Today:

Hey Folks!

In today's newsletter, we dive into the recent developments in AI and its impact on the future of work. Sam Altman and Anna Makanju discussed the role of AI and its potential consequences. Altman suggests that AI, while powerful, is more of a tool for augmenting human abilities than a job-replacing force, contrary to some earlier fears.

Sam Altman has even suggested that the arrival of Artificial General Intelligence (AGI) may not be as disruptive as once feared. AGI, which could mimic human intelligence, might not drastically change jobs or society as anticipated.

WATCH THE VIDEO ON YOUTUBE

DeepMind approaches gold standard in complex maths in latest AI breakthrough

Google's DeepMind has nearly matched the top high school students in solving tough geometry problems, a big step in AI. Their program, AlphaGeometry, answered 25 out of 30 questions from the International Mathematical Olympiad, just shy of the human gold medal level. 

This shows AI's growing smarts in math, but also the challenges left. AlphaGeometry combines language skills and logical thinking, a method inspired by the idea of fast pattern recognition paired with deep thought. Despite its success, the AI struggled with some complex problems. DeepMind aims to tackle math puzzles even humans can't solve yet, but they're not entering the Olympiad just now.

FINANCIAL TIMES

Google introduces ‘Circle to Search, a new way to search from anywhere on Android using gestures

Google just rolled out "Circle to Search" for Android, making it a snap to search from any app. You can use gestures like circling, tapping, or scribbling to look stuff up. Say you're watching a cooking video and spot a Korean corn dog. Just circle it and ask, "Why are these so popular?" Or, if you're texting about a restaurant, tap its name to get more info. 

The search results vary, sometimes giving traditional answers, other times using AI, especially if you're in Google's Search Generative Experience experiment. This feature isn't just cool; it's Google beefing up its search game. 

With websites getting spammy and AI chatbots stepping in, Google's trying to stay relevant. Circle to Search starts January 31 on new Galaxy S24s, Pixel 8, and other top Android phones, in all languages and places where these phones sell. More phones will get it down the line.

TECHCRUNCH

Tiktok Owner Bytedance Quietly Launched 4 Generative AI Apps Powered By OpenAI’s GPT

ByteDance, TikTok's Chinese parent company, has stealthily released four AI apps outside China. These apps, Cici AI, Coze, ChitChop, and BagelBell, don't mention ByteDance's involvement. They use OpenAI's GPT tech through Microsoft Azure. Cici AI, ChitChop, and Coze let users create chatbots, while BagelBell crafts interactive stories. 

With over 10 million downloads, Cici is the hit. Coze is the only one available in the U.S. ByteDance's move reflects the AI race among tech giants. There's concern about data privacy and potential Chinese government influence, as ByteDance might share user info within its group. OpenAI recently cut ByteDance's direct API access, but ByteDance still uses its tech via Microsoft.

FORBES

Samsung and Google Cloud Join Forces To Bring Generative AI to Samsung Galaxy S24 Series

Samsung and Google Cloud are teaming up to bring some cool AI tech to the new Samsung Galaxy S24. This means the S24 will be the first to have Google's Gemini Pro and Imagen 2, making it a big deal in the smartphone world. 

Gemini Pro's a smart cookie that gets text, images, and videos and works with different apps on the phone. It's all about helping folks summarize stuff in their notes, recordings, and texts. Plus, with Imagen 2, users can edit photos like a pro, adding cool AI-generated touches. 

This partnership is a big step for Samsung, using Google's top AI for complex tasks and making their phones smarter. It's all about giving users a more powerful and creative experience with their phones.

SAMSUNG

Android Auto is getting new AI-powered features, including suggested replies and actions

Google's Android Auto is stepping up its game with some cool new AI features. It's getting smarter at handling texts while you're driving. Instead of reading out every message in a group chat, it'll give you a quick summary. Plus, it'll suggest replies or actions based on your texts. Say your buddy asks when you'll show up – Android Auto can automatically offer to send your ETA. 

Also, it'll make it easier to start navigating to a location or call someone with just one tap. Your Samsung Galaxy's wallpaper and icons will soon show up on your car's screen, making the switch from phone to car super smooth. 

TECHCRUNCH

🧠RESEARCH

stabilityai/stable-code-3b

Stable-code-3b, developed by Stability AI, is a powerful 2.7 billion parameter model trained on diverse code and text datasets. Excelling in 18 programming languages, it demonstrates top-notch performance in coding tasks. Its Fill in Middle feature supports long contexts, and it's trained on open-source datasets with a special focus on mathematical domains. However, users should fine-tune and evaluate it carefully for specific applications, considering potential biases and limitations from its training data.

