Today:

OpenAI's Simulator STUNS the Entire Industry! UNREAL Physics Model, Emergent Abilities and AGI.

OpenAI is making waves with its latest AI, Sora, pushing the boundaries of video generation tech. Sora can create videos so lifelike, it's hard to tell them from reality. Behind its showcase lies a vast, unseen potential, hinting at OpenAI's deep, strategic play. 

Sora's standout feature is its ability to mimic real-world physics and details in video games, making scenes like Minecraft come to life with stunning realism. This leap in technology suggests Sora isn't just a tool for creating videos but a data-driven physics engine that simulates complex, real or imaginary worlds. 

WATCH THE VIDEO ON YOUTUBE

Reddit has reportedly signed over its content to train AI models

Reddit's stepping into the AI game, striking a $60 million deal to let its mountain of posts and comments train AI brains. This move comes as Reddit gears up for a huge $5 billion public offering. 

Last year, Reddit tried to charge for API access, sparking a massive protest from its community, including a brief shutdown and hacker threats. Now, by possibly feeding user content into AI, Reddit's stirring the pot again. Users are wary, debating the ethics of using their digital chatter to power AI advances. Amidst this, Reddit's also rolling out changes like a new badge system and stricter ad rules, which hasn't exactly been met with open arms.

MASHABLE

Google’s Chess Experiments Reveal How to Boost the Power of AI

Google's chess AI experiments have leveled up the game, showing that mixing different strategies can outsmart even their top AI, AlphaZero. Tom Zahavy, a Google DeepMind scientist, got hooked on chess again during the COVID-19 lockdowns, inspired by chess legends and Netflix's "The Queen's Gambit." Diving deep, he explored how AI could tackle complex chess puzzles that stump even the smartest programs. 

The breakthrough? A supercharged AI that combines up to 10 different strategies, outplaying AlphaZero by embracing a variety of tactics. This AI dream team not only excels at chess but hints at solving real-world problems with a mix of creativity and diverse approaches, proving two (or ten) heads are better than one.

WIRED

Sora can create video collages, too.

OpenAI's latest text-to-video tool, Sora, can also whip up video collages, as demonstrated by one of its employees. 

While the AI's capabilities are undeniably impressive, folks are scratching their heads over the peculiar content in the upper right frame.

THE VERGE

Google Teases Innovative New Android Abilities With Gemini AI

Google's Gemini AI, set to join Android, impresses with practical functions over flashy feats. It efficiently organizes emails, finds dining spots, and streamlines tasks, aiming to simplify smartphone use. While not flawless—struggling with certain commands—it hints at a future of seamless integration and time-saving conveniences. The potential extends beyond personal assistance, envisioning improved smart home management and expanded capabilities. 

Despite ongoing development, Gemini signifies a shift towards leveraging AI for everyday efficiency, promising a more user-friendly mobile experience beyond gimmicks and complexities.

FORBES

FTC seeks to ban impersonation fraud as AI enables deepfakes

The FTC is moving to outlaw impersonation scams, eyeing a crackdown on deepfake tricks targeting individuals. With AI tech advancing, regulators aim to protect consumers from fake videos and voices. Proposed rules would make it illegal to use AI for impersonation, tackling voice cloning and phony videos. 

Already, the FCC banned AI voices in scam calls. Experts warn of the peril as scammers could mimic trusted contacts or famous figures. This push aligns with a broader trend of AI regulation, as seen in New York's proposal to criminalize deceptive AI use. It's the start of AI oversight amid growing concerns.

AXIOS

Google open sources file-identifying Magika AI for malware hunters and others

Google has released Magika, an AI-powered file identifier, to boost cybersecurity efforts. Magika swiftly detects file types, aiding in malware detection and intrusion prevention. It’s integrated into Google services like Gmail and Chrome for enhanced security. The move aligns with Google’s AI Cyber Defense Initiative, aiming to empower defenders against evolving threats. Phil Venables and Royal Hansen advocate for AI's pivotal role in cybersecurity, urging proactive measures. 

Despite Magika's 99% accuracy, it occasionally misclassifies files. Google plans to train startups and fund research, fortifying global cybersecurity efforts. Magika signifies Google's commitment to leveraging AI for a safer digital landscape.

THE REGISTER

🧠RESEARCH

Hierarchical State Space Models for Continuous Sequence-to-Sequence Modeling

Hierarchical State-Space Models (HiSS) is a new method for predicting sequences from raw sensor data, like those from medical devices to robotics. Traditional models struggle with the non-linearity and noise in real-world sensors. HiSS uses a layered approach for better accuracy, beating other sequence models by at least 23% on mean squared error across six datasets. It scales well with small datasets and works with current data-filtering techniques. 

Zero-Shot Unsupervised and Text-Based Audio Editing Using DDPM Inversion

The paper explores innovative zero-shot editing techniques for audio using DDPM inversion on pre-trained diffusion models. It introduces a method for text-based audio editing and a novel unsupervised approach to discover semantically meaningful editing directions. Applied to music, it allows for various interesting modifications, like adjusting specific instruments or melody improvisation. 

GES: Generalized Exponential Splatting for Efficient Radiance Field Rendering

Generalized Exponential Splatting (GES) is a method enhancing 3D scene modeling efficiency by utilizing the Generalized Exponential Function, reducing memory usage compared to traditional Gaussian Splatting. GES accurately captures sharp edges in scenes, outperforming Gaussian methods in natural signal representation and reducing memory footprint. It also achieves faster rendering speeds and competitive performance in novel-view synthesis with less memory. 

