Last Week in AI #288 - new AI video APIs, Hollywood studio embraces AI, world model for robots
AI video rivalry intensifies as Luma announces Dream Machine API hours after Runway, Lionsgate Signs Deal With AI Company Runway,1X Unveils AI World Model for Robot Training, and more!
Top News
AI video rivalry intensifies as Luma announces Dream Machine API hours after Runway
Text-to-video startup Luma AI has announced an API for its Dream Machine video generation model which allows users — including individual software developers, startup founders, and engineers at larger enterprises — to build applications and services using Luma's video generation model. The API is powered by the latest version of Dream Machine (v1.6) and offers advanced video generation tools such as text-to-video, image-to-video, keyframe control, video extension and looping, camera motion control, and variable aspect ratios. The API is priced at $0.32 per million pixels generated, making it accessible to smaller developers. Luma AI also introduced a "Scale" option for larger companies and organizations, providing higher rate limits and personalized onboarding and engineering support.
Just hours prior to this, Runway, an AI startup that is also focused on video creation, had launched its own API that allows developers to integrate its generative models into third-party platforms, currently offering its Gen-3 Alpha Turbo model with two pricing plans. There are two versions of its API for smaller teams and large enterprises, and both via require filling out a Google Form to get on waitlists. In contrast, Dream Machine’s API is available to begin using now.
Lionsgate Signs Deal With AI Company Runway, Hopes That AI Can Eliminate Storyboard Artists and VFX Crews
Lionsgate, a major Hollywood studio, has announced a pioneering partnership with Runway to develop an AI model that can generate "cinematic video" and potentially replace storyboard artists and VFX crews. The model will be trained on Lionsgate's extensive portfolio of film and TV content, with the aim of saving "millions and millions of dollars" in production costs. Lionsgate's vice chairman, Michael Burns, envisions the AI tool being used to create backgrounds and special effects, and believes it will offer "cutting-edge, capital-efficient content creation opportunities." However, the move has sparked controversy, with Runway currently facing a lawsuit from a group of visual artists for copyright infringement.
Norwegian Startup 1X Unveils AI World Model for Robot Training
Norwegian startup 1X Technologies has developed an AI-based world model to serve as a virtual simulator for training robots, addressing the challenge of reliably evaluating multi-task robots in dynamic environments. The company used machine learning to train these models on thousands of hours of video footage, enabling the models to predict how objects and environments would respond to different robotic actions and generate realistic visual simulations. Despite some limitations, such as occasional inconsistencies in object coherence and unrealistic object behavior, the models can handle complex interactions and generate multiple potential outcomes from the same initial conditions. To encourage further research and innovation, 1X Technologies has launched the "1X World Model Challenge," offering data, pre-trained models, and cash prizes.
Other News
Tools
Copilot Wave 2 supercharges productivity with AI across all your Microsoft 365 apps - Microsoft's Copilot AI chatbot, now in its 'Wave 2' phase, enhances productivity in Microsoft 365 apps by enabling collaborative document creation, narrative building in PowerPoint, data analysis in Excel using Python, intelligent email summarization in Outlook, and the introduction of AI assistants called Copilot Agents.
China's Alibaba launches over 100 new open-source AI models, releases text-to-video generation tool - The Hangzhou-headquartered firm is looking to increase competition with domestic rivals such as Baidu and Huawei, as well as U.S. titans like Microsoft and OpenAI.
Adobe Introduces Tools to Measure Impact of AI-Generated Content - These new tools will help brands show the return on investment (ROI) of the image and copy generation tools.
Amazon introduces Amelia, an AI assistant for third-party sellers - Amazon introduces Amelia, an AI assistant for third-party sellers, designed to quickly resolve account issues and fetch sales and inventory data, using generative AI and retrieval-augmented generation.
Amazon releases a video generator — but only for ads - Amazon has launched an AI-powered video generator for advertisers, creating custom product showcase videos from a single image, with plans for further development and expansion.
Snap is introducing an AI video generation tool for creators - Snap introduces AI video generation tool for creators, allowing them to generate AI videos from text and image prompts, with plans to make it available to a small subset of creators in beta.
Kling AI launches new 1.5 model along with Motion Brush feature - Kling AI releases version 1.5 with updates including 1080p HD video generation, motion brush feature, high-quality mode, improved image quality, and better prompt relevance.
Plaud’s $169 ChatGPT-powered NotePin has a permanent place in my travel bag - A $169 ChatGPT-powered NotePin by Plaud.AI offers a compact and efficient solution for transcribing conversations and taking notes, addressing the challenges of traditional methods and providing a deliberate design for intentional recording.
Mistral AI Released Mistral-Small-Instruct-2409 - Mistral AI has released Mistral-Small-Instruct-2409, a new open-source large language model designed to enhance AI performance, accessibility, and natural language processing tasks, while promoting transparency and collaboration in the AI community.
Gemini Live is rolling out to all Android users - for free - Google's Gemini Live, a conversational AI-powered assistant, is now available for free to all Android users, allowing them to have free-flowing conversations with the assistant whenever they'd like.
Business
Cruise robotaxis return to the Bay Area nearly one year after pedestrian crash - Cruise is resuming testing of its autonomous vehicles in the Bay Area after a pedestrian crash, with plans to progress to supervised AV testing later this fall.
Microsoft’s Hypocrisy on AI - Microsoft's AI technology is being marketed to fossil-fuel companies to maximize production and find new reserves, while the company publicly commits to reducing emissions and fighting climate change, leading to internal conflict and criticism from employees.
