:28.000 is the kind of details that are handled 00:21:26.640 --> 00:21:30.000 by the intern because they have very 00:21:28.000 --> 00:21:32.079 good recall and but you still have to 00:21:30.000 --> 00:21:33.679 know for example that um you know 00:21:32.079 --> 00:21:35.279 there's underlying tensor there's an 00:21:33.679 --> 00:21:37.200 underlying view and then you can 00:21:35.279 --> 00:21:38.319 manipulate view of the same storage or 00:21:37.200 --> 00:21:40.080 you can have different storage which 00:21:38.319 --> 00:21:41.519 would be less efficient and so you still 00:21:40.079 --> 00:21:43.439 have to have an understanding of what 00:21:41.519 --> 00:21:45.759 this stuff is doing and some of the 00:21:43.440 --> 00:21:47.600 fundamentals um so that you're not 00:21:45.759 --> 00:21:50.798 copying memory around unnecessarily and 00:21:47.599 --> 00:21:53.279 so on but uh the details of the APIs are 00:21:50.798 --> 00:21:55.599 now handed off so it um you're in charge 00:21:53.279 --> 00:21:57.038 of the taste the engineering the design 00:21:55.599 --> 00:21:58.240 um and that it makes sense and that 00:21:57.038 --> 00:21:59.519 you're asking for the right things and 00:21:58.240 --> 00:22:01.279 that you're saying that okay that these 00:21:59.519 --> 00:22:03.918 have to be unique user IDs that we're 00:22:01.279 --> 00:22:06.079 going to tie everything to um and so 00:22:03.919 --> 00:22:07.360 you're doing some of the design and 00:22:06.079 --> 00:22:08.879 development and the engineers are doing 00:22:07.359 --> 00:22:10.158 the fill in the blanks and that's 00:22:08.880 --> 00:22:11.600 currently kind of like where we are and 00:22:10.159 --> 00:22:13.679 I think that's what everyone of course 00:22:11.599 --> 00:22:15.359 is seeing I think right now 00:22:13.679 --> 00:22:18.559 >> do you think there's a chance that this 00:22:15.359 --> 00:22:20.079 um taste and judgment matters less over 00:22:18.558 --> 00:22:21.359 time or will the ceiling just keep 00:22:20.079 --> 00:22:22.720 rising 00:22:21.359 --> 00:22:25.439 >> um yeah it's a good question I would 00:22:22.720 --> 00:22:28.319 Okay. 00:22:25.440 --> 00:22:30.240 Um, I mean, I'm hoping that the that it 00:22:28.319 --> 00:22:31.519 improves. I think probably the reason it 00:22:30.240 --> 00:22:33.200 doesn't improve right now is again, it's 00:22:31.519 --> 00:22:36.558 not part of the RL. There's probably no 00:22:33.200 --> 00:22:39.840 aesthetics cost or reward or it's not 00:22:36.558 --> 00:22:41.200 good enough or something like that. Um, 00:22:39.839 --> 00:22:42.480 I do think that when you actually look 00:22:41.200 --> 00:22:44.480 at the code, sometimes I get a little 00:22:42.480 --> 00:22:46.079 bit of a heart attack because it's not 00:22:44.480 --> 00:22:47.120 like super amazing code necessarily all 00:22:46.079 --> 00:22:48.480 the time and it's very bloaty and 00:22:47.119 --> 00:22:50.239 there's a lot of copy paste and there's 00:22:48.480 --> 00:22:52.480 awkward abstractions that are brittle 00:22:50.240 --> 00:22:55.759 and like it works but it's just really 00:22:52.480 --> 00:22:57.839 gross. Um, and I do I do hope that this 00:22:55.759 --> 00:22:59.839 can improve in future models. Um, a good 00:22:57.839 --> 00:23:02.079 example also is this uh you know micro 00:22:59.839 --> 00:23:04.639 GPT project which where I was trying to 00:23:02.079 --> 00:23:06.639 simplify uh LLM training to be as simple 00:23:04.640 --> 00:23:08.799 as possible. The models hate this. They 00:23:06.640 --> 00:23:10.960 can't do it. I tried to I keep I kept 00:23:08.798 --> 00:23:13.599 trying to prompt an LLM to simplify more 00:23:10.960 --> 00:23:15.519 simplify more and it just can't you feel 00:23:13.599 --> 00:23:18.240 like you're outside of the RL circuits. 