MotionCtrl: A Unified and Flexible Motion Controller for Video Generation

The MotionCtrl SVD by TencentARC is an innovative tool designed for motion control. This space utilizes advanced algorithms to manipulate and control motion data effectively. It's particularly useful for animation and gaming, where precise and dynamic motion control is crucial. The technology behind MotionCtrl SVD demonstrates TencentARC's commitment to pushing the boundaries in digital motion and animation. 

Scalable Pre-training of Large Autoregressive Image Models

Apple  introduces AIM, a set of vision models similar to large language models, showing two main findings: bigger models and more data improve visual feature performance, and the model's objective function score predicts its success in practical applications. A notable example is a 7 billion parameter AIM trained on 2 billion images, achieving 84% accuracy on ImageNet-1k, with potential for even better performance, suggesting a new approach to training large-scale vision models without needing image-specific strategies.

E^2-LLM: Efficient and Extreme Length Extension of Large Language Models

Alibaba introduces E^2-LLM, a method for efficiently extending the context length of large language models (LLMs) without needing extensive computational resources or long-context training data. It uses short training data (e.g., 4k length), one-time training, and two augmentation methods based on RoPE position embeddings to make the model more adaptable to varying context lengths during inference. This approach significantly reduces training costs and improves model robustness for long-context tasks, as demonstrated in various benchmark datasets.

Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translation

Microsoft presents Contrastive Preference Optimization (CPO), a new approach to improve machine translation (MT) in moderate-sized large language models (LLMs). Unlike traditional supervised fine-tuning that often struggles with the quality of reference data, CPO trains models to avoid producing just adequate translations. This method, applied to the ALMA model with minimal training data, achieved significant performance enhancements, matching or surpassing the winners of the WMT competitions and GPT-4 on various WMT test datasets.

HexaGen3D: StableDiffusion is just one step away from Fast and Diverse Text-to-3D Generation

Qualcomm introduces HexaGen3D, a novel approach for fast and diverse text-to-3D generation. It overcomes the challenge of data scarcity in 3D asset generation by leveraging pretrained 2D diffusion models. HexaGen3D fine-tunes these models to predict six orthographic projections and a corresponding latent triplane, which are then decoded into textured meshes. This method generates high-quality, varied objects from text prompts in just 7 seconds, offering a significant improvement in quality-to-latency trade-offs compared to existing methods.

🛠️TOP TOOLS

Friz - AI-powered tool that streamlines social media management for direct-to-consumer brands. It offers services like AI brainstorming, post generation, scheduling, and analytics in one place. 

MarketMind -  AI-powered market analysis tool that streamlines the market research process, providing real-time, comprehensive insights. 

Personage - AI-powered platform designed to help creators monetize their audience using a Telegram chatbot. Users can create AI companions or clones and integrate them with Telegram bots. 

Equals - AI Assist for next-generation spreadsheets. Powered by OpenAI's GPT-4 Turbo, it provides proactive error detection, one-click fixes, and suggested prompts.

Gotalk.ai - AI voice generator, transforming written text into lifelike speech quickly. Ideal for YouTube, podcasts, and phone systems, it's easy to use, provides natural-sounding voices with customization options, and is suitable for various applications across industries.

📲SOCIAL MEDIA

🗞️MORE NEWS

Top ex-Google researchers raise $30 mln for Tokyo-based AI lab

Sakana AI, a new Tokyo-based artificial intelligence startup by ex-Google researchers, bagged $30 million in seed funding. The goal is to make Tokyo an AI hub, like OpenAI in San Francisco. Their approach differs by using multiple small models instead of one large one, with a focus on Asian languages. This initiative is part of a wider trend of former Google researchers founding successful AI ventures. For more details, you can read the full article on Reuters here. REUTERS

OpenAI Board’s Search For New Directors Includes A Rival

OpenAI's board is considering adding Ali Ghodsi, CEO of Databricks, as a director. This move is notable because Databricks is seen as a competitor to OpenAI. The invitation suggests potential changes within OpenAI's board and challenges for CEO Sam Altman. This strategic choice indicates a shift in OpenAI's approach to collaboration and competition in the AI industry. THE INFORMATION

Google adds AI-powered overviews for multisearch in Lens

Google has enhanced its Google Lens feature with AI-powered overviews in multisearch. Users can now point their camera or upload an image to Lens and ask questions for AI-generated insights. This update adds informative answers, like plant care instructions, to visual searches. It's integrated with Google's new Circle to Search gesture feature but differs from Google's GenAI Search Generative Experience. The update, available in the U.S. in English, aims to provide more accurate and relevant search results in a cluttered, SEO-dominated web. It's not confined to Google Labs and is part of broader generative AI advancements in Google's services. TECHCRUNCH

What’s Rabbit working on with Perplexity AI?

