Need a personal assistant? We do too, that’s why we use AI.

Ready to embrace a new era of task delegation?

HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.

Learn how to integrate AI into your own processes, allowing you to optimize your time and resources, while maximizing your output with ease.

Get the free guide here.

Today:

• Ilya's Vision for Future of Superintelligent AI 

• Musk Wanted For-Profit OpenAI Model

• Meta Opposes OpenAI’s For-Profit Shift

• xAI Launches Faster Grok Chatbot

• Meta Releases Invisible Watermark for Videos

• Pika 2.0 Adds Customization Features

Ilya Sutskever "Superintelligence is Self Aware, Unpredictable and Highly Agentic" | NeurIPS 2024

Ilya Sutskever left OpenAI to start Safe Super Intelligence (SSI), a new AI company that has attracted over a billion dollars in investment. In 2014, Ilya and colleagues at Google published a groundbreaking paper on teaching computers to translate languages using neural networks, which won a major award. 

Ten years later, Ilya reflected on AI advancements and highlighted challenges that might slow progress. He discussed the future of superintelligent AI, including ideas like self-awareness. Ilya emphasized the importance of large data and powerful networks but noted that future AI will need new approaches as data growth slows.

WATCH THE VIDEO ON YOUTUBE

Musk Wanted For-Profit OpenAI Model

Elon Musk initially supported OpenAI’s mission but pushed to make it a for-profit company in 2017. He demanded majority ownership and control, including being CEO, which OpenAI rejected to stay true to its goal of benefiting humanity. Negotiations failed, leading Elon to leave and start his own AI company, xAI, in 2023. OpenAI continued as a capped-profit organization, securing funding from partners like Microsoft. 

Despite Elon’s legal challenges, OpenAI remains committed to developing safe artificial intelligence for everyone, emphasizing innovation and ethical practices over individual control. The company aims to lead global AI advancements responsibly.

Meta Opposes OpenAI’s For-Profit Shift

Meta is urging California Attorney General Rob Bonta to prevent OpenAI from changing from a nonprofit to a for-profit company. In a letter, Meta claims OpenAI is misusing its nonprofit resources for large private profits. Meta supports Elon Musk’s opposition to this shift, despite their past disagreements. 

OpenAI, led by CEO Sam Altman, argues that becoming for-profit is essential to attract investors and sustain its successful ChatGPT product, which earns billions annually. OpenAI assures that the nonprofit will continue to exist and uphold its mission to ensure artificial intelligence benefits all people. Meta fears this move could negatively impact tech startups.

xAI Launches Faster Grok Chatbot

Elon Musk’s AI company, xAI, is launching an upgraded version of its Grok chatbot on X, the platform he owns. The new Grok 2 is three times faster and more accurate, supporting multiple languages better. Free users can ask Grok up to 10 questions every two hours, while premium subscribers enjoy higher limits. Additionally, xAI introduced a “Grok button” on X to help users find relevant information and engage in trending topics. 

The company also improved its API by making it more efficient and cheaper. Soon, xAI’s image AI, Aurora, will be available through the API, enhancing their offerings despite ongoing legal issues with OpenAI.

Meta Releases Invisible Watermark for Videos

Meta has introduced Video Seal, a new tool that embeds hidden marks in AI-generated videos to identify their source even after edits. These watermarks are invisible to viewers but can be detected later to verify where the video originated. Meta has made the entire Video Seal system available for free online, including the watermarking model, research paper, and all necessary code. 

They also released an interactive demo for users to test the technology. Additionally, Meta launched Omni Seal Bench, which compares different watermarking methods across various media, and Meta Watermark Anything, another free tool. These initiatives aim to ensure the authenticity of AI-created content.

Pika 2.0 Adds Customization Features

Pika, a startup from Palo Alto known for creating realistic AI video tools, has launched Pika 2.0. This new version offers users more control and customization when making AI-generated videos. Following the successful release of Pika 1.5, which gained over five million users and billions of video views, Pika 2.0 introduces features like “Scene Ingredients.” 