00:23:15.519 --> 00:23:20.240 It feels like you're obviously you know 00:23:18.240 --> 00:23:23.759 you're pulling teeth. It's not like 00:23:20.240 --> 00:23:25.120 light speed. So I think um I do think 00:23:23.759 --> 00:23:26.400 that people are still remain in charge 00:23:25.119 --> 00:23:27.599 of this. But I do think that there's 00:23:26.400 --> 00:23:28.640 nothing fundamental again that's 00:23:27.599 --> 00:23:30.399 preventing it. It's just the labs 00:23:28.640 --> 00:23:31.038 haven't done it yet almost. 00:23:30.400 --> 00:23:33.360 >> Yeah. 00:23:31.038 --> 00:23:36.480 >> So I'd love to come back to this idea of 00:23:33.359 --> 00:23:38.158 uh jagged forms of intelligence. you 00:23:36.480 --> 00:23:39.519 wrote a little bit about this with a 00:23:38.159 --> 00:23:42.640 very thoughtprovoking piece around 00:23:39.519 --> 00:23:44.400 animals versus ghosts. Um, and the idea 00:23:42.640 --> 00:23:46.799 is that we're not building animals, we 00:23:44.400 --> 00:23:48.559 are summoning ghosts. Um, and these are 00:23:46.798 --> 00:23:51.440 jagged forms of intelligence that are 00:23:48.558 --> 00:23:54.000 shaped by data and reward functions, but 00:23:51.440 --> 00:23:57.038 not by intrinsic motivation or fun or 00:23:54.000 --> 00:24:00.000 curiosity or empowerment. Uh, things 00:23:57.038 --> 00:24:02.879 that kind of came about via evolution. 00:24:00.000 --> 00:24:04.480 um why does that framing matter and what 00:24:02.880 --> 00:24:07.120 does it actually change about how you 00:24:04.480 --> 00:24:08.960 build and deploy and evaluate or even 00:24:07.119 --> 00:24:12.558 trust them? 00:24:08.960 --> 00:24:13.759 >> Uh yeah, so yeah, I think the reason I 00:24:12.558 --> 00:24:15.200 wrote about this is because I'm trying 00:24:13.759 --> 00:24:16.640 to wrap my head around what these things 00:24:15.200 --> 00:24:18.319 are, right? Because if you have a good 00:24:16.640 --> 00:24:20.080 model of what they are or are not, then 00:24:18.319 --> 00:24:23.759 you're going to be more competent at uh 00:24:20.079 --> 00:24:25.918 using them. Um and I do think that um I 00:24:23.759 --> 00:24:28.558 don't know if it has I'm not sure if it 00:24:25.919 --> 00:24:29.520 actually has like real power. [laughter] 00:24:28.558 --> 00:24:33.119 I think it's a little bit of 00:24:29.519 --> 00:24:34.798 philosophizing. Um, but I do think that 00:24:33.119 --> 00:24:36.879 um 00:24:34.798 --> 00:24:38.639 I think it's just um coming to terms 00:24:36.880 --> 00:24:40.080 with the fact that these things are not, 00:24:38.640 --> 00:24:41.278 you know, animal intelligences. Like if 00:24:40.079 --> 00:24:43.199 you yell at them, they're not going to 00:24:41.278 --> 00:24:46.798 work better or worse or it doesn't have 00:24:43.200 --> 00:24:48.159 any impact. Um, and uh it's all just 00:24:46.798 --> 00:24:50.960 kind of like these statistical 00:24:48.159 --> 00:24:53.200 simulation circuits where the the 