Using a Swarm of AI Models for Complex Results - The AI Podcast

Summary

The AI Podcast: Episode Summary

Episode Title: Using a Swarm of AI Models for Complex Results
Release Date: December 1, 2024
Host: The AI Podcast

Introduction

In this episode of The AI Podcast, the host delves into the burgeoning landscape of artificial intelligence by spotlighting a groundbreaking Japanese startup, Saskana AI. With a fresh infusion of $100 million in Series A funding, Saskana AI is positioning itself as a formidable player in the global AI arena, challenging giants like OpenAI and Anthropic. The discussion navigates through the company's innovative strategies, funding dynamics, technological advancements, and its potential impact on Japan's AI supremacy.

Saskana AI: An Emerging Powerhouse

Saskana AI, founded by ex-Google researchers David Ha and Lillian Jones, alongside former diplomat Ren Ito, has recently secured an impressive $100 million in Series A funding. The host expresses enthusiasm about the company's potential, stating:

“[...] they have just raised $100 million more and they're essentially taking on OpenAI. They're taking on Anthropic as what they're calling a world class AI lab. This is a fascinating company.”
— Host [02:10]

This substantial investment underscores confidence from major industry players, positioning Saskana AI as a significant contender in the AI sector.

Funding and Strategic Investors

The funding round attracted prominent investors, including New Enterprise Associates (NEA), Coachella Ventures, Lex Capital, and tech giant Nvidia. The inclusion of Nvidia is particularly noteworthy:

“Nvidia just puts money into any new AI company because they know that all money in AI eventually gets funneled back to them [...]”
— Host [03:45]

Nvidia's investment is strategic, ensuring that advancements within Saskana AI lever their hardware and infrastructure solutions, creating a symbiotic relationship beneficial to both parties.

Innovative Technological Approach

Swarm of AI Models

Saskana AI adopts a novel approach by leveraging a swarm of smaller AI models to achieve complex results. The host draws parallels to OpenAI's GPT-4:

“They have some that are capable of generating Japan's kind of traditional Yukiko E Artwork [...] similar to what GPT-4 is doing with its 16 different expert AI models.”
— Host [07:30]

This method allows for specialized processing, enhancing the versatility and accuracy of AI outputs by directing tasks to the most appropriate model within the swarm.

Native Japanese Data Sets

Distinctively, Saskana AI builds its models using native Japanese data sets. The host remarks on the cultural and linguistic significance:

“They are building this on a native Japanese data set [...] I'd be curious to see how AI models differ if they're built off of it exclusively Chinese data sets or Japanese data sets or French data sets.”
— Host [10:15]

This focus ensures that the AI systems are finely tuned to Japanese language nuances, cultural contexts, and societal norms, potentially offering more relevant and precise interactions for Japanese users.

AI Scientist Research

Saskana AI has developed an "AI Scientist," an advanced LLM system designed to automate the entire research lifecycle. This includes ideation, coding, experimentation, result summarization, and even drafting research papers. The host highlights:

“They created an AI scientist that automates the entire research lifecycle [...] it will even do, you know, conduct some peer review stuff.”
— Host [12:50]

This innovation could revolutionize research practices, expediting discoveries and enhancing collaborative efforts within scientific communities.

Nature-Inspired Development

Embracing biomimicry, Saskana AI's development philosophy draws inspiration from natural ecosystems. The host explains:

“They are really looking to scale up their nature inspired approach to AI development [...] like a nature, animal ecosystem kind of approach.”
— Host [15:20]

This strategy aims to create more resilient, adaptive, and efficient AI systems by mimicking the complex interdependencies found in nature.

Partnership with Nvidia: Enhancing Capabilities

Nvidia plays a pivotal role in Saskana AI's expansion by providing cutting-edge GPU systems and access to their data centers in Japan. The host comments on Nvidia's support:

“They’re going to be offering a lot of this infrastructure support [...] providing all of the latest GPU systems to help them actually train and develop everything.”
— Host [18:05]

This partnership ensures that Saskana AI has the necessary computational power and infrastructure to sustain rapid growth and innovation.

