

Discover more from Last Week in AI
Last Week in AI #237: Salesforce's AI assistant takes flight ๐, California puts brakes on driverless trucks ๐, Adobe rewards AI trainers ๐ฐ, and more!
Salesforce introduces new AI assistant Einstein Copilot for all its CRM apps, California bill to ban driverless autonomous trucks goes to Newsomโs desk, Adobe starts paying bonuses to contributors
Top News
Salesforce introduces new AI assistant, Einstein Copilot, for all its CRM apps
Salesforce has announced an update to its AI suite, Einstein, at the Dreamforce 2023 conference. The update introduces Einstein Copilot, a generative AI conversational assistant integrated into Salesforce's customer relation management (CRM) apps. The assistant can automate tasks in sales, service, marketing, commerce, and development, be tailored to industry-specific needs, and customized by customer administrators to access and reference specific data. Einstein Copilot can also utilize third-party Language Learning Models (LLMs) like OpenAI's GPT-3.5. The assistant's operations are secured by Salesforce's Einstein Trust Layer, which screens every AI response and records all AI interactions for compliance and auditing purposes.
California bill to ban driverless autonomous trucks goes to Newsomโs desk
The California Senate has passed a bill, AB 316, that mandates the presence of a trained human safety operator in self-driving, heavy-duty vehicles on public roads, effectively banning driverless autonomous trucks. The bill, which still requires Governor Gavin Newsom's approval, has been opposed by autonomous vehicle (AV) companies and industry representatives who argue that it undermines the purpose of driverless technology and hampers technological advancement. Supporters of the bill argue that it ensures road safety and job security for truck drivers. The bill also requires the California Department of Motor Vehicles (DMV) to provide evidence of safety to policymakers and submit a report evaluating the performance of AV technology by January 1, 2029, or five years after testing begins. The DMV has opposed the bill, stating that it will not increase safety and will hinder the development of technology in California.
Adobe starts paying bonuses to Stock contributors whose content is being used to train Firefly
Adobe has announced that it will pay bonuses to Adobe Stock contributors whose content has been used to train its Firefly generative AI models. The bonus will be given to eligible contributors with photos, vectors, or illustrations in the standard and premium Stock collections, and will be paid out annually. The first bonus will cover the period from June 2, 2022, to June 3, 2023, which corresponds to the data set used to train Firefly. The bonus amount will be calculated based on a formula that considers the total number of assets a contributor has in the training dataset and the number of assets licensed, serving as a proxy for the commercial success and value of their assets. However, Adobe has not disclosed the total amount set aside for these bonuses or provided any range estimates.
Other News
Applications
Microsoft-backed self-driving car maker Wayve might finally have an answer to the driverless vehicle trust debate - London-based self-driving car maker Wayve has developed a first-of-its-kind technology called Lingo-1 that can explain to drivers how its AI is "thinking" and making driving decisions, aiming to address the lack of trust and transparency in autonomous vehicles.
How to access thousands of free audiobooks, thanks to Microsoft AI and Project Gutenberg - Project Gutenberg and Microsoft have collaborated to create thousands of free audiobooks using neural text-to-speech technology, allowing users to access high-quality audiobooks with natural-sounding voices and customizable narration.
eBay rolls out a tool that generates product listings from photos - eBay has introduced an AI tool that can generate product listings from a single photo, automatically writing a title and description, suggesting a category and subcategory, and providing information such as a product release date, list price, and shipping cost.
Text-to-Video AIs Runway Gen-2 and Pika Labs get new features - RunwayML and Pika Labs announce new features for their video AI systems, with RunwayML introducing camera control for AI-generated video and Pika Labs adding image animation and the ability to create videos at up to 24 frames per second.
Stability AI releases AI audio platform - Stability AI has launched Stable Audio, a text-to-audio generative AI platform that allows users to generate songs or background audio for various projects, offering more control over the length of the audio compared to previous models.
Business
Google.org to invest $20M into AI-focused grants for think tanks and academic institutions - Google.org is investing $20 million into AI-focused grants for think tanks and academic institutions as part of its Digital Futures Project, aiming to support researchers and public policy solutions around AI and address questions about fairness, bias, misinformation, security, and the future of work.