DreamMatcher: Appearance Matching Self-Attention for Semantically-Consistent Text-to-Image Personalization

"DreamMatcher" presents a novel method for text-to-image personalization that maintains semantic consistency. Unlike traditional approaches that struggle with accurately mimicking reference appearances, DreamMatcher uses a plug-in method focusing on semantic matching to replace target values with reference values. This ensures the generation of diverse structures without altering the structural integrity of pre-trained models. The approach also includes a semantic-consistent masking strategy, showing significant improvements in complex scenarios and compatibility with existing models.

How to Train Data-Efficient LLMs

This paper discusses data-efficient methods for training large language models (LLMs), focusing on optimizing the balance between model quality and resource consumption. Two main approaches are highlighted: Ask-LLM, which uses zero-shot reasoning of LLMs for assessing training example quality, and Density sampling, aimed at maximizing data coverage and diversity. Through extensive testing, these methods demonstrated significant improvements in training efficiency, with Ask-LLM enabling faster convergence and performance gains even when significantly reducing the dataset size.

🛠️TOP TOOLS

Shortwave - AI Assistant that simplifies email tasks like drafting, searching, and analyzing.

Univerbal - AI language tutor app designed to facilitate language learning directly from your pocket.

Circleback - innovative tool designed to automate the process of taking meeting notes, generating action items, and ensuring data privacy.

Superlist - a versatile list-making app perfect for both team projects and personal tasks, offering features like real-time collaboration, multiplatform support, and privacy control.

YourMove AI - transforming your profile and messages to boost matches and dates. 

Bluedot - AI-powered Chrome extension for Google Meet

📲SOCIAL MEDIA

🗞️MORE NEWS

Scale, Anthropic Show Pickup in AI Venture Deals

Investors’ interest in AI startups rebounded after a brief slowdown earlier this year, signaling renewed enthusiasm. Scale AI, specializing in teaching AI to recognize patterns, seeks funding at a valuation of up to $14 billion, twice its previous valuation. Despite initial caution, investors are eager to back promising AI ventures. THE INFORMATION

Cooler Screens plans to use AI to change how consumers shop

Cooler Screens, a smart screen developer, plans to revamp the freezer aisle in grocery stores using AI to tailor ads to customers. CEO Arsen Avakian explains how the technology personalizes content based on shoppers' preferences. This innovation aims to enhance the shopping experience and help consumers find products more suited to their needs. TECHCRUNCH

Clubhouse’s new feature turns your texts into custom voice messages

Clubhouse introduces a new feature allowing users to send texts converted into custom voice messages, aiming to maintain relevance amidst declining user engagement. The app's AI recreates users' voices for seamless communication, emphasizing the feeling of real-time conversation. The move follows a trend of tech companies integrating personalized AI features. TECHCRUNCH

Little-known startup takes the AI weather prediction crown

In the competitive realm of AI weather prediction, Stanford graduates' startup, WindBorne Systems, outpaces tech giants like Huawei and Google DeepMind. Utilizing a fleet of inexpensive weather balloons and AI techniques akin to ChatGPT, WindBorne surpasses DeepMind's accuracy, garnering government contracts and eyeing commercial markets. Their innovative approach combines deep learning and physics models for enhanced forecasting accuracy, showcasing the potential of AI in real-world applications. SEMAFLOR

Panorámica's interactive-AI machine lets "anybody be a Mexican designer"

Panorámica's AI machine showcased at Zona Maco art fair in Mexico City lets anyone create Mexican-inspired furniture. Users input prompts like historical periods and color preferences, generating unique designs. The collective challenges traditional views of design authorship, believing the machine symbolizes a new era in creativity, redefining Mexican design globally. DEZEEN

Chinese scientists use massive databank and AI to try to predict dementia 15 years before symptoms start

Chinese scientists have made strides in predicting dementia onset up to 15 years in advance using blood samples and AI. Analyzing proteins from 50,000 individuals, they developed a predictive model, aiming for early intervention. Despite limitations, the study offers hope for understanding and treating dementia. AI played a crucial role in this research, identifying key biomarkers for future dementia prediction from plasma proteins. The study utilized a massive UK Biobank cohort and AI algorithms to assess dementia risk, potentially revolutionizing early diagnosis and intervention strategies. SCMP

Deepfake porn is a huge problem — here are some of the tools that could help stop it

Deepfake porn, generated by AI, poses serious threats, with 96% of deepfakes being pornographic. Tools like digital watermarks and defensive measures like Nightshade offer some protection, while regulation, though challenged, aims to deter malicious use. Legislation varies globally, but friction in creation and distribution can mitigate harm. BUSINESS INSIDER

Could AI personas attend your work meetings for you? One tech CEO says yes — and by the end of the year

AI avatars could soon attend work meetings on behalf of employees, according to Otter's CEO. These avatars would mimic employees' behavior and speech patterns, potentially saving time and boosting productivity. Challenges remain in ensuring these avatars can interact appropriately and convey emotional cues. BUSINESS INSIDER

What Sam Altman's chimerical trillions say about AI hype

Sam Altman's proposal for increasing silicon supply for AI, which comes with a price tag of $5 trillion to $7 trillion, has sparked discussions about the feasibility and realism of such grandiose figures. While Altman's pitch underscores the momentum and ambition behind the AI revolution, critics argue that these numbers are far-fetched and detached from reality. The broader AI discourse seems to have drifted away from practicality, with Altman's proposals being a prime example of the inflated expectations surrounding AI's potential. AXIOS