Impact on Japan's AI Landscape

Saskana AI's efforts are part of a broader movement to rejuvenate Japan's position in the global AI sector. The company's mission, as articulated in their blog, is to:

“Help Japan regain a technological edge in AI and increase its global competitiveness [...] assist Japan, its institutions and its citizens to overcome its AI challenges in the road ahead.”
— Saskana AI Blog [20:40]

The host underscores the significance of this mission:

“Japan's trying to change that and this company is kind of a big step in that direction.”
— Host [21:10]

With Japan historically being a technological leader, Saskana AI's rise could herald a new era of AI innovation emanating from the region.

Competitive Landscape: OpenAI in Japan

The episode also touches on the competitive dynamics with OpenAI's presence in Japan. OpenAI has established a Tokyo hub and released a custom GPT-4 model optimized for the Japanese language. Additionally, rumors circulate about a forthcoming model, "GPT Next," purported to be a significant advancement over GPT-4.

“OpenAI has a Tokyo hub [...] they've already released a custom GPT4 model optimized for the Japanese language.”
— Host [24:25]

This competition highlights the intense race to dominate the AI landscape in Japan, with Saskana AI and OpenAI both striving to leverage localized advancements.

Conclusion

The host concludes with a strong endorsement of Saskana AI's potential and its innovative approaches:

“This is a very innovative AI lab that I would definitely recommend keeping your eyes on.”
— Host [27:00]

Emphasizing the company's unique strategies and substantial backing, the host positions Saskana AI as a key player to watch in the evolving AI ecosystem.

Additional Resources

For listeners interested in harnessing the power of podcasting, the host mentions a specialized podcast course designed to help both budding and established podcasters grow their audience organically. While advertisements are to be skipped in summaries, interested individuals can find more information via the podcast's show notes.

Stay Updated:
For regular insights and in-depth discussions on artificial intelligence, subscribe to The AI Podcast and stay tuned for more episodes exploring the frontiers of AI technology.

Transcript

A (0:00)

