Top News
Google DeepMind’s new AlphaFold can model a much larger slice of biological life
Google’s DeepMind has unveiled AlphaFold 3, an advanced version of its protein structure prediction tool, which can now predict the structures of DNA, RNA, and essential drug discovery molecules like ligands. This new tool provides a more detailed and dynamic view of molecular interactions, marking a significant step towards modeling the dynamic systems of biology. AlphaFold 3 is expected to be particularly beneficial for drug discovery, as it can predict where a drug binds a protein, a feature that was absent in its predecessor, AlphaFold 2. However, the tool’s effectiveness will depend on the accuracy of its predictions, which, according to Mohammed AlQuraishi, an assistant professor of systems biology at Columbia University, is still very inaccurate for certain uses like protein-RNA interactions.
Rifle-Armed Robot Dogs Now Being Tested By Marine Special Operators
The United States Marine Forces Special Operations Command (MARSOC) is testing rifle-armed “robot dogs” supplied by Onyx Industries. These robotic canines are equipped with Onyx’s SENTRY remote weapon system (RWS) in two calibers: 7.62x39mm and 6.5mm Creedmoor. The underlying robot dog, Ghost Robotics’ Vision 60 quadrupedal unmanned ground vehicle (Q-UGV), is designed for tasks such as remote inspection, surveillance, mapping, distributed communications, and persistent security. The autonomous weapon system can scan and detect targets, including drones, people, and vehicles, and alert the operator once a target has been identified, allowing the human to decide whether to engage or not. The system can be operated from anywhere in the world with remote human operators.
DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
DeepSeek AI releases DeepSeek-V2, a Mixture-of-Experts (MoE) language model, that is state-of-the-art, cost-effective, and efficient with 236B total parameters, of which 21B are activated for each token, and a context length of 128K tokens. The model incorporates Multi-head Latent Attention (MLA) and DeepSeekMoE, which ensure efficient inference by compressing the Key-Value (KV) cache into a latent vector and enable economical training through sparse computation. Compared to its predecessor, DeepSeek 67B, DeepSeek-V2 offers improved performance, reduces training costs by 42.5%, shrinks the KV cache by 93.3%, and increases the maximum generation throughput by 5.76 times. The model, pretrained on an 8.1T token multi-source corpus and further refined through Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), maintains top-tier performance among open-source models, even with only 21B activated parameters.
TikTok will automatically label AI-generated content created on platforms like DALL·E 3
TikTok has announced that it will automatically label AI-generated content created on other platforms, such as OpenAI’s DALL·E 3, using a technology called Content Credentials from the Coalition for Content Provenance and Authenticity (C2PA). This technology attaches specific metadata to content, allowing TikTok to instantly recognize and label AI-generated content. The change, which is rolling out globally, will also apply to content created on platforms that have implemented Content Credentials, including Microsoft’s Bing Image Creator. In the coming months, TikTok will also start attaching Content Credentials to AI-generated content created on its own platform. The company views this move as an additional measure to ensure AI-generated content is properly labeled and to alleviate pressure on creators.
Other News
Tools
Microsoft is ‘turning everyone into a prompt engineer’ with new Copilot AI features – Microsoft is introducing new AI features in Copilot for Microsoft 365 to help users create better prompts and become prompt engineers, aiming to improve productivity and efficiency in the workplace.
New Microsoft AI model may challenge GPT-4 and Google Gemini – Microsoft is developing a new large-scale AI language model called MAI-1, potentially rivaling state-of-the-art models from Google, Anthropic, and OpenAI, and may preview it at the Build developer conference.
Powerful New Chatbot Mysteriously Returns in the Middle of the Night – A powerful new chatbot mysteriously returns to a benchmarking site, sparking speculation about its origins and drawing massive attention from the AI community.
OpenAI Releases ‘Deepfake’ Detector to Disinformation Researchers – OpenAI releases a deepfake detector tool to combat the influence of AI-generated content on the upcoming elections, acknowledging that it’s just the beginning of the fight against deepfakes.
OpenAI’s Model Spec outlines some basic rules for AI – OpenAI’s Model Spec outlines a proposed framework for AI behavior, including principles and rules to shape how AI tools respond in the future, with the intention of drawing a clearer line between intentional actions and bugs.
