🔥 AI Startup Raises $1B: Safe Superintelligence, co-founded by ex-OpenAI chief scientist Ilya Sutskever

Read time: under 7 minutes

WhatsUp AI Subscriber!

It’s time for your weekly AI dose! AI is making waves again! 📸 Google Photos introduces the AI-powered ‘Ask Photos’ feature, making it easier to unlock and relive your memories. 🪄 Reflection 70B, an open-source model, is redefining AI excellence with its innovative capabilities. 🎉 OpenAI is set to revolutionize AI with its upcoming models, Strawberry and Orion. 🚗🔍 Tesla’s FSD roadmap raises more questions than answers, sparking discussions on the future of autonomous driving. 🌟 Gemini’s new AI features take image generation to the next level, showcasing impressive advancements.

Here are all the top trends making headlines!

What’s Trending 🚀 

📸 Unlocking Memories: Google Photos’ New AI-Powered ‘Ask Photos’ Feature

🪄Reflection 70B: The Open-Source Model Redefining AI Excellence

 🎉 OpenAI’s Next Big Leap: Strawberry and Orion Set to Revolutionize AI!

 🚗🔍 Tesla’s FSD Roadmap: More Questions Than Answers

🌟 Gemini’s New AI Features: A Leap Ahead in Image Generation!

Google Photos is introducing a new feature called Ask Photos, which is currently in early access for select US users on Android and iOS. This feature uses Gemini models to understand the context of your photo gallery, such as important people, hobbies, or favorite foods, and helps you find specific memories or uncover information about your life.

Key features of Ask Photos include:

• Conversational Search: You can use natural language queries to search your photo gallery, mimicking a chat conversation. If Ask Photos doesn’t find what you’re looking for, you can provide extra details to refine the search.

• Contextual Understanding: The feature can identify details like where you took photos with your camping gear or what dish is on the table in a restaurant photo.

• Photo Suggestions: Ask Photos can suggest the best photos from events like birthday parties for shared albums or summarize top activities from recent trips to share with friends.

On the privacy front, Google assures that Google Photos is never used for ads and is protected with industry-leading security measures. Queries may be reviewed by humans to improve the feature, but only after being disconnected from your Google Account to protect your privacy.

This innovative feature aims to make finding and sharing memories easier and more intuitive, leveraging advanced AI to enhance user experience.

Reflection 70B: The New Open-Source Champion

Matt Shumer, co-founder and CEO of AI writing startup HyperWrite, recently launched Reflection 70B, a groundbreaking open-source language model. This model has outperformed top closed-source models like OpenAI’s GPT-4o and Anthropic’s Claude Sonnet 3.5. Developed using a novel technique called Reflection-Tuning, Reflection 70B can detect and correct its own mistakes before finalizing an answer. This technique addresses common issues like model hallucinations and improves reasoning accuracy. The model outputs its internal reasoning in <thinking> tags and final answers in <output> tags, with additional <reflection> tags used for correcting any detected errors³.

Reflection 70B has achieved top positions in several benchmarks, including MMLU, MATH, IFEval, and GSM8K, demonstrating superior performance over GPT-4o and Llama 3.1 405B. The upcoming Reflection 405B model, expected next week, is anticipated to further elevate the standard for LLMs globally.

This week also saw the release of Qwen2-VL by Alibaba, a vision-language model that can chat via camera, play card games, and control mobile phones and robots. The advanced 72 billion model of Qwen2-VL achieved state-of-the-art visual understanding across 20 benchmarks, often surpassing even closed-source models like GPT-4o and Claude 3.5-Sonnet.

These advancements highlight the rapid progress in AI, with open-source models like Reflection 70B leading the way in innovation and performance.

OpenAI’s ambitious plans for its new AI models and corporate restructuring reflect its commitment to innovation and growth in the AI space. The company’s focus on advanced capabilities and significant funding efforts highlight its leading position in the industry. Here is the quick summary for you

• High Subscription Prices: OpenAI is considering subscription prices as high as $2,000 per month for its upcoming large language models (LLMs) like Strawberry and Orion.

• Advanced Capabilities: The new AI model, Strawberry, will be able to solve complex problems, perform high-level tasks, and conduct deep research beyond current AI capabilities.

• Corporate Structure: OpenAI is contemplating changes to its corporate structure to make it more attractive to investors while maintaining its mission to benefit humanity.

• Funding and Valuation: OpenAI aims to raise several billion dollars in a funding round, potentially valuing the company at over $100 billion. Potential investors include Apple, Nvidia, Microsoft, and Thrive Capital.

Tesla has revealed its AI product/self-driving roadmap for the next few months, but it raises more questions than it answers. The main issue is the lack of transparency in FSD intervention data, making it hard to measure improvements. Tesla’s CEO, Elon Musk, claims that the upcoming v12.5 will achieve a “3x improvement in miles between necessary intervention” compared to previous versions.

However, the data shows only slight improvements. For instance, v12.5.1 averages 32 miles between disengagement and 128 miles between critical disengagement, compared to v12.3.6’s 30 miles and 189 miles, respectively. To achieve an unsupervised self-driving system, Tesla would need to reach 50,000 to 100,000 miles between critical disengagement.

• Crowdsourced Data: Shows minor improvements in miles between disengagements.

• Critical Disengagements: Need to be significantly reduced for safe, unsupervised driving.

• Driver Frustration: Many drivers take control due to slow speeds and navigation errors.

Tesla’s roadmap brings more questions than answers, mainly due to the lack of official intervention data. The company needs to address these issues to make meaningful progress in self-driving technology.

Tesla’s new humanoid robot can perform complex tasks now

Google has announced new features for its AI chatbot, Gemini, at its annual developer conference, Google I/O. These features are now available to users, including the upgraded Imagen 3 image generation model, which promises higher image quality compared to its predecessor, Imagen 2. This upgrade is available to Gemini Advanced, Business, and Enterprise users.

Key highlights include:

• Imagen 3’s Performance: In a recent evaluation, Imagen 3 outperformed other leading models like DALL-E 3, Midjourney v6, and Stable Diffusion 3 Large in terms of user satisfaction with the generated images.

• Free Image Generation: Unlike ChatGPT, which requires a subscription for image generation, Gemini offers this feature for free, making it a significant advantage for users.

• Customizable AI Assistants: Gemini introduces Gems, customizable AI assistants that can be tailored for specific tasks such as coding, writing, and career guidance. This feature is similar to ChatGPT’s custom GPTs and is available in over 30 languages and 150 countries.

Additionally, Google has integrated Synth-ID into Imagen 3, which watermarks AI-generated images to indicate their origin. The company is also rolling out the ability to generate images of people, with early access for Gemini Advanced, Business, and Enterprise users.

These updates position Gemini as a strong competitor in the AI chatbot market, offering advanced features and customization options to enhance user experience.

Funding Frenzy

* Fastn’s $2.6M Boost: Revolutionizing Composable Architecture for Rapid Integration

* You.com secures $50 million to enhance AI-driven productivity tools, focusing on complex queries and enterprise solutions.

*Sedric AI raised $18.5 million in Series A funding to enhance its AI-powered compliance platform for financial institutions¹.

* Safe Superintelligence, co-founded by ex-OpenAI chief scientist Ilya Sutskever, raised $1B to develop general AI systems.

*Anytime AI raises $4M in seed funding to advance its AI-powered legal assistant technology.

*Tokyo-based artificial intelligence firm Sakana AI raised over $100 million in a Series A funding round and formed a collaboration with Nvidia

Got thoughts? Hit us up! We'd love to hear your feedback or any cool tidbits you'd like to share. Just reply to this email