• Posts
  • OpenAI Launches GPT Mentions | Brave Browser Adds Mixtral | Lying LLMs & Lucid Dreaming

OpenAI Launches GPT Mentions | Brave Browser Adds Mixtral | Lying LLMs & Lucid Dreaming

PLUS: Grok 1.5 Launches Next Month, Improved Windows 11 Features and more...

Sponsored by


Your info is on the dark web

Every day, data brokers profit from your sensitive info—phone number, DOB, SSN—selling it to the highest bidder. And who’s buying it? Best case: companies target you with ads. Worst case: scammers and identity thieves.

It's time you check out Incogni. It scrubs your personal data from the web, confronting the world’s data brokers on your behalf. And unlike other services, Incogni helps remove your sensitive information from all broker types, including those tricky People Search Sites.

Help protect yourself from identity theft, spam calls, ID theft and health insurers raising your rates. Plus, just for our readers: Get 55% off Incogni using code PRIVACY.

AI NEWS | OpenAI Launches GPT Mentions | Brave Browser Adds Mixtral | Lying LLMs & Lucid Dreaming

OpenAI is introducing "GPT Mentions," a feature allowing users to summon various GPT agents for specific tasks within ChatGPT, much like using a command in Slack or Discord. This innovation aligns with OpenAI's vision, discussed in November, of an "army of agents" performing programmed tasks. 

Brave browser has incorporated Mixtral, an open-source AI model, to enhance its AI assistant, Leo. Leo can now summarize web pages, transcribe videos, and answer questions more efficiently.

Lastly, an intriguing AI application for lucid dreaming is being developed. Named Morpheus One, this technology aims to enable and stabilize lucid dreams, potentially allowing users to control their dream experiences. Beta testing is set for Spring 2024.

Tesla to build Dojo supercomputer at NY gigafactory, Elon Musk confirms

Tesla is set to build its advanced Dojo supercomputer at its Gigafactory in Buffalo, New York. This move, part of a $500 million investment, aims to enhance Tesla's AI capabilities, particularly in autonomous driving. Dojo will process huge volumes of data from Tesla vehicles, potentially making it one of the world's most powerful computers. 

This project, announced by New York Governor Kathy Hochul, is expected to create numerous jobs and stimulate regional innovation. Additionally, a separate AI supercomputer project will receive substantial funding at the State University of New York (SUNY).

Ex-Salesforce Co-CEO to Win Funding for AI Startup at Near-$1 Billion Value, Sources Say

Bret Taylor, previously a big cheese at Salesforce and now an OpenAI board member, is about to score a huge deal for Sierra, his AI startup focused on business needs. He started it last year with a buddy from Google. The scoop is that Sequoia Capital, a major player in the investment world, is about to throw down $85 million. 

This cash infusion could shoot Sierra's value close to a cool billion. Neither Sierra nor Sequoia are spilling the beans on this yet, but it's a pretty big deal in the AI scene.

ElevenLabs reportedly banned the account that deepfaked Biden's voice with its AI tools

ElevenLabs, a company providing AI voice cloning, banned a user for creating a fake audio of President Biden. This deepfake, used in a New Hampshire robocall, falsely told folks not to vote in the state primary. Security experts at Pindrop figured out ElevenLabs' tech was used. They cleaned up the call's audio and compared it to over 120 other voice fakes, finding a near perfect match. 

ElevenLabs, aware of Pindrop's findings, is still digging into it but has already suspended the culprit's account. This incident highlights the potential abuse of voice cloning tech in messing with elections and spreading false info.

Grok-1.5 to arrive next month, featuring improvements: Elon Musk

Elon Musk, head of Tesla and xAI, announced that a new version of their Grok language model is coming next month, boasting major upgrades. This update follows claims that Grok, like its rival ChatGPT, often spews out false info. Musk promises the new version will be better all-around, hinting at less nonsense talk. 

Grok, xAI's first release, is a subscriber-only AI based on the "Hitchhiker's Guide to the Galaxy," designed to answer and even suggest questions. It offers a playful "Fun Mode" or a standard "Regular Mode." Since xAI's debut last summer, it's become a benefit corporation, aiming to make a positive social and environmental impact. Musk, who's been wary about AI risks, insists on significant control over AI and robotics development at Tesla.

Windows 11’s AI-powered Voice Clarity feature improves your video chats, plus setup has a new look (finally)

Windows 11's latest preview build, aimed at early testers, brings a cool AI feature called Voice Clarity to improve your video chats. Originally just for Surface gadgets, this feature now helps anyone using Windows 11 sound clearer on calls by cutting out echo and background noise. But, it only works if the apps you're using are set up for it, like Microsoft's Phone Link or WhatsApp. Plus, there's a neat trick where photos taken on your Android phone can instantly pop up on your Windows 11 PC for easy editing. 

