Episode #17 – News News News

This episode of BigCheese.ai features a mix of tech and entertainment discussions. We kick things off with the grand opening of The BigCheese Podcast Studio in Indianapolis, a space offering professional facilities for content creators. Then, we take a trip down memory lane with movie quotes and explore the challenges and possibilities in AI development, including OpenAI’s recent advancements and Groq’s innovative hardware. Stay tuned for more exciting topics next week!

Show Notes

Participants:

Sean Hise
Brandon Corbin
Jacob Wise
Andre Harakas

Topics Covered:

The BigCheese Podcast Studio

DeAndre has created a podcast studio for professional use in Indianapolis.
The studio contains various dynamic backgrounds suitable for different types of content creation.
The studio is equipped with high-end cameras and lighting setups, making content production easier for podcasters.
The space serves as a prototype for a larger version that will be introduced soon.

Relevant Quote: "We've got five separate walls in here where you can flip around a wall and have a completely different background." - DeAndre Herakis

Remembering Movie Quotes

Sean Hise celebrates the memory of movie quotes.
Participants share anecdotes connecting day-to-day conversations with movie scenes.

Relevant Quote: "Every time somebody says a statement, I connect with the movie." - Sean Hise

AI News and Model Discussions

The team catches up on the latest AI news after a series of guest episodes.
They discuss the recent addition of memory functionality to AI models.

Relevant Quote: "OpenAI is introducing memory now, though, right?" - Brandon Corbin

Groq – AI-Based Company

Introduction of Groq.com, an AI company formed by former Google engineers.
Groq.com has developed LPUs (Language Processing Units) for high-speed responses in language models.
Participants highlight the impressive speed and efficiency of Groq.com's AI platforms compared to other models like ChatGPT.

OpenAI's Ambitions and Challenges

Discussion about the future of OpenAI.
Concerns raised on whether OpenAI will sustain its competitive edge in the rapidly evolving AI industry.
Speculations around the company's potential survival and the significance of its 'memory' feature.

The Concept of Memory in AI

The panel debates the importance of memory in AI models.
Reflections on how AI needs to be developed further in areas beyond generating content.

Sam Altman and the AI Hardware Overhang

They ponder upon Sam Altman's statement on the future costs for AI-driven chips.
Considerations on the amount required and the potential implications of creating such advanced hardware.
Discussion on Altman's approach to AI and its alignment with OpenAI's mission.

Gemini's Image Algorithm and the Hiatus

Google's decision to shut down Gemini's image algorithm due to bias concerns is brought into focus.
Arguments centering on the challenge of maintaining historical accuracy in responses from AI models.

Don't forget to tune in next week for more discussions on AI advancements and the digital landscape.

