Google Building its AI Search Engine

Plus: Stability AI's Language Models, Microsoft's own AI chips, OpenAI's read team, and more!

Hey there 👋

Welcome to yet another interesting week in the world of AI! This time, our favorite entrepreneur, Elon Musk, has caught our attention with his announcement of TruthGPT - a fresh competitor to OpenAI's well-known ChatGPT, all under his brand new venture, X AI Corp. As if things weren't lively enough already!

Hold on, there's more! Google has finally entered the AI-powered Search arena, proving that it's never too late to join the party. Amidst the clamor from tech leaders for a temporary halt on training large language models, OpenAI's Sam Altman reassures us that GPT-5 isn't in the works just yet.

And if you thought that was all, we also have over 100 new AI research papers and tools to keep your week engaging, as we've come to expect in the AI domain. So sit back, relax, and join us as we explore the fascinating world of artificial intelligence!

This issue covers:

1. Latest Developments 🌍

2. News from the Industry 🧑‍🏫

3. Tools of the Trade ⚒️

4. Hot Takes 🔥

5. AI Meme of the Week 🤡

Latest Developments 🌍

Our Pick 👌

DINOv2 by Meta: SOTA computer vision models using self-supervised learning for high-performance features without fine-tuning.

• StableLM by Stability AI: An open-source language model suite, including models with 3 billion and 7 billion parameters, to improve transparency, accessibility, and support for users.

• Paella by LAION: Text-to-image model that uses a quantized latent space and a CNN architecture, aims to make text-to-image models more accessible.

• Multimodal C4: A billion-scale corpus of images interleaved with text for few-shot learning and complex prompts.

• SAM meets Inpainting: Inpaint Anything, a mask-free image inpainting system based on SAM, offering "click and fill" features: Remove, Fill, and Replace Anything.

• SEEM (Segment Everything Everywhere All at Once): A versatile, interactive, and semantic-aware model for visual segmentation with prompt-based capabilities.

• SAM-Adapter: Improves performance of SAM in challenging tasks such as camouflaged object detection and shadow detection.

• FaceLit by Apple: A generative framework for generating 3D faces from 2D images with photorealistic results and illumination and view controls.

• RedPajama: A project that aims to create leading open-source LMs, starting with the reproduction of the LLaMA training dataset of over 1.2 trillion tokens.

• MiniGPT-4: Advanced vision-language model with enhanced capabilities and computational efficiency.

• LLaVA: Large Language and Vision Assistant - a SOTA multimodal model for visual instruction tuning and chat capabilities.

• CAMEL (Communicative Agents for "Mind" Exploration of Large Scale Language Model Society): A framework for studying autonomous cooperation among communicative agents using role-playing.

• Low-code LLM: A visual programming approach for more effective utilization of LLMs in complex tasks through user-friendly interactions.

• Robust Prompts on Vision-Language Models: By integrating multiple-scale image features into prompts, improving robustness on base and novel classes.

• Chinese Open Instruction Generalist: A Chinese instruction dataset for generalist LMs, has around 200k high-quality Chinese instruction tuning samples.

• Solving Math Word Problems: Combining LMs with symbolic solvers improves accuracy in solving complex math word problems.

• Personalized Avatar Scene: Pipeline for zero-shot avatar personalization using text-to-3D pose diffusion model trained on large-scale image datasets.

• Soundini: Adds sound-guided visual effects to videos using denoising diffusion probabilistic models, producing realistic effects that reflect sound properties.

• Generative Disco: Text-to-video generation that uses LLMs and text-to-image models to create music visualizations.

• Zip-NeRF: Combines grid-based representations with anti-aliasing techniques, for faster training and improved accuracy.

• Nerfbusters: Removing artifacts such as floaters or flawed geometry and improve scene geometry in casually captured NeRF.

• Avatars Grow Legs (AGRoL): Generates smooth human motion for full-body avatars from sparse tracking inputs, focusing on the lower body movement.

---

News from the Industry 🧑‍🏫

Our Pick 👌

Adobe expands Firefly to include video and audio with features such as text to colour, custom sound generation, and script analysis.

• Google is developing an AI-powered search engine and upgrading its existing one, named Magi, in response to competition from AI-powered rivals like Bing, after losing out on contracts with Samsung.