Also, Microsoft finally updated the Windows 11 setup look – it's about time! This change is part of several updates in the new build, making Windows 11 more user-friendly, especially for remote work or family video calls.


"Diffuse to Choose: Enriching Image Conditioned Inpainting in Latent Diffusion Models for Virtual Try-All," by Amazon addresses the challenge of virtual product visualization in online shopping. It introduces a new image-conditioned inpainting model that efficiently balances fast processing with high-detail retention. This model improves upon traditional diffusion models and DreamPaint by directly incorporating fine-grained features from reference images, ensuring accurate, detailed, and fast virtual product visualization. The model's effectiveness is demonstrated through extensive testing on various datasets​.

"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence" introduces the DeepSeek-Coder series, a collection of open-source code models ranging from 1.3B to 33B in size. Developed to enhance code intelligence in software development, these models are trained on a 2 trillion token high-quality code corpus. They outperform both open-source and closed-source counterparts like Codex and GPT-3.5 in various benchmarks. DeepSeek-Coder is unique for its extensive training and permissive license, enabling both research and commercial use​. 

"Rethinking Patch Dependence for Masked Autoencoders" challenges the need for self-attention in masked autoencoders (MAE) for image processing. It introduces Cross-Attention Masked Autoencoders (CrossMAE), which use cross-attention between visible and masked tokens, omitting self-attention. This approach significantly reduces computation while maintaining or improving performance. CrossMAE demonstrates superior results in ImageNet classification and COCO instance segmentation tasks, using less decoding compute compared to traditional MAEs.

"Unitxt: Flexible, Shareable and Reusable Data Preparation and Evaluation for Generative AI" presents Unitxt, an innovative library designed for customizable text data preparation and evaluation in generative language models. It revolutionizes traditional text processing by offering a structured, modular, and customizable solution. Unitxt integrates with popular libraries, allowing for easy customization and sharing. It includes model-specific formats, task prompts, and other dataset processing tools, centralized in the Unitxt-Catalog to encourage collaboration and exploration.

 "FP6-LLM: Efficiently Serving Large Language Models Through FP6-Centric Algorithm-System Co-Design" presents a novel approach for efficiently running large language models (LLMs) by utilizing six-bit quantization (FP6). This method reduces the size of LLMs while maintaining their quality across various applications. The paper introduces TC-FPx, a full-stack GPU kernel design that supports various quantization bit-widths. This enables more efficient LLM inference, achieving up to 2.65 times higher throughput compared to the FP16 baseline, and making it possible to run massive models like LLaMA-70b on a single GPU.

"Human-LRM," a Large Reconstruction Model designed to predict human Neural Radiance Fields (NeRF) from a single image. It demonstrates remarkable adaptability, training on extensive datasets with 3D scans and multi-view capture. The paper proposes a novel strategy to distill multi-view reconstruction into single-view via a conditional triplane diffusion model, addressing variations in human body shapes when observed from a single view. This allows for the reconstruction of the full human body from an occluded image. Human-LRM outperforms previous methods significantly on various benchmarks. 


Weco - a platform that focuses on improving workplace well-being and performance. It uses AI to analyze various aspects of an organization, including culture, leadership, and team dynamics. 

Axonn - AI-driven platform for content creation and social media management, specifically designed to enhance personal branding. 

Yuna - AI-powered mental health app offering 24/7 support. It uses cognitive behavioral therapy patterns to aid personal growth, emotional clarity, and anxiety relief. Yuna ensures privacy and provides research-driven, confidential conversations.

Holly - AI-powered virtual recruiter designed to automate hiring processes. It offers features like sourcing candidates from over 750 million profiles, AI-driven matching, and running processes akin to top recruitment agencies. 

GetWebsite.Report - AI-powered web service that optimizes landing pages by analyzing user interface, experience, design, content, and SEO.

Vx.dev - GitHub-powered AI tool that simplifies development by generating code based on user-submitted GitHub issues.



Google Update Reveals AI Will Start Reading All Your Private Messages

Google's new AI, Bard, will analyze Android users' private messages to enhance messaging with features like mood-based responses and tailored assistance. While it promises on-device analysis for privacy, concerns remain about data collection and usage. Bard's integration into texting apps signals a major shift in smartphone privacy and AI chat integration, challenging users to balance convenience with privacy concerns. FORBES

Musk Plans to Buy AMD Chips as Tesla Loads Up on AI Hardware

Elon Musk, Tesla's CEO, plans to buy AMD chips to boost AI capabilities. Recently, AMD launched the MI300 series, boasting superior AI software performance. Musk's announcement follows his reveal of a $500+ million investment in Nvidia hardware this year. The move highlights Tesla's significant investment in AI technology. BLOOMBERG