Welcome back to the Big Cheese AI Podcast. I am the world's second best moderator behind Jason Calacanis. My name is DeAndre Herakis. I'm joined by the usual suspects, Sean Hise, Brandon Corbin, and Jacob Wise. And we have yet another setup, and we'll continue to get better. Yeah, new set. We got the band back together fully. Everyone's here today. Yeah, we've got a bunch of stuff to talk about, too. So before we begin, I think we should talk a little bit about the studio that you're building. Because there's a lot of work that goes into this. As you guys have-- if you've been following us, you know that our backgrounds have changed quite a bit. But that DeAndre here has been setting up a-- for lack of a better word, just a podcast studio here in Indianapolis that could potentially be used by other people who would want to be podcasting. Sure, a little life update, everybody. I have accepted a COO role with a marketing company called Maverick Marketing. They know about this. I would hope so. That's good that they know about it. As a part of that, we've had this really cool space in Keystone at the Crossing, Route 465, the Indiana Wesleyan building. And we used it for my startup's office. And we just were like, hey, let's turn this into a content studio where people can come in. So we've got five separate walls in here where you can flip around a wall and have a completely different background. So you can create really professional-looking podcasts. We've got amazing cameras and all the lighting's there. So it's really, really fun. You even have areas where you can go and you can shoot solo content, where you're kind of sitting in front of this cool, dynamic background. It looks like you're in this really nice space and you spend all this time. But really, you walk in, you hit a button, and we can actually do the edits for you and help you deliver some high-level quality content with ease. Because everybody wants to make cool content, but it's hard to actually build up the setups. And so that's kind of the vision for this space. And we have even more updates coming out, because probably next week I'll announce what even bigger things-- this is like the prototype of what we're building. The next version is going to be this times 10. It's going to be much bigger. I can't wait to talk about it. But thanks, Brandon, for the shout-out, because it is kind of exciting. I like what we've done here so far. Great job. I like what you've done here. What movie is that from? That would be Wayne's World. Was it? So you're the guy who knows all of the zingers. Every time somebody says a statement, I connect with the movie. It's from-- and then I can't help myself but say it out loud. So my old business partner was the same way, where he just knew every single zinger that came from every movie. And I am like the absolute-- so he would be like, ba-ba-ba-ba. And I'm like, nothing. That scene's from Wayne and Garth. Garth has that big, huge margarita glass at the-- and then they're at the restaurant. And the guy's got, I have two cashier's checks for 0.27 per million tokens, where ChatGPT4 is like 7 trillion. How do you raise $7 t r i ll i o nb y S am A lt man ? I t h o ug h tt ha tw a s a - - I t h o ug h t, w h e n I h e a r d t ha t f or t h e f i rs tt im e I w a s a tG o, t h eyscre w e d u p . T h ey m u s t ha v e m e an t bi ll i o nan d n o tt r i ll i o n, b ec a u se t ha t^{'} s bi gg er t han A ppl e an d G oo g l e^{'} s ma r k e t s ha reco mbin e d . R i g h t . A n dd i gg in g in t o t ha t s t ory a l i ttl e bi t, i tt u r n so u t i tw a s n^{'} t e x a c tl y h e^{'} s t ry in g t or ai se t ha t . T hi s i s t h ee x p ec t e d cos t o v er a llt o d o t ha t kin d o f - - F or t h e f u ll - - - - e n d e a v or . A n d a l o t o f t ha t i ss p ec u l a t e d t o b e in p o w er . S o l ik e, j u s t f orsc a l e, t h e l a r g es t c hi p man u f a c t u rer in t h e w or l d, T ai w an - - m y n o t es - - an y w a y, T ai w an eseco m p an y, t h ey a re - - L ik e TSCM or w ha t e v er i t i s . Y es, t ha t^{'} scorrec t . T ha t^{'} s a g eo p o l i t i c a l i ss u e . O h, t ha t s t u f f^{'} s in T ai w an r i g h t n o w . Y e ah, e x a c tl y . B u tt h e i rre v e n u e - - L e t m ec h ec km y n o t es . I t hinki tw a s l ik e$ 500 million-- or billion, I mean, last year. $500 bi ll i o n ? Wh y mak e bi ll i o n s w h e n yo u co u l d mak e - - N o, g l o ba l - - sorry . Gl o ba l c hi p s a l es l a s t ye a r w ere$ 526 billion. Jeez. So global chip sales were 1/14 of what he's suggesting it's going to cost to meet demand for the future of AI. That was one thing I was just like, OK, these numbers are not-- They're not today or tomorrow. He's looking at it as a plateau, where there's these people, like these guys are coming out with these new chips. Yeah. You know what I mean? Well, the thing about it is he wants to build-- this is all for a new chip. So the new chips would be this startup he's got-- God, I have to check my notes. Rain, Neuromorphic. Why do I keep doubting this guy and saying he's going to fail? I know. Neuromorphic. It's-- Yeah. Yeah, the guy hasn't-- he's never failed. So basically, these new chips are analog instead of digital. And analog chips are much better at actually processing the type of stuff that LLMs and the work that they're doing. It's faster, lower power, and just cheaper in general to produce. So that's where-- he wants to get this overhang of compute power, right, and then throw software at it. But that's a new term, by the way, I want to talk about. I think it's called hardware overhang. And it's this concept of-- think about today. When you build software, you have to-- the hardware generally has to catch up, right? We are constantly at this threshold. Software is always outpacing hardware. And then the hardware eventually catches up. So there's not much damage software can do. But what if hardware all of a sudden jumped 30, 40, 50 years