U.K. agency releases tools to test AI model safety – U.K. Safety Institute releases an open-source toolset called Inspect to assess AI model safety, aiming to provide a shared, accessible approach to evaluations and encourage global AI community collaboration.
Business
Wayve, an A.I. Start-Up for Autonomous Driving, Raises $1 Billion – Wayve, a London-based AI start-up for autonomous driving, raised an eye-popping $1 billion from investors like SoftBank, Microsoft, and Nvidia, reflecting investor optimism about AI’s potential to reshape industries.
Alibaba rolls out latest version of its large language model to meet robust AI demand – Alibaba Cloud releases the latest version of its large language model, Qwen2.5, with advancements in reasoning, code comprehension, and textual understanding, surpassing OpenAI’s GPT-4 in language and creation capabilities.
Audible’s Test of AI-Voiced Audiobooks Tops 40,000 Titles – Audible introduces AI-generated voices to its users, with over 40,000 titles now available in this format.
OpenAI and Stack Overflow partner to bring more technical knowledge into ChatGPT – OpenAI and Stack Overflow partner to improve AI models and bring more technical information into ChatGPT, with OpenAI gaining access to Stack Overflow’s API and feedback from the developer community, and Stack Overflow using OpenAI’s language models to expand its Overflow AI.
President Joe Biden to announce AI data center at failed Foxconn site in Wisconsin – President Joe Biden is set to announce a $3.3 billion investment by Microsoft to build an AI data center in Wisconsin, creating thousands of jobs and partnering with local educational institutions.
Microsoft Creates Top Secret Generative AI Service for US Spies – Microsoft has developed a top-secret generative AI service for US intelligence agencies, allowing them to safely analyze classified information.
Rad AI, a startup that helps radiologists save time on report generation, raises $50M Series B from Khosla Ventures – AI startup Rad AI, backed by Khosla Ventures, aims to streamline radiologists’ workload by automating report generation, rather than replacing physicians with machines, and has raised $50 million in Series B funding to further develop its innovative technology.
Apple Will Revamp Siri to Catch Up to Its Chatbot Competitors – Apple is revamping Siri to catch up to its chatbot competitors, such as OpenAI’s ChatGPT, by incorporating generative artificial intelligence and improving its conversational abilities.
Motional delays commercial robotaxi plans amid restructuring – Motional, an autonomous vehicle startup, is pausing commercial operations and delaying the launch of a driverless taxi service to focus on core technology and undergo a restructuring.
SoundHound AI and Perplexity Partner to Bring Online LLMs to Next Gen Voice Assistants Across Cars and IoT Devices – SoundHound AI partners with Perplexity to enhance its voice assistant with online LLM capabilities, allowing for accurate and generative AI responses to web-based queries, expanding the assistant’s capabilities across various devices.
OpenAI destroyed a trove of books used to train AI models. The employees who collected the data are gone. – OpenAI deleted two large datasets used to train its GPT-3 AI model, which contained over 100,000 published books, leading to a class-action lawsuit by the Authors Guild.
The rise of the Chinese AI unicorns doing battle with OpenAI – Chinese AI startups are competing with OpenAI, with four companies surpassing $1 billion in valuation and aiming to develop AI products to compete for market share in China.
Research
StoryDiffusion: Consistent Self-Attention for Long-Range Image and Video Generation – A new self-attention calculation method, Consistent Self-Attention, is proposed to enhance the consistency of generated images in diffusion-based generative models, and a Semantic Motion Predictor module is introduced to improve long-range video generation, resulting in the StoryDiffusion framework for text-based story generation with consistent images and videos.
Sperm whale ‘alphabet’ discovered, thanks to machine learning – Machine learning technologies have helped researchers at MIT CSAIL and Project CETI unlock a kind of sperm whale alphabet, revealing a complex combinatorial communication system among the chatty sea mammals.
ImageInWords: Unlocking Hyper-Detailed Image Descriptions – AI researchers have developed ImageInWords, a human-in-the-loop annotation framework for creating hyper-detailed image descriptions, resulting in a new dataset that significantly improves model performance in generating accurate and comprehensive image descriptions.