00:24:50.960 --> 00:24:55.519 substrate is pre-training so like 00:24:53.200 --> 00:24:57.919 statistics and then but then there's RL 00:24:55.519 --> 00:25:00.400 bolting on top. So, it kind of like 00:24:57.919 --> 00:25:02.159 increases the dispendages and um maybe 00:25:00.400 --> 00:25:04.080 it's just kind of like a mindset of what 00:25:02.159 --> 00:25:05.840 I'm coming into or what's likely to work 00:25:04.079 --> 00:25:07.759 or not likely to work or how to modify 00:25:05.839 --> 00:25:09.359 it. But I don't actually I don't know 00:25:07.759 --> 00:25:11.278 that I have like here are the five 00:25:09.359 --> 00:25:12.639 obvious outcomes of how to make your 00:25:11.278 --> 00:25:14.640 system better. It's more just being 00:25:12.640 --> 00:25:16.480 suspicious of it and um 00:25:14.640 --> 00:25:18.400 >> figuring out over time. 00:25:16.480 --> 00:25:20.000 >> That's where it starts. Um okay, so you 00:25:18.400 --> 00:25:22.559 are so deep in working with agents that 00:25:20.000 --> 00:25:24.880 don't just chat. They have um real 00:25:22.558 --> 00:25:26.240 permissions. They have local context. 00:25:24.880 --> 00:25:28.240 they actually take action on your be 00:25:26.240 --> 00:25:30.079 your behalf. What does the world look 00:25:28.240 --> 00:25:31.278 like when we all start to live in that 00:25:30.079 --> 00:25:34.000 world? 00:25:31.278 --> 00:25:35.599 >> Uh yeah, I think I think every a lot of 00:25:34.000 --> 00:25:38.240 people probably here are excited about 00:25:35.599 --> 00:25:40.240 what this agent uh you know native 00:25:38.240 --> 00:25:41.359 agentic environment looks like and 00:25:40.240 --> 00:25:42.480 everything has to be rewritten. 00:25:41.359 --> 00:25:44.558 Everything is still fundamentally 00:25:42.480 --> 00:25:46.798 written for humans and has to be moved 00:25:44.558 --> 00:25:48.319 around. I still use most of the time 00:25:46.798 --> 00:25:49.679 when I use uh different frameworks or 00:25:48.319 --> 00:25:51.359 libraries or things like that, they 00:25:49.679 --> 00:25:53.120 still have docs that are fundamentally 00:25:51.359 --> 00:25:55.678 written for humans. This is my favorite 00:25:53.119 --> 00:25:57.038 pet peeve. Like I don't uh why are 00:25:55.679 --> 00:25:58.400 people still telling me what to do? Like 00:25:57.038 --> 00:26:00.227 I don't want to do anything. What is the 00:25:58.400 --> 00:26:02.880 thing I should copy paste to my agent? 00:26:00.227 --> 00:26:04.798 [laughter] Like uh so it's just um every 00:26:02.880 --> 00:26:06.000 time I'm told, you know, go to this URL 00:26:04.798 --> 00:26:07.359 or something like that, it's just like 00:26:06.000 --> 00:26:10.319 ah [laughter] 00:26:07.359 --> 00:26:12.240 you know. [snorts] So um everyone is I 00:26:10.319 --> 00:26:14.079 think excited about how do we decompose 00:26:12.240 --> 00:26:16.159 the workloads that need to happen into 00:26:14.079 --> 00:26:18.240 fundamentally sensors over the world, 00:26:16.159 --> 00:26:20.080 actuators over the world. How do we make 00:26:18.240 --> 00:26:23.359 it agent native? Uh basically describe 00:26:20.079 --> 00:26:27.839 it to agents first. um and then have a 00:26:23.359 --> 00:26:30.158 lot of automation around um you know the 00:26:27.839 --> 00:26:32.959 um yeah around data structures that are 00:26:30.159 --> 00:26:34.400 very legible to the LLMs. Uh so