into the future? We had a lot of access capability. Well, then you could build some software that could be very, very good. Or it could be very, very bad, dramatically bad, right? And it could hit a global scale. So that's the concern there is you have the ability to do very disruptive things. Like launch a DDoS attack that kills every single network just because you have so much power. Exactly. It's limiting today. You couldn't do something like that because there's only so many computers, right? But where it's kind of-- so not only does he have to figure out how he's going to raise the money. Yeah. Then we have the problem of actually finding the company who's going to produce the chips, most likely. It seems like he's playing this game that might be more like guys that are wearing Metallica shirts in a basement are going to solve, like without raising any money. There's somebody that's going to build Doom, right? Nobody built Doom. They built Wolfenstein. They built Doom, right? They just did it because they loved it and they figured it out. Like some of these kids are going to figure this crap out, right? That's kind of how I feel about what Grok is, though. That's what I'm saying. Right. Those guys, they're already there. They already have the manufacturing partnerships. They already know what they're building towards. And it's like, OK, why wouldn't Sam, instead of you saying, we need a new chip, just say, OK, we're going to use these language processing units as our primary foundation. And I'm just going to go and invest a couple hundred million dollars into Grok. I think his ambition is the issue. It's like he wants to do it on a scale so that we could do this worldwide in an instance and flip a switch kind of thing. Well, let's be honest. It also aligns with his style. Like he started opening AI and literally went dark for like five years. He didn't talk to anybody. He didn't tweet like more than once. They were just in the lab. That's his style. Like he's going to-- let's get a bunch of runway. Let's not talk to anyone. And let's do this. So maybe it just aligns with his natural way of doing things. Yeah, but something else I read in the article you sent me was this is not in line with open AI in general. [LAUGHTER] Their whole thing is like, oh, AGI is not possible because we don't have enough hardware, essentially. And then he's like, you know what we could do? We could manufacture a shit ton of hardware. And it's like, OK-- It's like him at a board meeting. OK, so aligning with their mission of making open AI safe and standard. But what if we just make a billion dollars? [LAUGHTER] Right, yeah. So there's a kind of a controversial thing that I think is actually interesting for a different reason than every other clickbait article you're going to read on this. But Google this week shut down Gemini's image. Gemini is the rebirth of Bard through a rebrand, essentially, right? I mean, I'm not sure. Good idea by them, right? Bard was horrible. Bard was a horrible name. Horrible. Literally, the worst fucking name. Just go to bard.google.com. Digga-digga-dee. Anyways, they call it Gemini. And Gemini is their chat GPT. It's their interface for doing AI stuff. It's very similar. You can prompt it. It can generate images. It can generate text. There's some nuances there. It's Google, so obviously there's polish there. But they shut down their image generation system. And it was because of a-- I opened up Twitter. And like Brandon said, Twitter is, for me, my feed is much different than it used to be. But traditionally, my feed is very self-important, self-righteous tech bros or really good developers that have a big following. And so they just leverage their following. People that are really involved in mostly engineering, software engineering and tech, and influencer types. But they shut down this image generation system because it was-- the news articles and the hype was that basically it wouldn't generate images of white people. It was too woke. Don't say that. I know. No, but no, no, no, no. I say that only in jest, right? That it was quote, unquote, "too woke." Because all of these large-- everybody's afraid of being not inclusive anymore. And so they kind of put in all of these guardrails to make sure that if I'm asking for something that's racist, don't do it. Well, Washington Post did a study last year. And it was basically like, hey, they went and analyzed, does AI have bias? And they said, hey, show me a picture of a successful person. And nine times out of 10, it was a white guy in a suit. And it said, OK, show me a picture of a social worker. And it was usually an African-American woman. And so there's this implicit bias in AI that may not be built in from, OK, we're going to scan all the data, oh, and then put this prompt in. It's like, OK, what in the real world, white people are rich, and black people are-- it was just there's an implicit bias in the nature of the data somehow, some way. And it's very, very, very, very damaging, potentially, that AI could perpetuate bias. Because in this world, we're trying to not be like that. Well, Gemini's-- the use case that got exposed on the internet was historical information being misrepresented. So people said, hey, generate an image of an 1800 senator. And the prompt response, here are pictures of a diverse set of 1800 senators. And as we all know, there was no diversity in the 1800s in the Senate. And also, they said, please generate image of Nazis. And they generated images of Indian, African-American, people of color that were-- A very diverse set of Nazis. Quit dropping your iPad. But it got-- Andre and I had this conversation on the way to work out this morning, on the way home from working out. And I was like, I understand that. Like, I think that Verve-- if you're going to read about this, read the Verge article. It's more balanced. It's like, get off the clickbait crap. Like, it's like, yeah, we can understand this. It delivers