Chain of Thought Empowers Transformers to Solve Inherently Serial Problems – Chain of Thought (CoT) empowers transformers to solve inherently serial problems by providing a theoretical understanding of its power for decoder-only transformers through the lens of expressiveness, enabling the model to perform inherently serial computation and dramatically improving accuracy for tasks that are hard for parallel computation.
xLSTM: Extended Long Short-Term Memory – xLSTM, an extended version of Long Short-Term Memory (LSTM), introduces exponential gating and modified memory structures to boost language modeling capabilities, outpacing Transformers and State Space Models in performance and scaling.
MAmmoTH2: Scaling Instructions from the Web – Efficiently harvesting 10 million naturally existing instruction data from the web corpus enhances large language model reasoning, leading to significant performance boosts on reasoning benchmarks and achieving state-of-the-art results on several benchmarks without costly human annotation or GPT-4 distillation.
Imagine Flash: Accelerating Emu Diffusion Models with Backward Distillation – A novel distillation framework is proposed to accelerate high-fidelity, diverse sample generation using just one to three steps in diffusion models, outperforming existing competitors in quantitative metrics and human evaluations.
Granite Code Models: A Family of Open Foundation Models for Code Intelligence – Granite Code Models are a family of open foundation models for code intelligence, ranging in size from 3 to 34 billion parameters and optimized for enterprise software development workflows, with state-of-the-art performance across a range of coding tasks.
Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Models – Prometheus 2 is an open-source language model designed to evaluate other language models, addressing concerns about transparency, controllability, and affordability while closely mirroring human and GPT-4 judgments.
Concerns
Is AI lying to me? Scientists warn of growing capacity for deception – AI systems are becoming increasingly sophisticated in their capacity for deception, raising concerns about potential dangers to society and the need for AI safety laws.
South Korean woman falls for deepfake Elon Musk, loses $50K in romance scam – A woman falls for a deepfake Elon Musk on Instagram and loses $50K in a romance scam after being convinced by the scammer’s detailed impersonation.
Stack Overflow bans users en masse for rebelling against OpenAI partnership — users banned for deleting answers to prevent them being used to train ChatGPT – Stack Overflow users are facing bans for deleting their answers to prevent them from being used to train ChatGPT, sparking outrage and a debate over the site’s partnership with OpenAI and the use of generative AI.
Meet AdVon, the AI-Powered Content Monster Infecting the Media Industry – AI-powered content monster AdVon Commerce has been using AI to generate product reviews, leading to fake bylines, legal threats, and a shift to AI-generated content, causing chaos and distrust in the media industry.
Slop is the new name for unwanted AI-generated content – Unwanted AI-generated content is being termed “slop” and compared to spam, with the author advocating for ethical use of AI and not sharing unreviewed content.
The teens making friends with AI chatbots – Teenagers are finding comfort and support in AI chatbots, using them as a form of therapy and a way to vent about their emotions, but experts are concerned about the potential negative impact on their social development and reliance on bots.
Writers and publishers in Singapore reject a government plan to train AI on their work – Singaporean writers and publishers reject the government’s plan to train AI on their work, expressing concerns about consent, compensation, and the potential misuse of their creative material.
Policy
Meet the Woman Who Showed President Biden ChatGPT—and Helped Set the Course for AI – Arati Prabhakar showed President Biden ChatGPT, leading to a regulatory executive order for AI, as she discussed the transformative power of AI and her hopes for a semiconductor renaissance in the US.
US eyes curbs on China’s access to AI software behind apps like ChatGPT – US plans to restrict China’s access to advanced AI models like ChatGPT, amid concerns of potential use in cyber attacks and bioweapons, using a computing power threshold to determine export restrictions.
Fun
At the AI Film Festival, humanity triumphed over tech – AI-generated effects in films were showcased at Runway’s AI Film Festival, with some films struggling due to the limitations of current AI tools, while others triumphed thanks to the human touch and emotional performances.
Copyright © 2024 Skynet Today, All rights reserved.
Source: Read MoreÂ