Defence AI startup Helsing breaks the record for European AI, raising a 223M Series B - Helsing, a defense AI startup backed by Spotify founder Daniel Ek, has raised โฌ209 million ($223 million) in a Series B funding round, making it potentially the largest European AI company and defense tech unicorn.
AI chip startup Enfabrica raises $125 mln, with backing from Nvidia - Enfabrica, a chip startup focused on networking chips for AI data centers, has raised $125 million in venture capital, with Nvidia as a strategic investor, to develop a network chip that improves the efficiency of Nvidia GPUs by allowing them to draw data from multiple sources simultaneously.
Nvidia's dominance in AI chips deters funding for startups - Nvidia's dominance in AI chips has deterred funding for startups, with the number of U.S. deals falling 80% from a year ago, as venture financiers are unwilling to provide big cash infusions due to the difficulty of breaking into the market.
EY launches AI platform and LLM after $1.4 billion investment - EY has launched EY.ai, a comprehensive platform to help clients boost AI adoption, with alliances with companies including Microsoft, Dell Technologies, IBM, SAP, ServiceNow, Thomson Reuters, and UiPath, and an investment of $1.4 billion as the foundation for the platform.
Databricks raises over $500 mln at $43 bln valuation - Databricks, a data analytics platform using AI, has raised over $500 million in a funding round, valuing the company at $43 billion, as it plans to build out foundation models in partnership with Nvidia.
Druid, a conversational AI platform for enterprises that integrates with ChatGPT, raises $30M - European AI startup Druid raises $30 million in funding to expand its conversational AI platform, which integrates with ChatGPT, and focuses on U.S. growth.
Alibaba opens AI model Tongyi Qianwen to the public - Alibaba has gained Chinese regulatory approval to mass-market its artificial intelligence model Tongyi Qianwen, and will open it to the public, with several organizations already reaching cooperation agreements to train their own large language models or develop language model applications based on it.
Adobeโs Firefly generative AI tools are now widely available - Adobe has announced that its Firefly generative AI model is now commercially available across Adobe Creative Cloud, Adobe Express, and Adobe Experience Cloud, allowing users to access Firefly-powered workflows and explore generative capabilities without subscribing to specific Adobe Creative Suite applications.
Anthropic and BCG form new alliance to deliver enterprise AI to clients - Anthropic has partnered with Boston Consulting Group (BCG) to provide BCG clients with direct access to Anthropic's AI technology, including its large language model assistant Claude 2, for applications such as synthesizing long-form documents, fraud detection, demand forecasting, and writing-related tasks.
AI and Machine Learning Usage up 15.5% Among Marketing Teams, Snowflake Modern Marketing Data Stack Report Finds - AI and machine learning usage among marketing teams has increased by 15.5% as marketers explore the potential impact of large language models and generative AI on their work, according to a report by Snowflake.
AI expert is a hot new position in the freelance jobs market - The rise of generative AI is rapidly reshaping the freelance tech job market, with AI-related job posts and searches surging across platforms like LinkedIn, Upwork, and Fiverr, as businesses seek AI experts to implement technologies like ChatGPT and AI app development.
Character.ai Surpasses ChatGPT In Mobile App Usage In The US - Character.ai, an AI app that allows users to design their own AI characters, is gaining ground on ChatGPT in terms of mobile app usage, with 4.2 million monthly active users in the US, appealing to a younger demographic and showing potential for further expansion.
Founders of shuttered Argo AI launch autonomous trucking startup - The founders of Argo AI, a self-driving car startup abandoned by Ford and Volkswagen, are launching an autonomous trucking business with $1 billion in backing from SoftBank, aiming to revolutionize the trucking and freight industries.
Concerns
Spies, scientists, defense officials and tech founders canโt agree on how to keep AI under control: โWeโre running at full speed toward a cliffโ - Spies, scientists, defense officials, and tech founders are divided on how to regulate AI, with some arguing for immediate action to prevent harm and others cautioning against stifling innovation.
More writers sue OpenAI for copyright infringement over AI training - A group of U.S. authors, including Pulitzer Prize winner Michael Chabon, has sued OpenAI for copyright infringement, accusing the Microsoft-backed program of misusing their writing to train its AI-powered chatbot ChatGPT.
