Hello Tuners,

First, Sakana AI, the Japanese startup with a blend of ex-Google wizards and diplomatic finesse, has netted $100 million in funding. With powerhouse backers like Nvidia and Khosla Ventures, they're not just making waves. They’re diving into the deep end with AI innovations that might just make even the most sophisticated algorithms look like fish in a barrel.

On the other side of the AI galaxy, Ilya Sutskever’s Safe Superintelligence Inc. is scaling the heights of superintelligent AI with a cool $1 billion in funding and a mission to build a brainy digital overlord that’s more “peacekeeper” than “planet conqueror.” Lastly, DeepMind’s GenRM is shaking up LLM accuracy with generative reward models, proving that sometimes, to solve complex problems, you need a little extra “generative” magic.

• Sakana AI Gears up to Enter Big Leagues

• Patching With Houston, Deploying Tune Assistant! 🧑‍🚀

• Safe(er) Superintelligence Making OpenAI Sweat

• DeepMind’s GenRM, Are We Repeating RLHF?

• You Get an AI Assistant As Soon As You Make(s) It

Sakana AI Gears up to Enter Big Leagues

Move over, ChatGPT! Just as Alibaba’s Qwen2-VL was making headlines, Sakana AI, a Japanese startup with a dream team of ex-Google researchers and a former diplomat, has reeled in a whopping $100 million in funding. With Nvidia and Lux Capital in their corner, they’re not just playing in the shallow end but diving headfirst into the deep sea of AI. Sakana’s secret sauce? Think less Skynet and more schools of fish working together to create brighter, faster AI models.

But this isn’t just about cool AI-powered ukiyo-e art (though that’s a flex). Sakana’s aiming big, with plans to make Japan the AI powerhouse it deserves to be. They've got Nvidia backing them with GPUs galore, and soon, they’ll be throwing AI hackathons to scoop up the best talent in Japan. Rivals like OpenAI and Cohere might be circling, but Sakana’s swimming upstream with some severe tech muscle. Vinod Khosla even said they’re leading the “path to innovation”, so no pressure there!

Patching With Houston, Deploying Tune Assistant! 🧑‍🚀

Tune Assistant is now live in production, showcasing a range of advanced features designed to boost its functionality and integration capabilities. Here’s a rundown of the new features:

• Define Custom Instructions: Tailor the assistant’s tasks to your specific needs.

• Custom Function Model: Select a model to determine the appropriate function to call.

• Custom Response Model: Merge outputs from various tools into a unified response.

• Web Search: Instantly fetch answers from the web.

• Image Generation: Create stunning visuals on the fly.

• Actions (OpenAPI spec): Connect and interact with external applications.

• Function-Calling (10x smarter): Access real-time data and execute custom functions.

Head to Tune Studio Now to get the most out of your Models and Credits! 🚀

Safe(er) Superintelligence Making OpenAI Sweat

In the ultimate plot twist, Ilya Sutskever, OpenAI’s co-founder and part-time coup enthusiast, has emerged from the AI wilderness with a shiny new $1 billion investment in hand. His latest venture, Safe Superintelligence Inc. (SSI), has some of Silicon Valley’s heaviest hitters, including Andreessen Horowitz and Sequoia, backing it, giving Sutskever more clout than a startup tech bro on Twitter. With a $5 billion valuation and a team of just ten people (talk about scaling!), Sutskever's mountaintop tweet ("Mountain: identified. Time to climb.") sounds like he's less interested in overthrowing CEOs and more focused on building AI that's smarter than all of us. What could go wrong, right?

Unlike your run-of-the-mill AI startup trying to get your microwave to flirt with you, SSI’s significant goal is to build safe superintelligence, a digital brain that’s smarter than humans but, you know, not planning world domination. With no pesky product deadlines or commercialization distractions, Sutskever’s team will focus purely on science (and maybe some peace of mind for us regular humans). They've got enough computing power and VC backing to make OpenAI nervous, so buckle up for the AI wars ahead.

