In this episode of the Tech Talks Daily Podcast, we venture into the cutting-edge world of artificial intelligence with Yonatan Geifman, CEO and co-founder of Deci.
Deci, renowned for its pioneering work in deep learning and artificial intelligence, is at the forefront of creating AI that builds better AI. Today, we focus on the transformative technology Deci is developing and its groundbreaking partnership with Qualcomm Technologies. This collaboration sets new benchmarks in AI efficiency and makes generative AI more accessible and cost-effective across various industries. But first, let's extend our gratitude to [Sponsor Name] for their support in bringing this episode to life.
Yonatan shares with us the essence of Deci's innovative journey, particularly its strategic partnership with Qualcomm. This collaboration is a game-changer, challenging the status quo and offering high-performance, energy, and cost-efficient solutions in the AI ecosystem. Through their joint efforts, Deci and Qualcomm are introducing two revolutionary Generative AI models tailored for the Qualcomm® Cloud AI 100. These models, DeciCoder-6B and DeciDiffusion 2.0, boost the cost efficiency and scalability of GenAI-based services and address the pressing GPU shortage in the market.
DeciCoder-6B, a state-of-the-art code generation model, and DeciDiffusion 2.0, a text-to-image diffusion model, exemplify Deci's commitment to enhancing AI accessibility. Developed using Deci's proprietary Neural Architecture Search Technology, AutoNAC™, these models are optimized to leverage the full potential of Qualcomm's hardware, setting new standards in performance and efficiency.
As we delve deeper, Yonatan enlightens us on the significant impact of these innovations across various sectors, from healthcare to finance, and the pivotal role of open-source in advancing AI technology. The democratization of Generative AI, facilitated by Deci and Qualcomm, promises to unlock new levels of creativity and efficiency, making the transformative power of AI available to all.
Join us as we explore how Deci's collaboration with Qualcomm is reshaping the landscape of artificial intelligence, making it more accessible, efficient, and transformative for industries worldwide. What does the future hold for AI development, and how will these advancements impact our approach to technology and innovation?
Share your thoughts and join the conversation.
[00:00:00] Do you find yourself wondering about this incredible journey of AI development and its
[00:00:08] transformative impact across every sector?
[00:00:11] Well, today I want to dive deep into the heart of innovation with the CEO of a company called
[00:00:18] DESI.
[00:00:19] They are a trailblazing company at the forefront of Tech Talks daily, and in today's digital age where data breaches are all too common, securing sensitive information has never been more critical, right? Well enter KiteWorks, a pinnacle of Managed File Transfer Security or MFT security. So with its FedRAMP moderate authorisation, a prestigious certification that they've
[00:01:42] held since 2017 by the Department of Defence, that people are building and integrating into their applications. We started from the computer vision world and now we are working strongly in the generative AI space, both in the computer vision side, in text to image models and LLMs and enabling
[00:03:01] companies to integrate those new set of models into their applications to drive more innovation be ready or to be accurate. And the inference is the stage where you put models in production and let them work for your users or for your customers. And that's those are slightly different workloads. And Qualcomm, a new chip that is called the AI 100
[00:04:20] that is now integrated into cloud providers like AWS and
[00:04:23] we collaborated in order to build models that we released are available in open source in our Hanging Face repository. If you look there, you can find those models with instructions how to run them in general and how to run them on time exponentially. So if you look on the models like 10 years ago they were significantly smaller and required much less computers compared to the models that we are working with today. So the models are increasing exponentially we need more
[00:07:03] and more compute to run those models and while those models in an automated way in order to make them more efficient and solve that problem of the computational complexity. Today, if you're using one of the largest models like GPT-4 or GLUT2, it will probably take you a couple of seconds to get a response. If you're building a real-time application, that's what probably won't be a reasonable
[00:08:24] time for your users to wait improve the efficiencies in the in the enterprises workflow, for example, developers tools in internal developers, key and internal content creation tools
[00:09:40] and things like that tools that are a large organizational
[00:09:43] building in house in order to improve the efficiency across to be a completely customer facing, customer chatbot. So this is an example of how we take probably the most simple or evident use case of LLMs into both the two sides of internal facing applications and external facing application. And obviously, in an adoption of new technology,
[00:11:03] there are risks.
[00:11:04] And it's easier to take risks on internal use cases
[00:11:08] rather than on of use cases for generative AI for example, summarizing customer calls and giving some insights about those requests of the customers and things like that and many, many, many more use cases that can be optimized for specific how to like the Qualcomm in the example that we gave earlier. And also your partnership with Qualcomm that we mentioned right at the beginning of the podcast, that aims to make generative AI ultimately accessible for a broad range of applications.
[00:13:40] So how do you envision that democratization impacting the development of AI-powered applications and services?
[00:13:46] Because such a huge talking point right now. of the closed source models like GPT-4 and RDR. So if we connect those two trends together, we'll see more and more companies that doing the POC level maybe on one of the closed source model providers and then shifting in order to get more customization, more control and more access to the model by shifting towards open source where in parallel,
[00:15:01] this year we'll see a significant improvement in model.
[00:15:04] So the future is open source,
[00:15:06] we're supporting that future by releasing, decicoder 6B, what we see that it outperformed famous models like CodeLama 7B by Facebook and other models, even StarCoder 15B speed and significant improvement in accuracy that puts that model as one of the best in the category to use on on the Qualcomm AI 100 and not only on that chip for code generation.
[00:18:46] the LLM-7B. So we are strong supporters of Open Source. We believe that the future in the AI will come from Open Source. One of the nice things, the series that basically
[00:18:52] Google released the Transformer architecture as an academic paper, they could keep that
[00:18:58] transformers the building block of all the large language models and generative AI in
[00:19:04] general, one of the building blocks in the organization and like double the size of the organization. One of the things that I was really focused about is how can we keep moving fast while going to the
[00:20:22] organization. Going to the organization takes space where you can find all the models that we released with the models.
[00:21:40] You get all the information about the models, etc. increase adoption. But thank you for sharing that with me today. Thank you very much. So as we wrap up today's conversation, I think it's clear that the journey of AI is one of continuous evolution and boundless possibilities. And Desi, in collaboration with Qualcomm Technologies, is not just navigating this journey, they're helping pave new pathways
[00:23:01] for others to follow. And I think their commitment to enhancing AI efficiency, coupled with the