Today:

Hey Folks! 

Today, we're diving deep into a game-changing technology that's been on the radar of some of the biggest names in tech, including Jeff Bezos and Nvidia. Meet Perplexity AI, the search engine that's quietly earning its spot as a go-to resource for many, outshining giants like Google and even ChatGPT. 

Perplexity is a search engine enhanced by AI, allowing for focused, efficient searches across various domains - from academic papers to YouTube and Reddit. It stands out by providing real-time, source-supported answers, minimizing the risk of misinformation. With features like the ability to attach images, PDFs, and customize searches (even specifying the type of AI model used), Perplexity offers a tailored search experience.

WATCH THE VIDEO ON YOUTUBE

Bard generates photos now, finally

Google's Bard chatbot is catching up to its rival, ChatGPT Plus, by introducing AI image generation using Google's Imagen 2 text-to-image model. Bard, now powered by Google's Gemini Pro large language model, was expected to have image generation, but the more powerful Gemini Ultra model is still in development. 

Bard aims to compete with ChatGPT Plus, which relies on a paid subscription for image generation. Google emphasizes responsibility, watermarking AI-generated images, and implementing safeguards against explicit content. 

Additionally, Google released ImageFX, an experimental photo tool powered by Imagen 2. Bard also expands to over 40 languages and offers a double-check feature in 40 languages.

THE VERGE

Amazon announces AI shopping assistant called Rufus

Amazon introduced a new AI shopping buddy called Rufus. This digital helper makes it easier for folks to find stuff they want to buy. You just type or talk to it in the Amazon app, and it gives you answers. You can ask all sorts of questions like, "What's the difference between trail and road running shoes?" or "Tell me about drip and pour-over coffee makers." 

Rufus uses Amazon's product info, reviews, and web data to come up with the goods. They're testing it with a small group in the U.S. before releasing it everywhere. CEO Andy Jassy's all about AI, and Amazon's got more AI tricks up its sleeve.

CNBC

Tim Cook confirms Apple’s generative AI features are coming ‘later this year’

Apple CEO Tim Cook has confirmed that the company is working on generative AI software features set to release "later this year." This aligns with reports of iOS 18 being a significant update. Cook mentioned their dedication to artificial intelligence without giving specific details. 

Apple's cautious approach to discussing their work is evident, but this hints at ambitious plans to integrate AI into iOS, iPadOS, and macOS. The move follows competitors like Google and Samsung, emphasizing AI features in their smartphones. Cook believes there's a substantial opportunity for Apple in the realm of generative AI. The full details will be disclosed later this year.

THE VERGE

Meta to deploy in-house custom chips this year to power AI drive - memo

Meta Platforms, the parent company of Facebook, is set to deploy a new custom chip, known as "Artemis," in its data centers this year to bolster its artificial intelligence (AI) initiatives. This chip, a second-generation in-house silicon, aims to reduce Meta's reliance on Nvidia chips and the associated costs of running AI workloads. With this move, Meta hopes to support its generative AI products in Facebook, Instagram, WhatsApp, and smart devices like Ray-Ban smartglasses. 

The deployment of its own chip could save the company hundreds of millions in annual energy costs and billions in chip purchasing costs. This development marks a positive shift in Meta's AI chip project after a previous iteration was abandoned in 2022 in favor of purchasing Nvidia GPUs. 

The "Artemis" chip focuses on inference, where models make ranking judgments and generate responses, with future plans for a chip capable of both training and inference. This step could lead to substantial cost savings for Meta's recommendation models.

REUTERS

OpenAI reassures: GPT-4 gives 'a mild uplift' to creators of biochemical weapons

OpenAI conducted a study on GPT-4's potential role in assisting with bioweapon creation. Concerns were raised that AI models like GPT-4 might provide instructions for creating pathogens, potentially enabling malicious use. OpenAI tested 100 participants, half with biology PhDs and half with biology coursework, allowing some to use GPT-4 without safety restrictions. 

The results showed only a "mild uplift" in accuracy and completeness, indicating GPT-4 doesn't significantly aid in creating bioweapons. Obtaining the necessary materials and executing an attack remain complex challenges. While the threat increased slightly, it's not conclusive, necessitating further research and discussion. The study's peer review status is unclear.

THE REGISTER

🧠RESEARCH

EE-Tuning: An Economical yet Scalable Solution for Tuning Early-Exit Large Language Models

EE-Tuning by Alibaba is a cost-effective method for enhancing large language models with early-exit capabilities. Unlike traditional full-model training, EE-Tuning adds special layers to existing models, making them smarter and quicker without the usual heavy costs. This technique not only saves on resources but also works well with current tech, making high-performance AI more accessible. 

Efficient Exploration for LLMs

Efficient Exploration for LLMs by Google explores the significant impact of efficient exploration techniques in improving large language models (LLMs) through human feedback. By adopting a method where an agent creates queries and refines a reward model based on received feedback, the study finds that the best results come from using double Thompson sampling guided by a neural network for uncertainty. This approach dramatically reduces the number of queries needed for high performance, emphasizing the crucial role of uncertainty estimation and exploration strategy in the process. 

ReplaceAnything3D:Text-Guided 3D Scene Editing with Compositional Neural Radiance Fields

ReplaceAnything3D (RAM3D) is a new way to edit 3D scenes by swapping out objects using text instructions. It cleverly removes an object from a scene and puts in a new one, making sure everything looks right from different angles. This tool can change up various real-world scenes, ensuring new objects fit seamlessly without messing up the scene's vibe.

