Google Launches Genimi: The ChatGPT Killer

Today’s top AI Highlights:

1. Google released the most awaited LLM Gemini to compete with GPT-4

2. Google Cloud enables next-generation of AI workloads with AI Hypercomputer

3. Meta AI releases Imagine to rival OpenAI’s DALL.E and Midjourney

4. Generate the perfect font combination for your next project using AI

& so much more!

Read time: 3 mins

Latest Developments 🌍

Google Unveils Gemini: Multimodal AI Set to Rival GPT-4 🚀

Google has finally released Gemini, a multimodal AI model, meaning it can understand, operate across, and combine different types of information, including text, images, audio, video, and code. This AI model is not just an answer to OpenAI's GPT-4 but a leap forward in multimodal AI technology.

Key Highlights:

• Varied Variants: Gemini is available in three sizes: Nano for Android, Pro integrated into Google Bard, and the upcoming Ultra variant, most capable and largest model for highly-complex tasks.

• Unified Multimodal Approach: Unlike OpenAI's model-specific approach (DALL.E for images, Whisper for audio), Google has developed Gemini as a single, integrated model adept in handling text, images, audio, and video.

• Benchmark Brilliance: In a series of tests, Gemini outperformed GPT-4 in 30 out of 32 benchmarks, particularly excelling in video and audio processing due to its holistic multimodal design.

• Real-World Impact: Beyond benchmarks, Gemini shines in practical applications. Its AlphaCode 2 system, designed for coding, has shown remarkable performance, surpassing many in coding competitions.

Google has also released AlphaCode 2, Google’s next-gen code-generating AI model, utilizing Gemini Pro with sophisticated search and reranking strategies. It is a significant enhancement of the original AlphaCode, and excels in competitive programming, outperforming 85% of human competitors. It also showcases an improved ability to solve programming problems, achieving a solution rate of 43% in evaluated contests.

Google Enables Next-Gen AI Workloads 🏋️‍♀️

Google Cloud has launched the Cloud TPU v5p and AI Hypercomputer, marking a significant leap in AI computing capabilities. These are designed to accelerate the training and serving of complex AI models, reshaping the landscape of AI-driven applications.

Key Highlights:

• Cloud TPU v5p: It offers unparalleled power, scalability, and adaptability. It's capable of performing more than double the floating-point operations per second (FLOPS) and features triple the high-bandwidth memory compared to its predecessor, TPU v4.

• Efficiency in Training LLMs: With the Cloud TPU v5p, Large Language Models (LLMs) can be trained 2.8 times faster, and embedding-dense models can be processed 1.9 times quicker than before.

• AI Hypercomputer: A novel supercomputer architecture, specifically engineered for AI workloads. It integrates performance-optimized hardware with open software, supporting ML frameworks like JAX, TensorFlow, and PyTorch.

Meta Releases DALL.E Competitor 🏞️

To close out the year, Meta is testing more than 20 new ways generative AI can improve our experiences across Facebook, Instagram, Messenger, and WhatsApp — spanning search, social discovery, ads, business messaging, and more.

Our favorite today is Imagine, Meta’s new text-to-image generator powered by Emu. It is quite similar to DALL.E, Midjourney, and Stable Diffusion, and generates four high-quality images for each prompt.

Tools of the Trade ⚒️

• Hexomatic: A no-code platform that offers over 100 automations for sales, marketing, and research tasks, streamlining operations without the need for programming skills.

• Embedefy: Embedefy offers free open-source embeddings, allowing seamless generation of embeddings for various AI applications like semantic search, clustering, and recommendations.

• Flightcontrol: Flightcontrol simplifies the deployment and management of AWS services, offering an intuitive platform that integrates seamlessly with tools like GitHub, supports a variety of AWS services, and ensures efficient resource management.

• Fontjoy: Fontjoy uses AI to generate harmonious font pairings, simplifying design decisions and enhancing the visual appeal of your work.

Hot Takes 🔥

1. ok but i can’t lie Gemini Ultra is a terrible name ~ roon

2. Hot-take on Gemini 1) Surprised that Google managed to produce something roughly on GPT-4 level. They might be the only company other than OpenAI who will though. 2) The technology is SATURATING hard, scaling laws != competence 3) The elephant in the room is the AUTONOMY GAP - these models are still as useless as ever unless there is a human in the loop almost at every single step, reinforcing that they are TOOLS, not AGENTS. 4) Many of their demos look dated compared to what you can do on ChatGPT already 5) Google pulled this off mostly because of access to DATA, the DATA IS THE MOAT! 6) They got GPT-4 level performance with fewer training tricks (dense model) PS. Don’t trust benchmarks ~ Machine Learning Street Talk

Meme of the Day 🤡

That’s all for today!

See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!