Chip Startup Groq Backs Saudi AI Ambitions With Aramco Deal - AI startup Groq Inc. partners with Aramco to build a massive data center in Saudi Arabia, aiming to establish a regional hub for AI systems in the Middle East, Africa, and India.
Black Forest Labs is raising $100M at a $1B valuation, say sources - AI startup Black Forest Labs, known for its generative image models and high-profile customers, is reportedly raising $100 million at a $1 billion valuation, attracting attention from investors and facing pressure to deliver on its promising technology.
11xAI raises $24M led by Benchmark to build AI digital employees - AI startup 11x.ai raises $24 million to build AI digital employees for process automation, with plans to expand its team and develop new AI agents.
Campfire secures $3.95 million for AI-native gaming - Campfire secures $3.95 million for AI-native gaming and launches Cozy Friends as a showcase for AI-native games, allowing developers to create AI characters that can hold conversations with users and accompany them on online adventures.
Sam Altman departs OpenAI’s safety committee - Sam Altman leaves OpenAI's internal safety committee, which will now be an independent board oversight group chaired by Zico Kolter, amid concerns about the company's policies and profit incentives.
Google outlines plans to help you sort real images from fake - Google plans to use C2PA technology to identify if images were taken with a camera, edited by software, or produced by generative AI models, and will integrate this into search results, ad systems, and potentially YouTube.
With AI, dead celebrities are working again – and making millions - AI is being used to bring back the voices and images of deceased celebrities, allowing their estates to continue earning revenue from their intellectual property.
TikTok owner ByteDance taps TSMC to make its own AI GPUs to stop relying on Nvidia — the company has reportedly spent over $2 billion on Nvidia AI GPUs - ByteDance, the parent company of TikTok, is developing its own AI GPUs to reduce reliance on Nvidia and is expected to enter mass production by 2026, aiming to address the shortage and high cost of Nvidia GPUs.
OpenAI to Decide Which Backers to Let Into $6.5 Billion Funding - OpenAI is nearing completion of its $6.5 billion fundraising, and prospective investors will soon find out if they will be part of the deal.
Research
EzAudio: Enhancing Text-to-Audio Generation with Efficient Diffusion Transformer - EzAudio introduces a transformer-based T2A diffusion model that enhances convergence speed, training stability, and memory usage, while delivering realistic listening experiences and maintaining a streamlined model structure and low training costs.
AI tool cuts unexpected deaths in hospital by 26%, Canadian study finds - AI-based early warning system at St. Michael's Hospital in Toronto, called Chartwatch, has led to a 26% decrease in unexpected deaths among hospitalized patients, showing promising results for AI technology in healthcare.
Denoising diffusion models for high-resolution microscopy image restoration - AI models are being developed to denoise high-resolution microscopy images, improving image restoration for better analysis and interpretation.
Takin: A Cohort of Superior Quality Zero-shot Speech Generation Models - A new series of zero-shot speech generation models, Takin AudioLLM, including Takin TTS, Takin VC, and Takin Morphing, are introduced, capable of producing high-quality, customizable speech for audiobook production.
AI model can reveal the structures of crystalline materials - AI model Crystalyze can determine the structures of powdered crystalline materials, aiding in the characterization of materials for various applications such as batteries and magnets.
To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning - Chain-of-thought (CoT) is beneficial for tasks involving math and logic, providing strong performance benefits primarily in these areas, but yielding smaller gains on other types of tasks.
A Controlled Study on Long Context Extension and Generalization in LLMs - Study finds that long-context language models require consistent base models and extension data, reaffirming the critical role of perplexity as a performance indicator and highlighting challenges with approximate attention methods.
Concerns
A.I. Pioneers Call for Protections Against ‘Catastrophic Risks’ - A.I. pioneers warn of catastrophic risks posed by fast-developing technology and call for a global system of oversight to address the potential harm.
AI could soon be beyond our control—and the scientists who created it are worried - Leading AI scientists are urging world governments to collaborate on regulating AI technology to prevent catastrophic outcomes, proposing measures such as creating government AI safety bodies, requiring developer AI safety pledges, funding independent research and tech checks on AI, and establishing ethical norms for engineers.
James Earl Jones' controversial AI decision will let Darth Vader live on, but it raises concerns among actors - James Earl Jones' decision to use AI to preserve his voice as Darth Vader raises concerns among actors about the potential impact on their work and the need for consent and compensation transparency.
The widespread scam half of us don’t even know is possible - AI technology is being used to replicate people's voices in order to scam their relatives for money, and many people are unaware of this possibility, making them more likely to fall victim to these convincing scams.
Project Analyzing Human Language Usage Shuts Down Because ‘Generative AI Has Polluted the Data’ - Generative AI spam has led to the shutdown of a project analyzing human language usage.
Policy
White House Launches AI Data Center Task Force with Industry Experts - White House launches AI data center task force with industry experts to address massive infrastructure needs for artificial intelligence projects, aiming to ensure US leadership in the field and drive electricity demand up by 15% to 20% over the next decade.
Governor Newsom signs bills to combat deepfake election content - Governor Newsom signs bills to combat deepfake election content, including legislation to protect the digital likeness of actors and performers and require disclosure of AI-generated or altered content in political advertisements.
Governor Newsom signs bills to protect digital likeness of performers - California passes bill to protect performers' digital likeness from unauthorized use by AI, ensuring their professional representation in negotiating contracts.
Explainers
Understanding 1.58-bit Large Language Models - The article explains the concept of 1.58-bit large language models, including the advantages of ternary quantization, the meaning of 1.58 bits, practical implementations, the BitNet b1.58 model, and its performance compared to full-precision models.