Executive demos can make AI look more capable than it is

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 41:42

   done across data centers they're just hiring and fire executive ai psychosis basically it is so think of it like this you are a guy who makes 50 million dollars of rsus and cash a year probably more on the stock side and you have this chat bot every idea you've ever had it's like you're a genius you're the smartest man you're so that that's exactly right you should fire all of them you don't need hr and these people they get this tool and they're able ai so mobitar fantastic you're very pro like he likes ai fantastic youtube guy has a good point that ai is really good at a demo it's good at making like a very rough thingy and that's all it takes to brainwash an executive Nick Suresh quote all you need for an executive is to cobble something together looks close enough and these people don't do work most CEOs go to lunch leave lunch and

2. 42:28

   barely read their emails you know one thing I self-driving cars are obviously a very relevant they're interesting yeah and so you know self-driving cars the industry tried to get us very excited you know like a decade ago and they weren't ready because there are a lot of edge cases i think there's similarity to what you're saying with you know the demo idea which is basically a demo it's like it's mostly there and then people way underestimate how much it takes but imagine it all the way imagine if they used waymo and just they just fired all the cap drives every single one it's like that's good enough and then you just every time it rained you would just have the waymos jumping in but then it'll be a great opportunity for every business that doesn't like that that's just like competitive dynamics in business what business what do you

3. 43:08

   mean what do you mean like businesses that hold on to their if your cases that all these companies are going to find that they needed those employees the companies that don't react this way will outmaneuver them like that yeah probably that's competition yeah that's in fact in the erp space of the oracle they're finding a bunch of churn to other erp enterprise resource planning companies that's happening i think meta could die in 10 or 15 years because of the things that zuckerberg has done to it i think microsoft people at microsoft are miserable because copilot copilot co-pilot. Satya Nadella, MBA. Sundar Pichai at Google, MBA. McKinsey. Mark Zuckerberg ran on MBA. Tim Cook, MBA. I think John Ternis is MBA. Point is, these people don't experience problems, so they have no idea how to solve them. Sorry, you're saying Zuckerberg has an MBA? He doesn't

4. 43:49

   have an MBA. He hasn't written a line of code since 2006, apparently. Okay. Which is really funny. But getting back to the main point, which is, I think people credit CEOs with Thank you. Way more intelligence than they have. I also think they think that these guys at the top, like Google, for example, they'd never make a big mistake. Go and use Google right now and tell me if that looks like a good product or it looks like if a startup released a search engine that randomly responded like a weirdo to you in 2014, Michael Arrington would have them shot.

5. 44:17

   I just, I don't get the, like, moral lens. Like, if these people are wrong, they're going to lose a bunch of money. And, like, if you don't like them, you can be happy about that. Like, what's, what's the issue? Like, they're trying. tens of thousands of people losing their jobs the people across the world who have spent three years being scared being told this thing that cannot do it but regardless still as if it could being fired being threatened they open the newspaper every day there's a completely fictitious story like strategy for them like are you long like just like human heavy businesses or like long humans i don't invest in the market i people have this weird thing like oh what are you showing i'm not i like writing i actually care about this stuff and i think people love love that i'm just saying what's

6. 45:00

   like the positive vision for for the world what what do you mean well i'm just saying it's very focused on on sort of like the negative like my view of the ai industry is that people are building interesting stuff that they think i mean i do think it will help with drug discovery research all sorts of not really generative though you're you're mushing there are there are different some of the things that's the thing though but there are different techniques like that's the marketing technique though they want you to think all ai is one blob when ai is many different things and the thing everyone's so your your criticism is specifically text generative generative ai so video video but like the techniques that like allow for alpha go and alpha i do not believe is generative right but there are related techniques involved and like

7. 45:45

   certainly demis like who is creating them is across all this stuff are you skeptical of him Yes. So Demis, Nobel, Nobel scientist. Yeah, but chemistry. And also AI techniques within chemistry, not generative. Every time I see him on stage, I think that in 10 years, the computer is going to wake up. And irresponsible, disgusting. I think it's disgraceful to constantly do this thing of AGI is just 10 years away. You want to know why they're talking about AGI? Because the current stuff isn't good enough.

8. 46:20

   I don't find a lot of coders who don't, like coders seem to believe that this is impacting their career. Like, do you disagree with that? Some are. Carl Brown, the Internet of Bugs, puts it where it's like, it makes the easy things easier, the hard things harder. There are people that use it who just ignore the outputs. And the thing is, you can make really clumsy fudge software that sort of kind of works until it doesn't before LLMs.

9. 46:45

   There was a bunch of software within major hyperscalers that just didn't work. This has compounded that problem. the problem with llms one of the most dangerous things about them is they can make an incompetent imbecile seem semi-competent they can hang around and do more damage it's far more damaging to have someone with an llm who can't code or can barely code who is able to pretend they can and get responsibility than it is for that person to just not be able to do that not to sound like a tech person but what you're describing is you know democratizing intelligence you're you're saying that there's someone who doesn't know how to code that you're giving them more capability than they had before and that there's something wrong about that because you want to keep it as the reserve of the super intelligent coder you are crediting the models with intelligence they don't have you're

10. 47:31

    not giving these people intelligence you're giving them a very complex autocomplete which is still what it is and people love that when i say that you are giving them something that can do things that it's seen before it can do shit off of stack overflow or what have you they can do things from its training data. And it can, when you hit it enough, produce something that will work. Work does not mean stable. Work does not mean functional even. Work means functional enough that no one gets on their back. It is not democratizing coding.

11. 48:00

    To give an imbecile something that is as dumb as it, but able to do a better impression of someone who can do their job. And I think the problem is, is that people conflate those things.