DeepMind’s GenRM, Are We Repeating RLHF?

Google DeepMind and collaborators from top institutions have introduced GenRM, a new approach to improve the accuracy of large language models (LLMs) through generative reward models. Unlike traditional verifiers that use pre-trained models to score responses, GenRM taps into LLMs' text-generation capabilities. By training verifiers with next-token prediction and using techniques like chain-of-thought (CoT) prompting, GenRM can produce intermediate reasoning steps, catching subtle errors that classic models often miss.

Tested across various reasoning tasks, GenRM consistently outperformed existing methods, including GPT-4 and Gemini 1.5 Pro, particularly on math problems. The model's ability to scale and enhance verification with advanced prompting techniques positions it as a significant breakthrough for improving LLM performance in complex reasoning applications.

You Get an AI Assistant As Soon As You Make(s) It

You.com just scored $50 million in Series B funding, setting the stage for a significant shift in enterprise AI. This round, led by Georgian and backed by Nvidia and Salesforce Ventures, brings their total funding to $99 million. You.com, founded by Richard Socher, You.com aims to tackle "AI sprawl" with its all-in-one platform, offering AI-powered productivity tools tailored to knowledge workers.

With features like the "Research Assistant" and "Genius Assistant,"  You.com promises accurate, real-time answers backed by citations, unlike many AI tools prioritizing speed over precision. Their enterprise-focused strategy, including multiplayer AI for team collaboration, is geared to enhance productivity and drive AI adoption across organizations.

LLM Of The Week

Qwen2-VL

Qwen2-VL is the latest evolution of the Qwen-VL model, marking a significant leap in AI capabilities. This new version boasts state-of-the-art visual understanding performance across benchmarks like MathVista and RealWorldQA. It excels in interpreting images of different resolutions and ratios and can process videos over 20 minutes long, making it ideal for advanced video-based tasks. Qwen2-VL integrates seamlessly with mobile phones and robots, providing automated operation based on visual and textual inputs.

The model's architecture includes critical updates like Naive Dynamic Resolution, allowing it to handle arbitrary image images with a more human-like processing approach. Its Multimodal Rotary Position Embedding (M-ROPE) enhances its ability to manage textual, visual, and video information effectively. With support for multiple languages beyond English and Chinese, including major European and Asian languages, Qwen2-VL is designed to serve a global audience, advancing the frontiers of multimodal AI.

Weekly Research Spotlight 🔍

Agentic Retrieval-Augmented Generation for Time Series Analysis

Time series modelling often encounters challenges such as complex spatiotemporal dependencies and distribution shifts. A new approach has been proposed to address these issues using an agentic Retrieval-Augmented Generation (RAG) framework. This method employs a hierarchical, multi-agent system where a master agent directs tasks to specialized sub-agents. These sub-agents, equipped with fine-tuned pre-trained language models and a shared repository of prompts, are designed to handle specific time series tasks by retrieving relevant historical knowledge to improve predictions.

This modular RAG framework has demonstrated exceptional flexibility and performance on benchmark datasets. By effectively managing complex dependencies and adapting to shifts in data distribution, it surpasses traditional, task-specific methods, offering a more robust solution for time series analysis.

Best Prompt of the Week 🎨

Today's Goal: Try new things 🧪

Acting as a Business Development Partner

Prompt: I want you to act as a business development planner. You will create a structured daily plan tailored to help an individual start a corporate merchandise business focused on serving B2B clients. You will identify a target market, develop essential strategies and action steps, select the tools and resources for efficient business operations, and outline any additional activities needed to ensure the successful launch and growth of the business. My first suggestion request is: "I need help creating a daily activity plan for someone who is planning to start a corporate merchandise business targeting B2B customers."

This Week’s Must-Watch Gem 💎

This Week's Must Read Gem 💎