Today:

Google wants you to label AI-generated images used in Merchant Center

Google requires users to label AI-generated images in its Merchant Center feed, preserving metadata tags like trainedAlgorithmicMedia. The move aims to ensure transparency regarding image creation methods. Google's policy applies to various image attributes within Merchant Center, emphasizing the need for disclosure. 

While currently limited to Merchant Center, future expansion to other Google Search properties is anticipated. This initiative reflects Google's commitment to transparency and user trust. Adhering to these guidelines is crucial for businesses leveraging AI-generated imagery, as failure to comply may lead to policy violations and potential repercussions.

SEARCH ENGINE LAND

Hear your imagination: ElevenLabs to launch model for AI sound effects

ElevenLabs, a rising AI startup, is unveiling a groundbreaking text-to-sound model, revolutionizing content creation. This innovation enables creators to generate sound effects simply by describing their vision in words. Unlike existing AI tools, ElevenLabs' model fills the gap by providing default audio for AI-generated content. From chirping birds to bustling streets, it offers a wide range of sound effects, enriching digital experiences. 

While the model isn't publicly available yet, interested users can sign up for early access. This move reflects ElevenLabs' commitment to advancing responsible AI and capturing a share of the rapidly growing $5 billion AI speech market by 2032.

VENTUREBEAT

OpenAI Completes Deal That Values the Company at $80 Billion

OpenAI, a leading AI company in San Francisco, is now valued at over $80 billion after a deal led by Thrive Capital, tripling its worth in under 10 months. This deal allows employees to cash out their shares, indicating confidence in the company's future despite recent turmoil, including firing its CEO, Sam Altman, last November. 

The investment frenzy in AI firms, fueled by their ability to generate text, sounds, and images, showcases Silicon Valley's eagerness to fund such ventures. OpenAI's valuation surge reflects investors' optimism in the potential of AI technology.

THE NEW YORK TIMES

Chinese EV maker Xpeng to hire 4,000, invest in AI

Chinese electric vehicle manufacturer Xpeng plans to hire 4,000 new employees and invest millions in artificial intelligence, aiming to navigate intense competition in the auto market. CEO He Xiaopeng announced this move after the Lunar New Year holiday, emphasizing the company's determination to thrive amidst economic challenges. Xpeng aims to release 30 new products in three years, contrasting with rivals like Nio cutting costs due to declining domestic demand. 

Despite challenges, Xpeng remains optimistic, with Volkswagen's significant investment highlighting confidence in its growth potential. He foresees 2024 as a pivotal year for the company's expansion and performance enhancement.

REUTERS

AWS launches program to boost AI and tech competency for small and medium businesses

Amazon Web Services (AWS) rolled out a program to aid small and medium-sized businesses (SMBs) in enhancing their tech prowess, including artificial intelligence (AI) skills. The initiative aims to bridge the gap for SMBs lacking technical expertise. AWS partners with experienced firms to offer services ranging from AI to cloud solutions. 

These 30 initial partners, 17 in the U.S., cater to SMBs' specific needs, ensuring a tailored approach. For instance, Magellantv collaborated with Mission Cloud to utilize AI for language dubbing, unlocking new market potential. AWS vets partners based on their SMB focus and track record, offering tailored solutions upon request.

FOX BUSINESS

Kakao Ventures Backs Korean Startup That Makes AI Models Faster And Cheaper To Run

SqueezeBits, a South Korean startup, aims to make AI more affordable and accessible by reducing costs associated with running AI models. Recently backed by Kakao Ventures and other investors, the company raised $2 million in pre-Series A funding. With AI model operation costs soaring, SqueezeBits offers efficiency solutions to democratize AI. By optimizing models and computation processes, they claim to make models three to five times faster and reduce memory usage by four times. 

Despite competition, SqueezeBits sees a market gap due to the slower pace of hardware development compared to AI algorithms, positioning themselves as crucial optimization providers.

FORBES

Singapore’s AI ambitions get a boost with $740 million investment plan

Singapore plans to invest over $743 million in artificial intelligence (AI) over five years, aiming to bolster its global business and innovation status. Deputy Prime Minister Lawrence Wong announced the initiative in the Budget speech, emphasizing the importance of AI in driving economic growth. The investment includes securing advanced chips crucial for AI development and partnering with leading companies to establish AI centers of excellence. Singapore's focus on AI aligns with its National AI Strategy 2.0, aiming to empower workers and businesses. The initiative also promotes responsible AI use, highlighted by the AI Verify tool for governance testing.

CNBC

🧠RESEARCH

Linear Transformers with Learnable Kernel Functions are Better In-Context Models

The paper introduces a tweak to the Linear Transformer model, making it better at learning directly from examples, similar to how humans pick up new info just by seeing it in action. This new approach uses a special math trick to improve how the model remembers and uses new information, showing promising results in language tasks. It's a big step for making language models more efficient and effective at understanding and generating text based on a few examples.

In Search of Needles in a 10M Haystack: Recurrent Memory Finds What LLMs Miss

BABILong is a new benchmark to evaluate models on their ability to extract and process information from long documents. It finds that while standard methods work for shorter texts, augmenting GPT-2 with recurrent memory significantly boosts its performance on tasks with up to 10 million elements, setting a record for the longest input processed by a neural network model. This represents a major advancement in handling extensive sequences in natural language processing.