This allows users to add and adjust characters, objects, and settings in their videos easily. Aimed at individuals and small brands, Pika 2.0 is user-friendly and affordable, making it easier for everyone to create engaging social media content. Major brands like Balenciaga and Vogue are already using Pika’s tools.

🧠RESEARCH

InternLM-XComposer2.5-OmniLive: A Comprehensive Multimodal System for Long-term Streaming Video and Audio Interactions

InternLM-XComposer2.5-OmniLive, a system designed to handle long-term video and audio interactions. It addresses the challenge of continuous, simultaneous processing of perception, memory, and reasoning in AI, overcoming limitations of current models that struggle with real-time interaction and long-term data storage.

Phi-4 Technical Report

Phi-4, a 14-billion parameter language model that focuses on data quality, incorporating both synthetic and organic data during training. Unlike its predecessors, phi-4 outperforms its teacher model (GPT-4) in STEM-related question answering. It achieves strong performance on reasoning tasks despite minor architectural changes.

Euclid: Supercharging Multimodal LLMs with Synthetic High-Fidelity Visual Descriptions

Euclid, a family of multimodal large language models designed to improve low-level visual perception, particularly in describing geometric details. It highlights the use of high-fidelity synthetic data, multi-stage training, and a data curriculum to enhance performance on tasks like image-based geometry understanding, outperforming existing models in this area.

Multimodal Latent Language Modeling with Next-Token Diffusion

Latent Language Modeling (LatentLM), a unified approach to handle both discrete (text, code) and continuous (image, audio, video) data using causal Transformers. It incorporates a variational autoencoder and next-token diffusion for efficient multimodal generation and understanding. Experiments show LatentLM outperforms existing models in image generation, text-to-speech, and scalability.

AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials

AgentTrek, a method for synthesizing high-quality GUI agent trajectories using web tutorials. By automatically gathering tutorial texts, converting them into task goals, and simulating execution in a digital environment, AgentTrek generates training data that improves agent performance. This approach is cost-effective and scalable compared to manual annotation.

🛠️TOP TOOLS

PlayPhrase - Allows users to search for and play video clips containing specific phrases from movies and TV shows. 

Star By Face - AI-powered celebrity look-alike app that allows users to discover which famous personalities they resemble. 

AI Time Machine - Transform celebrities into historical figures, offering a humorous and imaginative glimpse into the past.

Ask Your PDF - AI-powered platform that revolutionizes document interaction, allowing users to engage in intelligent conversations with their PDF files, extract key insights, and manage information efficiently across multiple platforms.

NetworkAI - AI-powered networking tool developed by Wonsulting to streamline and enhance the job search process. 

📲SOCIAL MEDIA

🗞️MORE NEWS

• AI developer Tibor Blaho found hints in ChatGPT's code about a limited preview of GPT-4.5 for Teams subscribers. Despite rumors and CEO Sam Altman dismissing a December release, a big announcement is expected soon.

• OpenAI launched "Projects" in ChatGPT, allowing users to organize files and conversations into folders, similar to Google’s NotebookLM. Available to Plus, Pro, and Teams subscribers, with wider access expected in January.

• Meta's new AI architecture, Byte Latent Transformer (BLT), addresses the flaw in current language models by processing data at the byte level instead of tokens. This approach improves accuracy and efficiency, especially for complex tasks.

• OpenAI CEO Sam Altman will donate $1 million to President-elect Trump's inaugural fund, expressing support for Trump's leadership in advancing AI. This follows similar donations from Mark Zuckerberg and Jeff Bezos.

• Generative AI startup Liquid AI raised $250 million in early-stage funding, led by AMD. The company develops Liquid Foundation Models (LFMs), which are smaller and more efficient AI models tailored for enterprise use.

• Keyu Tian, a former ByteDance intern accused of sabotaging colleagues' work, was awarded the Best Paper Award at the NeurIPS conference for his research on generating images. Some have raised objections to the decision.

• Suchir Balaji, a former OpenAI employee who raised concerns about copyright issues at the company, was found dead at 26 in his San Francisco apartment. The cause of death was ruled a suicide, according to the medical examiner.