Reka, a California AI startup, released Flash, a 21-billion-parameter model trained on 32+ languages that scores 73.5% on MMLU benchmarks, and Edge, a smaller 7B model for devices. The company raised $58M and plans a larger model called Core. Lenovo announced an AI operating system and AI-enabled computers coming in the first half of 2024.
Questions this video answers
$58 million
21 billion parameters
7 billion parameters
73.5% accuracy
Text from over 32 languages
Chapters
A groundbreaking tech startup from California called Raika is making waves with its latest AI Reika Flash. Founded by X, Google and Facebook experts including Singapore's Yi Tai, this team is fast becoming a big deal in AI. With a recent $58 million boost, they're on a roll, proving they're ones to watch in the tech world. At the heart of Reika's recent announcement are two distinct AI models, Rekka Flash and Raka Edge. Flash is a multimodal and multilingual model boasting 21 billion parameters. Designed to process and understand a blend of text and images across more than 32 languages. Its capabilities are said to rival and in some aspects outperform those of established models like Google's Gemini Pro and OpenAI's GPT 3.5, especially in reasoning, code generation and question answering. Raka Edge, on the other hand, is a more compact variant with 7 billion parameters tailored for scenarios requiring minimal resources. Edge is ideal for use on devices with limited computing power, making advanced AI technologies more accessible and versatile. Now Raca Flash's prowess is highlighted through its performance on several benchmarks. It has demonstrated superior capabilities in knowledge based question answering, reasoning and code generation, surpassing larger models like Llama 2, Grok 1 and even GPT 3.5. For instance, in the MMLU benchmark for knowledge based question answering, Flash achieved an accuracy of 73.5%, outperforming Gemini Pro and GPT 3.5. Similar superior performances were noted in other benchmarks, underscoring Flash's efficiency and effectiveness. A significant advantage of Flash Model is its multilingual capacity. Pre trained on text from over 32 languages, including English, German, Chinese and Japanese, among others. This broad linguistic foundation allows Rekka Flash to excel in multilingual reasoning and question answering, setting new standards for global AI applications. In the realm of multimodal benchmarks, which involve understanding and generating responses based on images and videos, Raka Flash remains competitive. It shows promising results in visual question answering and video captioning, maintaining a strong performance against models like Gemini Pro and outdoing others on various tasks. Raika has also ventured into refining AI's conversational abilities with both Flash and Edge through rigorous instruction, tuning and reinforcement learning techniques. These models have been optimized for engaging and realistic chat experiences in evaluations. Pitting Raika's models against others like GPT4 and Claude 2.1, Flash and Edge have demonstrated commendable performance, showcasing their potential in delivering nuanced and contextually aware interactions. Rekka Edge, despite its smaller size, has not lagged in performance it shines in language benchmarks rivaling and even outperforming other models of its scale and knowledge, question answering, reasoning and code generation. Its efficiency and the ability to work in resource constrained environments make it a valuable tool for developers and users needing lightweight but powerful AI solutions. Raka's ambition doesn't stop with Flash and Edge. The company has teased the upcoming release of Raka Core, its largest and most advanced model yet, promising further innovations in AI technology. The founding team's vision, rooted in their rich experience at leading tech firms like Google, DeepMind and Meta, is not just about creating competitive AI models. It's about pushing the boundaries of what AI can achieve, making sophisticated technologies accessible and usable across a wide range of applications, from on device utilities to complex server based processing tasks. Now, the launch of Flash and Edge AI models represents a significant milestone in the AI landscape. By offering models that combine efficiency, flexibility and advanced capabilities, Reka is not just challenging established giants like Google and Microsoft, but is also democratizing access to cutting edge AI technologies. With its continued focus on innovation and accessibility, the company is poised to make a lasting impact on how AI is developed and utilized, bringing us closer to a future where AI can more profoundly elevate human potential. So what's the big deal? Well, at CES 2024, a big tech event where companies show off their latest gadgets, Liujun, who's a big shot at Lenovo, shared some interesting news. Lenovo is cooking up its very own AI based operating system. Why? Because they want to dive deep into the world of AI, just like many other tech companies These days, it seems like AI is popping up everywhere in technology and Lenovo doesn't want to be left behind. But Lenovo is doing more than just creating a new operating system. They're planning to release a bunch of smart computers in the first half of 2024. These computers are special because they can learn how you like to use them. This means the more you use them, the easier and more comfortable they become to use as they get to know your habits and preferences. Lenovo's big idea here is to create a whole ecosystem of AI powered devices. So if you end up using their AI tech, you'll likely find that everything works together really well, making your digital life a bit easier and more connected. It's a bold move by Lenovo, aiming to carve out its own space in the AI tech scene with something that could really shake things up. Alright, that wraps up our video. If you liked it, please consider subscribing and sharing so we can keep bringing more content like this. Thanks for watching and see you in the next one.