DataDreamer: A Tool for Synthetic Data Generation and Reproducible LLM Workflows

DataDreamer is an open-source tool designed to enhance synthetic data generation and reproducible workflows with Large Language Models (LLMs). It addresses challenges related to the scale, closed source nature, and lack of standardized tools in LLM research, promoting open science and reproducibility. DataDreamer enables researchers to easily implement advanced LLM workflows through simple code, aligning with best practices for transparency in research. The tool and its documentation are accessible on GitHub for anyone interested in leveraging LLMs for various NLP tasks. 

LLM Comparator: Visual Analytics for Side-by-Side Evaluation of Large Language Models

LLM Comparator is a new visual analytics tool designed for the side-by-side evaluation of large language models (LLMs). It offers an interactive way to analyze automatic evaluation results, helping users understand the nuances of model performance and the qualitative differences in responses. Developed in collaboration with experts from a major tech company, the tool aims to tackle the challenges of scalability and interpretability in model evaluation. The study includes feedback from users who regularly assess their models, highlighting the tool's utility in enhancing understanding and decision-making in LLM development.

GaussianObject: Just Taking Four Images to Get A High-Quality 3D Object with Gaussian Splatting

"GaussianObject" presents a breakthrough in 3D object reconstruction from just four images using Gaussian splatting. It introduces innovative techniques to overcome challenges related to sparse view limitations, offering a framework that significantly enhances multi-view consistency and information completeness. The proposed method shows remarkable results on challenging datasets, outperforming existing state-of-the-art methods and paving the way for efficient and high-quality 3D rendering with minimal input.

🛠️TOP TOOLS

Commonbase - AI sales engineer, enhancing teams' ability to answer technical customer questions up to 10 times faster. It integrates seamlessly with tools like Slack, providing detailed, up-to-date answers about products and documentation.

Code to Flow - a tool that turns code into interactive flowcharts to simplify complex logic instantly.

Friday - create generalist computer agents that automate tasks across the operating system, showing significant improvements in general AI benchmarks.

Rask -  tool for video localization and dubbing, offering translation into 130+ languages.

Codium - boosts developer productivity by generating meaningful tests within IDEs, offering insights into code behavior and integrity. 

📲SOCIAL MEDIA

🗞️MORE NEWS

Google offers 300% hike to retain employee switching to AI startup

Google once offered a worker a 300% raise to stop them from leaving for Perplexity AI, revealing tech giants' efforts to retain talent. CEO Aravind Srinivas discussed tech industry layoffs targeting high-salary employees, with Google warning of more cuts. Sundar Pichai emphasized "tough choices" for investment capacity. Layoffs continue amid ambitious goals. MINT

Days after half a billion Asians went to the polls, Big Tech promises to counter 2024 election misinformation

In response to the recent Asian elections, tech giants like Google, Meta, and Microsoft pledge to combat AI-generated election misinformation. The "Tech Accord to Combat Deceptive Use of AI in 2024 Elections" aims to detect and address deceptive content, but it's vague commitments raise doubts about its effectiveness. THE REGISTER

ChatGPT Used by North Korean Hackers to Scam LinkedIn Users

North Korean hackers are reportedly utilizing AI tools like ChatGPT to conduct phishing scams on platforms like LinkedIn, aiming to gather sensitive information to fund their nuclear weapons program. Despite North Korea's restricted internet access for its citizens, state-backed hacking groups are employing AI to develop sophisticated malware as well. These cyberattacks primarily target employees from various sectors and involve creating credible-looking recruiter profiles on LinkedIn to establish trust with targets. TECH.CO

Commercial AI tool moderately successful at predicting hospitalization-related kidney injury

A recent study by Mass General Brigham evaluated a commercial AI tool's effectiveness in predicting hospital-acquired acute kidney injury (HA-AKI). The tool showed moderate success but struggled with higher-risk patients and severe cases. Further research is needed before clinical use to minimize false positives. The findings were published in NEJM AI. MEDICAL XRESS

Google’s AI Boss Says Scale Only Gets You So Far

Demis Hassabis, CEO of Google DeepMind, discusses advancements in AI, particularly the development of Gemini, a powerful AI model. He emphasizes the need for innovation beyond just scaling up computing power, highlighting the importance of fundamental research. Hassabis also addresses potential safety concerns as AI systems become more agent-like. WIRED

A peek inside Alphabet’s $7 billion growth-stage investing arm, CapitalG

Alphabet's growth-stage investing arm, CapitalG, saw Laela Sturdy take the reins as head last year, succeeding founder David Lawee. Sturdy, a former Google employee, emphasizes team diversity and collaboration. CapitalG's strategy includes leveraging Google's resources and expertise to support portfolio companies, aiming for strategic investments and long-term partnerships. TECHCRUNCH

Can California show the way forward on AI safety?

The new AI legislation proposed by California Senator Scott Wiener aims to regulate the development of powerful AI models, focusing on preventing mass harm. It defines "frontier" models costing over $100 million and requires safety measures like shutdown capabilities. While criticized for potential loopholes, it's seen as a step toward responsible AI governance. VOX

How much electricity does AI consume?

The energy consumption of AI, like training GPT-3, is significant, equivalent to powering 130 homes annually. Companies, secretive about their methods, hide energy use data. Inference tasks vary in energy usage, with image generation being notably high. Overall, the industry's energy consumption could rival that of entire countries by 2027. Efforts to mitigate this impact include calls for energy-efficient AI models and questioning the necessity of AI for certain tasks. THE VERGE