• Google has merged its AI research units, Brain and DeepMind, into a new unit Google DeepMind, to accelerate progress in AI and reshape the technology landscape.

• Elon Musk is reportedly working on a "maximum truth-seeking AI" called TruthGPT, which he believes would be safer for humanity.

• Microsoft reportedly accelerates development of its own AI chips, named Athena, to reduce reliance on Nvidia.

• OpenAI hired a “red team” of experts to test its GPT-4 for issues like toxicity, prejudice, and linguistic biases, and mitigate them before launching.

• OpenAI's CEO, Sam Altman, confirms that the company is not currently training GPT-5 and won't be doing so for some time.

• The European Data Protection Board has established a task force to establish privacy policies for ChatGPT, following Italy's lead.

• LangChain secures funding led by Sequoia at a valuation of at least $200 million in an exclusive funding round.

• Web LLM, a project from the team behind Web Stable Diffusion, now runs the vicuna-7b LLM in a browser using the WebGPU API, with impressive performance.

• Hugging Face is partnering with AWS to optimize Hugging Face Transformers for AWS Inferentia2, a new inference accelerator that offers increased throughput and reduced latency.

• Reddit plans to charge companies for access to its API containing its conversations, used to train models developed by companies like Google and OpenAI.

• Stack Overflow also plans to charge large AI developers for access to its data to train models like ChatGPT, Dall-E, Bing, following Reddit's footsteps.

---

Tools of the Trade ⚒️

Our Pick 👌

WOXO VidGPT: ChatGPT plugin to create engaging videos in minutes with text prompts.

• Vercel AI Playground: Compare and fine-tune top models like GPT-4, Claude, and receive dynamic OG cards and auto-code snippets.

• Bardeen Chrome Extension: Automate tasks, connect apps, and boost productivity with one-click automation.

• Forge: Create and monetize AI-powered applications without writing any code.

• Snack Prompt: Simplifies and optimizes your ChatGPT experience with curated prompts, community upvoted top prompts and other features.

• VirtuozyAI: Your go-to bot for music inspiration and production, powered by GPT-4.

• DYVO: Creates catchy product photos in seconds by removing backgrounds and applying different styles.

• Mnemonic AI: Provides fully automated and data-driven customer intelligence to help businesses achieve higher conversions.

• Relayed: AI-powered meetings for teams, with flexible video conferencing, async conversations, and easy sharing.

• Virtuo: Access ChatGPT from any website for seamless AI-powered assistance and workflow simplification.

• IntelliMail: Chrome extension for personal email assistant by automatically generating emails.

• Uizard Autodesigner: Generates multi-screen mockups for apps and websites with simple text prompts.

• Nekton: Automates workflows with AI and offers a guaranteed 10x return on investment.

• Bitesized: AI-generated news summaries for quick updates without filler or bias.

• SpeechFlow: Speech-to-text API that accurately transcribes audio in 13 languages with accuracy.

• Kaya: Personal AI that learns from your notes and content to provide insights, answers and share knowledge.

• Butternut AI: Creates stunning and fully customizable websites in 20 seconds, without requiring any coding.

• Thunderclap: Supercharge your Twitter game with viral tweet generation and smart reply generation.

• Traverse AI: Legal assistant for well-informed discussions with attorneys and empowering clients with knowledge.

• Eve AI: AI image generator with various models and features, including watermark removal and upscaling.

• Tiipe: Generates creative product descriptions for eCommerce websites 10x faster.

---

Hot Takes 🔥

Source: DataChazGPT

Source: AlphaSignalAI

Source: Shubham Saboo

Source: Bojan Tunguz

---

AI Meme of the Week 🤡

Source: Lorenzo Green

That’s all for this week!

Will see you next Saturday with more such content. Don’t forget to subscribe and give your feedback below.

BONUS 🎉

Share this newsletter with three other friends and stand a chance to win my book GPT-3: The Ultimate Guide to build NLP Products with OpenAI API. Winners will be selected on a monthly basis.

🎁 Every paid subscriber will also receive FREE learning resources on trending topics like Python, Data Science, Machine Learning, and NLP!