AI can help screen for cancerโbut thereโs a catch - AI-assisted cancer screening has the potential to lead to more overdiagnosis, but there is also a chance that AI could help address the problem by examining the trajectories of different patients' cancers over time.
Microsoft Publishes Garbled AI Article Calling Tragically Deceased NBA Player "Useless" - Microsoft's MSN news portal published a garbled, seemingly AI-generated article that derided former NBA player Brandon Hunter as "useless" in its headline, sparking condemnation and raising concerns about the future of AI-generated content in journalism.
Thanks to A.I., workers are struggling with โFOBOโโfear of being obsolete - Workers in the U.S. are increasingly worried about the threat of AI technology rendering their jobs obsolete, with nearly one-quarter expressing concern, according to a recent Gallup survey, although experts say the fear is overblown for most workers.
As his Kickstarter passes $1.3 million, publisher defends Terraforming Marsโ generative AI art: โItโs too powerful a technologyโ - Indie Game Studios' imprint Stronghold Games defends its use of generative AI in the production of expansions for the popular board game Terraforming Mars, despite controversy surrounding the technology's ethical implications and impact on artists.
Research
Microsoft open sources EvoDiff, a novel protein-generating AI - Microsoft has open sourced EvoDiff, a general-purpose framework that can generate "high-fidelity," "diverse" proteins given a protein sequence, potentially revolutionizing protein engineering and drug development.
Physically Grounded Vision-Language Models for Robotic Manipulation - Physically grounded vision-language models (VLMs) are proposed to improve the understanding of physical object concepts, which enhances the performance of robotic manipulation tasks that involve interaction and physical reasoning about common household objects.
AI model speeds up high-resolution computer vision - Researchers from MIT and the MIT-IBM Watson AI Lab have developed an efficient computer vision model called EfficientViT that can perform semantic segmentation accurately in real-time on devices with limited hardware resources, such as autonomous vehicles, making it up to nine times faster than previous models while maintaining the same or better accuracy.
Medusa: Simple Framework for Accelerating LLM Generation with Multiple Decoding Heads - Medusa is a framework that accelerates text generation with large language models by integrating additional decoding heads into the original model, offering a 2x to 3x boost in speed and addressing the challenges of speculative decoding.
Efficient Memory Management for Large Language Model Serving with PagedAttention - Efficient memory management for large language model serving is achieved through the use of PagedAttention, an attention algorithm inspired by virtual memory and paging techniques, resulting in improved throughput and reduced memory usage compared to existing systems.
PromptTTS 2: Describing and Generating Voices with Text Prompt - PromptTTS 2 is a new approach to text-to-speech that addresses the challenges of voice variability and limited availability of text prompt datasets, using a variation network and prompt generation pipeline to generate voices consistent with text prompts and offer users more choices on voice generation.
Can Large Language Models Reason? - Large language models (LLMs) have exhibited behaviors that resemble reasoning, but there is a debate over whether these behaviors are driven by true abstract reasoning abilities or by memorizing patterns in training data, which has implications for their generalizability and trustworthiness.
InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation - A new method called Rectified Flow has been used to create InstaFlow, the first one-step diffusion-based text-to-image generator with high-quality results, surpassing previous techniques and achieving an FID of 23.3 on MS COCO 2017-5k.
Machine Learning Uncovers Neural Pathways of Narcissistic Traits - Researchers have used advanced machine learning techniques to uncover the neural structure associated with narcissism, predicting narcissistic personality traits based on brain organization and other personality aspects, and identifying a specific brain circuit that plays a predictive role.
MADLAD-400: A Multilingual And Document-Level Large Audited Dataset - A new dataset called MADLAD-400 has been introduced, which is a multilingual and document-level large audited dataset for AI applications.
FLM-101B: An Open LLM and How to Train It with $100K Budget - A paper discusses the development of a cost-effective large language model (LLM) with 101B parameters and 0.31TB tokens, trained on a 100K budget, and introduces a systematic evaluation paradigm for IQ evaluation of LLMs.
Introducing Wรผrstchen : Fast Diffusion for Image Generation - Wรผrstchen is a fast and efficient diffusion model for image generation that works in a highly compressed latent space, reducing computational costs for training and inference by orders of magnitude.