I think 00:26:32.960 --> 00:26:36.960 um yeah I'm hoping that there's a lot of 00:26:34.400 --> 00:26:39.038 agent first um infrastructure out there 00:26:36.960 --> 00:26:40.960 and that you know for Menuguen famously 00:26:39.038 --> 00:26:42.400 when I wrote the uh not I'm not sure how 00:26:40.960 --> 00:26:44.159 famously but when I wrote the blog post 00:26:42.400 --> 00:26:46.240 about Menuguen [laughter] 00:26:44.159 --> 00:26:47.440 um a lot of the work a lot of the 00:26:46.240 --> 00:26:48.720 trouble was not even writing the code 00:26:47.440 --> 00:26:50.240 for Menugen it was deploying it in 00:26:48.720 --> 00:26:51.440 versell because I had to work with all 00:26:50.240 --> 00:26:52.640 these different services and I had to 00:26:51.440 --> 00:26:54.960 string them up and I had to go to their 00:26:52.640 --> 00:26:56.720 settings and the menus and you know 00:26:54.960 --> 00:26:59.759 configure my DNS and it was just so 00:26:56.720 --> 00:27:01.759 annoying and so that's a good example of 00:26:59.759 --> 00:27:04.480 I would hope that menu gen that I could 00:27:01.759 --> 00:27:05.839 give a prompt to an LLM build menu genen 00:27:04.480 --> 00:27:07.839 and then I didn't have to touch anything 00:27:05.839 --> 00:27:09.678 and it's deployed in that same way on 00:27:07.839 --> 00:27:12.158 the internet. Uh I think that would be a 00:27:09.679 --> 00:27:13.360 good kind of a test for whether or not 00:27:12.159 --> 00:27:14.960 uh a lot of our infrastructure is 00:27:13.359 --> 00:27:17.278 becoming more and more agent native. And 00:27:14.960 --> 00:27:19.360 then ultimately I would say yeah I I do 00:27:17.278 --> 00:27:21.278 think we're going towards a world where 00:27:19.359 --> 00:27:25.519 um there's agent representation for 00:27:21.278 --> 00:27:26.960 people and for organizations and um you 00:27:25.519 --> 00:27:28.720 know I'll have my agent talk to your 00:27:26.960 --> 00:27:30.798 agent uh to figure out some of the 00:27:28.720 --> 00:27:33.038 details of our meetings or or things 00:27:30.798 --> 00:27:34.798 like that. So, [laughter] 00:27:33.038 --> 00:27:36.720 um I do think that that's uh roughly 00:27:34.798 --> 00:27:37.839 where things are going, but um yeah, I 00:27:36.720 --> 00:27:38.240 think everyone here is excited about 00:27:37.839 --> 00:27:40.000 that. 00:27:38.240 --> 00:27:41.679 >> I really like the visual analogy of 00:27:40.000 --> 00:27:42.640 sensors and actuators. I actually hadn't 00:27:41.679 --> 00:27:43.038 thought of that. That's super 00:27:42.640 --> 00:27:43.440 interesting, 00:27:43.038 --> 00:27:45.359 >> right? 00:27:43.440 --> 00:27:47.679 >> Um okay, I think we have to end on a 00:27:45.359 --> 00:27:49.359 question about education. Um because you 00:27:47.679 --> 00:27:51.200 are probably one of the very best in the 00:27:49.359 --> 00:27:53.519 world at making complex technical 00:27:51.200 --> 00:27:56.319 concepts simple and deeply thoughtful 00:27:53.519 --> 00:27:59.759 about how we design education around it. 00:27:56.319 --> 00:28:02.480 Um, what still remains worth learning 00:27:59.759 --> 00:28:05.440 deeply when intelligence gets cheap as 00:28:02.480 --> 00:28:07.759 we move into the next a era of AI? 00:28:05.440 --> 00:28:09.200 >> Yeah. Uh, there was a tweet that blew my 00:28:07.759 --> 00:28:10.558 mind recently and I keep thinking about 00:28:09.200 --> 00:28:12.640 it like every other day. It was 00:28:10.558 --> 00:28:14.240 something along the lines of um, you can 00:28:12.640 --> 00:28:16.640 outsource your thinking but you can't 00:28:14.240 --> 00:28:17.679 outsource your understanding. 