both sides of the story. But to me, the issue that I have is misrepresenting historical facts. I think that we see it-- like, I use the analogy. When I was growing up, we had the history book. And that history book talked about the Trail of Tears. And the Trail of Tears, it was one paragraph on the side of the bottom of the book on the page about this one thing that-- the Trail of Tears was really bad. And it was really sad. And a lot of people died. The end. Moving on. And it's like-- And it's a race. But talk about-- here's a story about the white senator that did great, and here's the-- there's this concept of bias in the historical information that's presented. And I think that what you're seeing is-- what you don't want to see is if someone is using a tool-- and this is my take. This is my take. If you're using that as a search engine or an encyclopedia, you can't-- you've got to think about that. Because fact-- it's like Indiana Jones in "Raiders of the Lost Ark." Archeology is the search for fact. Facts. Yeah, I'll jump on that. Because yes, it's like, OK, for us, it all makes sense. We had history class. We had history books. We had actual books of history that we could go back and read. And we knew that this was, in general, accepted as canon and absolute. Well, in a world in the future, a few years from now, where I grew up on LLMs using these technologies, if there's any bias at all, all of a sudden, the history is completely convoluted. Like, where is the source of truth? Well, somebody could just inject a prompt that's like, hey, don't talk about this. And that's kind of what's happened in the real world. So maybe it's actually mimicking the real world a little bit. It's just expanded now. It's much easier to-- instead of a one book in one classroom that 100 kids get to see over the course of a few years, now it's like 100,000 kids because they're all using the same LLM on the internet. It just expanded the issue so much more. That's a concern for me. It is. It's absolutely a concern. And the fact is that we should not be re-whiting history. Not whiting, but I mean, history is what it is. It is what it is. And we did some horrible shit, and we need to accept that. We need to accept-- and move on from-- not move on, but realize that that's what it was. But we should not have these AIs basically trying to rewrite history. Yeah, and I'm not sure that-- I think if you're talking about historical-- it's like when Google first came out. People were like, do not use Google with your paper. You have to cite your source on LexisNexis. I was like, what the hell is that? But there's a point to it. If you just say that, oh, the LLM is the source of truth-- I think there's a very high likelihood that that is the case. And-- Like people are going to just rely on LLMs for the source of truth. I do think so. OK, so have you guys heard of perplexity? I was trying to-- No. OK, so basically, perplexity is trying to solve this by doing real-time internet searches and then getting that-- basically, taking that web page, find the most relevant results to your query, putting it in a rag, and then reading that based off your query, and then responding back from that article. So it's more or less using the LLM to discern what it is you're trying to ask, then get that extracted information from sources on the internet. I think that's probably where Google has to take this. These were never intended to be used for searching. That was the first thing everybody always did. I know. It's like-- It's the first thing-- It's all we knew. So one of my clients, the first thing-- their board logged into ChatGPT at a board meeting. And these people are-- this is not a joke board. The first thing they did was Google themselves on ChatGPT. Who is Brandon Corbin? I haven't done that yet. You haven't done that? Oh, my god. Yes, they search for facts. They search for facts. And that's not-- the use case for LLMs is creation. It's creation. Now, I will say-- so when I was playing with Gemini, sources were included way more than ChatGPT. And I think that that's where we're ultimately going to have to get is that these things need to be grounded in a source. The labeling mechanisms that they're doing when they're doing this stuff, all the auto-labeling and all that, just attach that source. Yeah. Attach that source. Just tag it with a source. Because-- The problem is a lot of times that source is like 150,000 different sources. Or that source is completely nefariously injected into the model, which is another concept for another day when we talk about copyright issues. My conclusion on all of this is in a world where, OK, the future, we've got all these new chips, and everyone's using LLMs. And they're way better. And we're creating everything. And the history and truth is now convoluted. And so now things are just what we create. I think it becomes so individualistic. And your truth is just like-- the way that you're experiencing the world and history is so specific to you. It consolidates power to an unbelievable level. If everybody is using the WeChat in the world for AI, for LLMs, and they're just like, we really don't need those books. They ain't fast. We really don't need that Google anymore. We just use our-- We don't need real links. We just need the information. Well, of course, I'd love that information. Can I get Uber Eats as well? Can I get-- I don't know why I said that. But can I just get everything? Can I just get it all? Like path of least resistance. But now, it's kind of scary if everything's coming from one place. And ultimately, Elon was actually right. Yeah, well, he's a smart man. He's a very smart man. Broken clock. That's right, twice a day. Yeah, even though I can't stand him anymore. Yeah. Yeah. Good lord. Everybody, this was the Big Cheese AI podcast. I am DeAndre Harris. You're just joined by Sean Hize, Brandon Corbin, and Jacob Wise. Catch us next week.

Social

Show Notes

Subscribe

Let's Connect