X blocks Taylor Swift searches after fake AI videos go viral

X, the social media platform owned by Elon Musk, has blocked searches for Taylor Swift after sexually explicit deepfake images of the singer spread widely on the platform. This incident highlights the challenge social media platforms face with deepfakes, which are generated using AI and can be used to portray individuals in compromising situations without their consent. X's action comes amid increasing pressure for platforms like X, Meta, TikTok, and YouTube to address the abuse of deepfake technology. FINANCIAL TIMES

Japan and U.S. agree on AI research for drones to assist new fighter jet

Japan and the U.S. are teaming up on AI research for drones to work with Japan's upcoming fighter jet, a project they're doing with Britain and Italy for a 2035 launch. The U.S. is still beefing up defense ties with Japan, focusing on smart, self-flying systems. The plan? Use cutting-edge AI and machine learning to upgrade unmanned air vehicles (UAVs), making them smarter companions for Japan's future jets. The drones will gather intel, flying close to the jets. This move follows Japan's search for new allies in jet development after hitting a snag with U.S. info-sharing rules. It's a big step for Japan, working on defense gear with countries other than the U.S., especially with China's growing assertiveness in mind. THE JAPAN TIMES

AI deep learning decodes perchlorate salts crystals

Researchers at Tokyo University of Science are using deep learning to safely study explosive perchlorate salts. Traditionally, these salts are tricky to study due to their tendency to blow up with heat or shock. The team, led by Professor Takashiro Akitsu, is using AI to avoid these dangers. They're focusing on salen-type metal complexes, which are hard to experiment on directly. The AI analyzes the structure of these salts, helping to figure out what makes them explosive. This method uses a ton of data from the Cambridge Crystal Database and a fancy AI technique called variational autoencoder. The big takeaway? This AI method could revolutionize how we understand crystal structures and might lead to breakthroughs in drug discovery and catalysis research. It's a safer, smarter way to study really reactive stuff. INTERESTING ENGINEERING

Instacart is using AI art. It's incredibly unappetizing.

Instacart is using AI to create recipe photos and write recipes, but the results are somewhat unappetizing. The AI-generated images often have unrealistic features, like a hot dog with a tomato interior or a lemon-lettuce hybrid. These images, confirmed by AI-detection software, are quirky and sometimes disturbing. Instacart acknowledges the AI's imperfection and advises users to adapt recipes as needed. This initiative is part of Instacart's broader use of AI, including an "Ask Instacart" model developed with OpenAI. Instacart's stock, however, is currently down 25% from its IPO price. BUSINESS INSIDER

Italy fines first city for privacy breaches in use of AI

Italy's privacy watchdog fined the city of Trento €50,000 for violating data protection laws in its AI-based street surveillance projects. This marks the first such penalty for a local administration in Italy. The GPDP found multiple privacy breaches, including inadequate data anonymization and improper sharing with third parties. Trento is considering an appeal, citing a need for updated legislation for AI use. This action aligns with Italy's focus on AI regulation during its G7 presidency and EU moves to regulate AI, especially in biometric surveillance. REUTERS

ByteDance has a new AI model that lets users instantly speak in another person's voice. Researchers admitted the AI can be used for 'fraud.'

ByteDance, the parent company of TikTok, has developed an AI tool called StreamVoice, which allows users to instantly mimic another person's voice with just a single utterance. This technology has raised concerns about potential misuse for fraud or spreading misinformation. Researchers at ByteDance and Northwestern Polytechnical University in China created StreamVoice, which operates in real-time with minimal latency. While the technology showcases the rapid advancement of AI in voice conversion, it also underscores the growing risks associated with deepfake technologies in various applications, including political manipulation. BUSINESS INSIDER

New Texas Center Will Create Generative AI Computing Cluster Among Largest of Its Kind

The University of Texas at Austin is launching the Center for Generative AI, featuring one of academia's largest GPU computing clusters, named Vista. This initiative is part of UT's aim to be at the forefront of AI research and application. The center will focus on areas like biosciences, healthcare, computer vision, and natural language processing, and will work closely with various industries. UT's strong foundation in AI, including hosting the powerful Frontera supercomputer, positions it as a leader in AI research and development. The center is expected to drive significant advancements in AI and its applications across multiple fields. UT NEWS

AI-model developed to detect extremist users, ISIS-related content on X

Researchers have developed an AI model to detect extremist users and ISIS-related content on social media. This model, using tweets from 2009-2021, applies machine learning and natural language processing to identify users sharing such content and classify images related to ISIS. The initiative could aid social media companies in timely identification and restriction of extremist accounts, helping to reduce their influence online. The research was conducted by Pennsylvania State University and aims to address the growing use of social media for extremist propaganda and recruitment. THE ECONOMIC TIMES

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Join the conversation

or to participate.