00:28:16.640 --> 00:28:21.278 And um, 00:28:17.679 --> 00:28:23.519 >> I think that's really nicely put. I so 00:28:21.278 --> 00:28:25.119 yeah because I still I'm still part of 00:28:23.519 --> 00:28:26.720 the system and I still I still have to 00:28:25.119 --> 00:28:27.918 somehow information still has to make it 00:28:26.720 --> 00:28:29.278 into my brain and I feel like I'm 00:28:27.919 --> 00:28:30.799 becoming a bottleneck of just even 00:28:29.278 --> 00:28:32.880 knowing what are we trying to build why 00:28:30.798 --> 00:28:34.639 is it worth doing uh how do I direct you 00:28:32.880 --> 00:28:37.840 know how do I direct my my agents and so 00:28:34.640 --> 00:28:39.759 on so I do still think that ultimately 00:28:37.839 --> 00:28:43.278 something has to direct the thinking and 00:28:39.759 --> 00:28:44.720 the processing and so on and um that's 00:28:43.278 --> 00:28:46.240 still kind of fundamentally constrained 00:28:44.720 --> 00:28:47.679 somehow by understanding and this is one 00:28:46.240 --> 00:28:49.599 reason I also was very excited about all 00:28:47.679 --> 00:28:51.360 the LM knowledge bases because I feel 00:28:49.599 --> 00:28:53.199 like that's that's a way for me to 00:28:51.359 --> 00:28:54.959 process information and anytime I see a 00:28:53.200 --> 00:28:56.798 different projection onto information. I 00:28:54.960 --> 00:28:58.720 always like feel like I gain insight. So 00:28:56.798 --> 00:29:00.319 it's really just a lot of prompts for me 00:28:58.720 --> 00:29:03.360 to do synthetic data generation kind of 00:29:00.319 --> 00:29:05.038 over over some fixed data. Uh so I I 00:29:03.359 --> 00:29:06.719 really enjoy uh whenever I read an 00:29:05.038 --> 00:29:07.759 article I have my uh you know my wiki 00:29:06.720 --> 00:29:09.519 that's being built up from these 00:29:07.759 --> 00:29:12.640 articles and I love asking questions 00:29:09.519 --> 00:29:15.119 about things or um and I I think that 00:29:12.640 --> 00:29:17.278 ultimately these are tools to enhance 00:29:15.119 --> 00:29:18.558 understanding in a certain way and this 00:29:17.278 --> 00:29:20.079 is still kind of like a bit of a 00:29:18.558 --> 00:29:22.879 bottleneck because then you can't direct 00:29:20.079 --> 00:29:25.359 the you can't be a good director if you 00:29:22.880 --> 00:29:26.960 still uh because the LM certainly don't 00:29:25.359 --> 00:29:28.959 excel at understanding you still are 00:29:26.960 --> 00:29:31.038 uniquely in charge of that. So, uh, 00:29:28.960 --> 00:29:32.558 yeah, I think, uh, tools to that effect, 00:29:31.038 --> 00:29:33.200 I think are incredibly interesting and 00:29:32.558 --> 00:29:34.558 exciting. 00:29:33.200 --> 00:29:36.159 >> I'm excited to be back here in a couple 00:29:34.558 --> 00:29:38.480 years and to see if we've been fully 00:29:36.159 --> 00:29:40.000 automated out of the loop and they 00:29:38.480 --> 00:29:41.440 actually take care of understanding as 00:29:40.000 --> 00:29:42.930 well. Uh, thank you so much for joining 00:29:41.440 --> 00:29:44.950 us, Andre. We really appreciate it. 00:29:42.930 --> 00:29:44.950 [applause]