Anything in Any Scene: Photorealistic Video Object Insertion

"Anything in Any Scene," showcases a new method for adding any object to any video scene, ensuring it looks real from every angle. It focuses on making sure the object fits the scene's geometry, matches the lighting, and blends in perfectly, thanks to a special network that adjusts the final video to look as natural as possible. This approach could change the game for creating high-quality videos for VR, editing, and other applications, making it easier and cheaper to get great results. 

CARFF: Conditional Auto-encoded Radiance Field for 3D Scene Forecasting

CARFF is a cutting-edge method for predicting future 3D scenes from past observations, like 2D images. It uses a clever combo of techniques to guess how scenes change over time, creating detailed 3D models. This tech could revolutionize planning and simulation in complex scenarios, like self-driving car navigation, by anticipating visual occlusions and other changes, enhancing safety and efficiency.

Make-A-Shape: a Ten-Million-scale 3D Shape Model

"Make-A-Shape: a Ten-Million-scale 3D Shape Model" introduces a novel 3D generative model designed for efficient training on a vast scale, capable of utilizing 10 million publicly-available shapes. It innovates a wavelet-tree representation to compactly encode shapes, making them generatable by a diffusion model. The model demonstrates superior performance in various applications, such as unconditional generation, shape completion, and conditional generation across a wide range of modalities. It not only delivers high-quality results but also generates shapes efficiently in just a few seconds, often achieving this in just 2 seconds for most conditions.

🛠️TOP TOOLS

Graphy - a tool designed to quickly visualize and share data. It offers an intuitive interface for creating dashboards, reports, and charts, catering to users who need to analyze and present data effectively. 

Dopt - AI-powered platform for in-app assistance, enhancing user success with personalized and contextual experiences. 

TechInterviewer AI - AI-driven platform for conducting technical interviews, offering realistic simulations to evaluate candidates' skills effectively and efficiently.

VectorShift -  AI platform offering end-to-end automations for businesses, streamlining processes like data analysis, decision making, and workflow management with AI technology.

PhotoRoom - AI photo editor that simplifies photo editing. It offers background removal, object removal, instant background creation, and batch editing. 

📲SOCIAL MEDIA

🗞️MORE NEWS

Google Maps is getting ‘supercharged’ with generative AI

Google is enhancing Google Maps with generative AI, aimed at helping users discover new places. This feature uses large language models (LLM) to provide recommendations for restaurants and shopping based on detailed information from over 250 million places and insights from 300 million contributors. Initially available in the US, it aligns with Google's goal to transform Maps into a tool for discovering places, not just for navigation. The AI's conversational style promises personalized suggestions, starting with Google's Local Guides community and expanding to all users later. THE VERGE

The Arc browser is getting better bookmarks and search results, all thanks to AI

The Browser Company is bringing AI to its Arc browser for Mac and Windows, introducing features like "Instant Links" that skip search engines and directly provide links based on user queries. Another feature, "Live Folders," offers live-updating data streams. Additionally, Arc Explore, a tool for automating web browsing using AI-generated summaries and links, is in development. The Browser Company aims to simplify internet usage with these AI-driven changes, part of its "Second Act" focusing on AI, in line with the broader trend of browsers integrating AI capabilities. This shift is transforming how users navigate the web. THE VERGE

Meta revenue soars as it pivots to AI and announces dividends for investors

Meta, formerly Facebook, reported strong fourth-quarter earnings, with revenue reaching $40.1 billion, up 25% year-over-year. This growth comes as Meta focuses on integrating artificial intelligence (AI) into its products, with CEO Mark Zuckerberg stating that AI will be the company's biggest investment area in 2024. AI will enhance advertising campaigns, support new products like AI chatbots, and drive advertising revenue. Despite regulatory concerns and competition from platforms like TikTok, Meta's revenue from advertising, its core business, reached $38.7 billion. The company also announced a $50 billion share buyback program and plans to pay a dividend to investors. THE GUARDIAN

AI Data Center Startup Astera Labs Plans March IPO

Astera Labs, a startup that provides data center components to major customers like Amazon Web Services and Microsoft, is considering going public with an initial public offering (IPO) in March. Backed by Sutter Hill Ventures and Intel, the company has been meeting with potential investors and may release its IPO prospectus this month. Astera Labs was valued at $3.15 billion in 2022 and could have a valuation of around $4 billion now, based on comparisons with a similar publicly traded company, Credo Technology. The move comes as the AI industry experiences significant growth and investment. THE INFORMATION

This baby with a head camera helped teach an AI how kids learn language

Researchers at New York University have conducted an experiment to teach an AI model to learn language like a child. They used 61 hours of video footage from a helmet camera worn by a child named Sam, capturing Sam's visual experiences from the age of six months to just over two years. The AI model was trained on 600,000 video frames paired with spoken phrases, learning to match words to objects. The study suggests that some aspects of language learning are achievable from a small dataset, challenging the notion that children must have innate understanding to learn language rapidly. MIT

💰FUNDING

DataSnipper, startup that uses AI to eliminate some of the ‘dread’ in accounting, is valued at $1 billion in latest funding round

AI healthcare software startup Cohere Health just raised $50 million

Zum Raises $140M At $1.3B Valuation To Help Kids Get to School Faster With AI

Silicon Valley’s Codeium raises $65m to combine AI with coding