There is a new startup called Saskana AI and they have just raised $100 million more and they're essentially taking on OpenAI. They're taking on Anthropic as what they're calling a world class AI lab. This is a fascinating company. There's a ton of stuff going on here. I'm excited to break down exactly what they're doing, why they're getting $100 million. This is our series a round, what they plan on doing that I wanted to say Today's episode is sponsored by my very own podcast course. If you have ever thought of starting a podcast or if you already have a podcast now that maybe isn't getting very many listens, you want to help grow it. I have a podcast course that explains how to start a podcast from start to finish and also how to grow an existing podcast, how I was able to get over 4 million podcast downloads, 100% organic. I didn't run ads, I hardly did. Social media is 100% organic and I've been able to replicate this over four podcasts that I personally run. And I've helped a ton of other people do this as well. So I have successfully discovered the strategies, which was not easy. It took me five years. My first podcast barely got 20 listens per episode until I really went hard and figured out a lot of different things. So my goal with this podcast course is to make something that can essentially let you skip the the thousands of mistakes I made over the first five years, the over $15,000 I've spent on podcasting equipment, studio, build out team and get to starting a profitable, fast growing podcast right from the start. So if you're interested in starting a podcast, check out the link in my description to my podcast course. It will not disappoint you. It will be the best podcast course you ever take in your life. If you're interested in starting a podcast. So go check that out. Link is in the description. Get into the episode today. What I think is really interesting is that this is actually a just a couple days after China's Alibaba. There's a whole bunch of headlines in regards to the Quinn 2 VL AI model. And so a lot is going on right now as far as AI. There's tons of frenzy taking over the world and it's not just American companies. So in regards to that Saska Seskana AI this is actually a Japanese startup, it's founded by some former researchers over at Google. There's David Ha and Lillian Jones who is a also they Have a former diplomat, Ren Ito. So all of them are working on this together. They just announced this $100 million Series A round of funding and I think it's impressive. They have a bunch of really big players in the industry that have invested into this. There's New Enterprise Associates, nea, right. One of the top VA venture capital firms in the us, Coachella Ventures, Lex Capital. They also have Nvidia who's jumping in at this point. I feel like Nvidia jumps in on every single large funding round because they know that all money in AI eventually gets funneled back to them, whether that's through Compute or you know, just buying their GPUs or you know, like spending money on a platform like Amazon or Google Cloud that is using Nvidia hardware. So like Nvidia just puts money into any new AI company because they know that that money is going to eventually like it all trickles back to them. So I think it's a smart move by Nvidia. But it is funny. I think this is going to say a lot as far as where the state of the industry is going right now. They just announced this round on X. They said, quote, excited to announce our Series A. We raised more than $100 million to grow Sascana AI into a world class AI lab in Japan. We're going to really push the frontiers of what's possible with AI. Then they also said as a founder mode startup we operate much faster than most frontier AI labs at a global level. So this kind of funny they're they're making, I don't know, they're, they're jumping on the founder mode trend which if you haven't been on X, everyone is, is talking about the difference of founder mode versus manager mode. Founder mode being, you know, someone that really is a, is a founder of the startup and is really going crazy pushing it to be successful. So overall I think it's, you know, a very bullish sign. A lot of people are, you know, congratulating them and stuff on X and whatnot. So that was kind of their announcement. So what have they actually been up to? What do they actually do? They had a really big appearance last year because they had some very high profile founders and it's really kind of an interesting approach to how they're developing some of their AI models, some of their foundational models. Really their idea was that they were trying to bring together a bunch of smaller AI models and then try to give really complex results. So their idea is they say they're kind of like A swarm, a swarm of small AI models that are going to give really complex results. What's interesting to me with this is this is not, I mean I'm sure they're doing something in a very novel, different way, but it sounds honestly very similar to what, like GPT4. Starting with GPT4 I believe Chat GPT has been doing where GPT4 actually has 16 different expert AI models. And when you ask it a question, it's got kind of like, I think the main model that determines which of those 16 AI models to send it to. It's a really, it's a really good approach. We're building something similar over at AI Box, my AI startup in our chat feature in our, in our playground where when you're asking it a question for something, it runs it through an original model that determines which model to send your query to, whether you're trying to generate images or text or video, whatever you're kind of, whatever you're trying to do. And then that model will send it over to another model that's actually able to process the request. So it seems like within the actual foundational models themselves they have similar things. And so anyways that's kind of like that approach. I'm not sure to what level. They say they have a swarm of AI models so I'm not sure if that's more than 16 or if it's going to be something similar, but this is definitely a unique approach. They also are building this on a native Japanese data set, a bunch of native Japanese data sets, which is interesting, right? Because right now most of our AI models are built off of American or English speaking data sets. I think largely I would be really curious to see kind of how AI models differ if they're