Perplexity AI, known for its generative AI search technology, is set to reveal a collaborative project with Rabbit, the creators of the popular Rabbit R1 AI gadget. The announcement of their joint venture is scheduled for tomorrow at 2PM PT / 5PM ET. Details about the nature of this collaboration or the specifics of the project are currently under wraps until the official announcement. THE VERGE

Amazon brings its AI-powered image generator to Fire TV

Amazon is introducing an AI-powered image generator on Fire TV devices, specifically for the second-generation Fire TV Stick 4K Max and Fire TV Omni QLED Series in the U.S. The feature, based on Amazon's Titan Image Generator, allows users to create images using voice commands through Alexa. Users can request specific scenes like a "fairy landscape" and then select from artistic styles to customize their TV background. This is part of the Fire TV Ambient Experience, which includes displaying art or personal photos. TECHCRUNCH

A new AI-based risk prediction system could help catch deadly pancreatic cancer cases earlier

A new AI system developed by MIT's CSAIL and a researcher from Beth Israel Deaconess Medical Center may improve early detection of pancreatic cancer, particularly its most common form, pancreatic ductal adenocarcinoma (PDAC). This AI system, named PRISM, analyzes electronic health records to predict the risk of developing PDAC within 6-18 months, aiming for early-stage detection. The system showed significant improvement over current screening methods, potentially leading to earlier treatment and better outcomes. MIT

China issues draft guidelines for standardising AI industry

China has released draft guidelines to standardize its artificial intelligence (AI) industry. The guidelines aim to establish over 50 national and industry standards for AI by 2026 and contribute to forming over 20 international AI standards. This initiative is part of China's efforts to catch up with global AI leaders like the U.S., following the impact of innovations such as OpenAI's ChatGPT. The guidelines focus on early opportunities in AI development, targeting key technologies and application projects. REUTERS

EU calls for laws to force greater algorithmic transparency from music-streaming platforms

The European Parliament is advocating for new legislation to increase fairness and transparency in music streaming. This includes a bill requiring platforms to disclose their recommendation algorithms and label AI-generated songs. The move aims to provide better visibility for European artists and prevent streaming manipulation. It also seeks to ensure fair revenue distribution among all artists involved in a recording. This initiative aligns with similar efforts in other countries to support local music industries and artists. TECHCRUNCH

Amazon’s new AI feature will write you a bedtime story about hard drives

Amazon is experimenting with a new AI feature in its app that allows customers to ask specific questions about products. This tool can provide information like product dimensions or battery life and even create creative content like jokes or stories related to the products. However, it's designed to stay within certain limits and doesn't engage in personal interactions like flirting. This feature is part of Amazon's broader integration of AI in its retail services, including review summarization and fake review detection. Amazon's cloud division, AWS, is also advancing in generative AI with tools like the Titan Image Generator. THE VERGE

FDA Clears AI-Powered Handheld Device That Can Detect Skin Cancer

The FDA has approved DermaSensor's innovative handheld device, which uses AI to non-invasively detect skin cancer. It can identify basal cell carcinoma, squamous cell carcinoma, and melanoma. The device analyzes skin lesions through an AI algorithm, providing immediate results. This clearance is a significant advancement in skin cancer detection, complementing traditional methods. DermaSensor's effectiveness has been highlighted in studies, and it's offered via subscription pricing. This development marks a milestone in AI's application in medical diagnostics. FORBES

Rapidly Growing Radiology AI Company, Rad AI, Strikes Partnership With Google Cloud

Rad AI, a tech startup in AI and radiology, has partnered with Google Cloud to enhance its platforms, Rad AI Reporting and Omni Impressions. These platforms aid radiologists by generating personalized report impressions. The partnership aims to utilize Google's MedLM models and cloud services to improve report accuracy and radiologist workflow. Rad AI, having raised nearly $25 million in Series A funding, now serves about 30% of the U.S. radiology market. This collaboration represents a significant step in applying AI to radiology, addressing issues like radiologist burnout and improving clinical care delivery. FORBES

OpenAI Changes Its Stance, Announces That It Is Working With the Pentagon

OpenAI recently revised its usage policy, removing specific prohibitions on military applications. Following this, the company announced a partnership with the Pentagon. This collaboration includes working on open-source cybersecurity software and exploring ways to help prevent veteran suicide. OpenAI's policy adjustment allows for national security collaborations that align with its mission, while still prohibiting the use of its tools for harm or weapon development. This change opens the door for OpenAI to pursue military contracts. THE MESSENGER

Google’s Defense Against OpenAI Talent Grab: Special Stock

Google is using a special stock compensation program to retain top AI researchers at its DeepMind unit amidst staff cuts. This move counters OpenAI's multimillion-dollar offers to lure AI talent. Researchers at DeepMind are receiving large grants of restricted stock, sometimes amounting to millions per individual, to ensure they remain with Google during this competitive phase. THE INFORMATION