When Less is More: Investigating Data Pruning for Pretraining LLMs at Scale - Investigating the use of data pruning techniques for pretraining large language models (LLMs) at scale, this work finds that a simple technique of using perplexity as a data quality estimator outperforms more computationally expensive scoring methods, suggesting that the majority of pretraining data can be removed while retaining performance.
MADLAD-400: A Multilingual And Document-Level Large Audited Dataset - A new dataset called MADLAD-400 has been introduced, which is a multilingual and document-level large audited dataset for AI applications.
Analysis
Why AI Can't Make Human Creativity Obsolete - Humans continue to enjoy watching and participating in activities like chess, writing, and art, despite the advancements of AI, because they are drawn to the imperfections and human qualities that AI lacks.
Policy
America leads the world in AIโbut we could fall behind on AI regulation by the end of 2023 - The U.S. is at risk of falling behind in AI regulation as other countries, such as the U.K. and the European Union, are moving quickly to shape future rules, with the EU finalizing its AI Act by the end of 2023 and Japan leading a G7 effort to establish common standards for AI governance.
2 Senators Propose Bipartisan Framework for A.I. Laws - Senators Richard Blumenthal and Josh Hawley are proposing a comprehensive framework to regulate artificial intelligence, including requirements for licensing and auditing, the creation of an independent federal office, liability for privacy and civil rights violations, and data transparency and safety standards.
8 More Companies Pledge to Make A.I. Safe, White House Says - Eight more companies, including Adobe, IBM, Palantir, Nvidia, and Salesforce, have pledged to follow safety, security, and trust standards in artificial intelligence, joining the industry-led effort initiated by Amazon, Google, Microsoft, and others, as lawmakers and regulators grapple with the implications of AI technology.
Microsoft president and Nvidia chief scientist to testify in Senate AI hearings - Microsoft president and Nvidia chief scientist are testifying in Senate AI hearings, discussing the need for regulation and liability in the development and use of AI technology.
Self-publishers must declare if content sold on Amazonโs site is AI-generated - Amazon has introduced new rules for self-published Kindle books, requiring authors to declare if the content is AI-generated, in response to concerns about the proliferation of AI-generated books on the platform.
Exclusive: California Bill Proposes Regulating AI at State Level - A senior California lawmaker is introducing a bill to regulate AI at the state level, proposing transparency requirements, legal liability for unintended consequences, and security measures to prevent misuse of advanced AI systems.
EU to let โresponsibleโ AI startups train models on its supercomputers - The European Union plans to allow AI startups to use its high-performance computing supercomputers to train AI models, but only if they comply with the bloc's AI governance program.
Gary Gensler confirms SECโs use of AI for financial surveillance - SEC Chair Gary Gensler confirms the agency's use of AI for financial surveillance, stating that they are already using it in market surveillance and enforcement actions to look for patterns in the market.
California lawmakers want to protect actors from being replaced by artificial intelligence - California lawmakers have introduced a bill that would protect actors and artists from being replaced by their digital clones, as concerns grow about the use of artificial intelligence in the entertainment industry.
Inside Elon Musk's Struggle for the Future of AI - Elon Musk's concern about the dangers of artificial intelligence led him to invest in DeepMind, co-found OpenAI, and start his own AI company called xAI, with the goal of making AI safe and preserving human consciousness.
UN calls for age limits for AI tools in schools - The United Nations has called for age limits on the use of AI tools in schools, warning that relying on such programs instead of human teachers could harm a child's emotional well-being and leave them vulnerable to manipulation.
Rishi Sunak considers banning Chinese officials from half of AI summit - Rishi Sunak is considering banning Chinese officials from half of his artificial intelligence safety summit in November due to concerns over widespread spying by Beijing on western governments.
Expert Opinions
LinkedIn co-founder Reid Hoffman dismisses calls to halt AI development as โfoolishโ and โanti-humanistโ - LinkedIn co-founder Reid Hoffman dismisses calls to halt AI development as โfoolishโ and โanti-humanist,โ arguing that the pace of development should be accelerated to help solve societal problems.
Copyright ยฉ 2023 Skynet Today, All rights reserved.