built off of it exclusively Chinese data sets or Japanese data sets or French data sets. I think that would be an interesting thing. I like oftentimes these are all built off of English USA because it's just kind of like one of, it's like the biggest data set for online stuff. But Japan has quite a large population that has been chronically online as well. So we could, I think there's some large data sets there and it could, could come up with some interesting perspectives. If they're like culture, traditions or way of thinking is like slightly different. I'd be curious to see how that comes through in the model. Be very, very curious. They have multiple models they've come up with right now they have some that are capable of generating Japan's kind of traditional Yukiko E Artwork, which is kind of interesting. Right. So already we're seeing them able to generate artwork from Japan that's a little bit more traditional because that's kind of in their data set. Most recently they actually shared research on the AI scientist. So this was an LLM system that they created that essentially is essentially automating the entire research lifecycle. There's a bunch of different things that go into that. There's, you know, coming up with the ideas or like ideation, there's writing the actual code, there's actually running the experiments and then kind of summarizing the results. And then I think after that it will even write the entire paper and it will even do, you know, conduct some peer review stuff. So that was an interesting product. They come up with their AI scientist, they've come up with a lot of really interesting things. I think the next step for them is they're really looking to scale up their nature inspired approach to AI development. Right. Like they talk about their swarm and a bunch of different things. They say they're like trying to use nature to help with development. This is like with AI concepts and how they're developing things. I think this is a great idea. I mean I feel like a lot of AI research that's been successful to date has in one way or another been reflective of obviously human intelligence and different psychological things. So I don't know. I think this is very interesting. We're kind of going in the same direction. They seem to be trying to follow more of a nature, animal ecosystem kind of approach to it, which I think is interesting. With this new round of funding that they just raised, you know, Nvidia is putting in money and all these other venture capital firms, they are planning on speeding up their hiring. They want to build a really quote, talent dense AI research organization. I mean that sounds fantastic. That also sounds like something everyone would say. I don't know, that's nothing special in itself. They're also going to get a bunch of resources from their end infrastructure from their partnership with Nvidia. So that kind of makes sense. Nvidia put money in and that's probably going to go straight back to Nvidia Chips. They did publish a blog post about all of this recently and they said that Jensen Huang is going to. And well, I think really they mean Nvidia, but they're, they're kind of like sometimes they reference them. They're going to be offering a lot of this infrastructure support and they said they're going to be doing on two fronts. First they're going to provide all of the latest GPU systems to help them actually train and develop everything. And they're also going to give them access to Nvidia's powered data centers that are actually inside of Japan and they're going to be using those for running all their experiments and stuff. So very, very exciting. They recently said in a quote on kind of this whole, whole thing, they said this is a daunting task, especially for a small startup company which will require years of R and D and building long term relationships with key stakeholders in the nation. We believe our technology will help Japan regain a technological edge in AI and increase its global competitiveness. We also aim to deploy our technology to assist Japan, its institutions and its citizens to overcome its AI challenges in the road ahead. So the company kind of said that. I believe that was in their blog post. All of this is kind of in this trend. I've seen a lot of like very pro AI things coming out of Japan. I think Japan has at different times been, you know, a big leader in technology and feels like maybe it's slipping right now with AI, not a lot of the big AI companies or big AI technologies coming out of Japan per se. So I think Japan's trying to change that and this is, this company is kind of a big step in that direction. So it's going to be really interesting to see, you know, how this is, how this rolls out. OpenAI Also, I believe has a Japan headquarters as well. So I think, you know, there's, there's other players in Japan that are going to be, that are going to be duking it out. OpenAI and Sam Altman, I think he, he kind of like launched the Tokyo hub back in April and they've already released a custom GPT4 model optimized for the Japanese language, which I think is really interesting. And then another funny thing, if you follow all of the, the rumors and drama with OpenAI is that Japan's open. OpenAI's Japan President tattoo, he was recently giving a presentation that sort of got leaked where he was showing off a new AI model called GPT Next, which some people are saying is going to be 100 times better than GPT4 and AKA it's like it's GPT5 but they're calling it GPT Next. So we'll see what happens there. A lot of really interesting things coming out of Japan. I think it's unclear whether like the Japan lab developed it or if they just had information on it and we're talking about it. But I would not count out the Japanese when it comes to technology. They've been a leader in this space. This is something that they're very, very, very good at. So I'd be very excited to see, you know, not just what OpenAI Japan is able to is able to come up with, but with all of this funding, $50 million, what Saskana AI comes up with. This is a very innovative AI lab that I would definitely recommend keeping your eyes on. I'll keep you up to date on everything they do. Now, if you're interested, as I mentioned, in starting a podcast, I will leave a link in the show notes as well to my very own podcast course. Hope you all have an amazing rest of your day and I will talk to you next time.