

Discover more from Last Week in AI
Last Week in AI #224: EU's landmark AI regulation, AI used to voice John Lennon in last Beatles record, Meta's new voice synthesis AI, and more!
EU Parliament approves AI Act, AI voice of John Lennon used in the film Get Back, Meta unveils AI model that can perform many voice tasks like translation, style transfer, and noise removal
Top News
Europe approves landmark AI legislation, challenging tech giants’ power
The European Parliament has approved the EU AI Act, a comprehensive package of regulations aimed at protecting consumers from potentially harmful applications of artificial intelligence (AI). The legislation takes a risk-based approach, introducing restrictions on AI applications based on their potential danger. It would ban unacceptable tools, such as systems allowing law enforcement to predict criminal behavior, and introduce new limits on high-risk technologies, such as recommendation algorithms and AI-generated content. The legislation also requires companies to publish summaries of copyrighted data used to train their AI tools. The EU's aggressive posture towards tech giants puts it on a collision course with American companies, with OpenAI warning that it may be forced to pull out of Europe.
Key Takeaway: The EU AI Act is a landmark legislation that aims to protect consumers from the potential dangers of AI. By taking a risk-based approach, the legislation introduces restrictions on AI applications based on their potential harm. This puts the EU on a collision course with American tech giants, who have been funneling billions of dollars into AI. OpenAI has warned that it may be forced to pull out of Europe if the legislation is too restrictive. The EU's aggressive posture towards tech giants puts it in a position of global leadership on tech regulation, as other governments, including the US Congress, are just beginning to grapple with the threat presented by AI. The EU's regulations are likely to influence policymakers around the world and set standards that could trickle down to all consumers.
Paul McCartney says AI tools helped rescue John Lennon vocals for ‘last Beatles record’
Paul McCartney discusses the use of artificial intelligence (AI) tools to rescue John Lennon's vocals for what will be the last Beatles record. McCartney mentions that AI was used in the film Get Back to separate John's voice from a cassette recording. For the upcoming record, they were able to use AI to get John's voice pure and mix the record as usual.
Our Take: The use of AI tools to rescue and enhance John Lennon's vocals for the last Beatles record raises interesting questions about the intersection of technology and art. While it is impressive that AI can separate and enhance specific elements of a recording, it also raises concerns about the authenticity and integrity of the final product. Can AI truly capture the essence of an artist's voice and intention? Will future generations be able to distinguish between genuine recordings and those manipulated by AI? As technology continues to advance, artists and listeners alike will need to grapple with these ethical and artistic considerations.
Meta announces Voicebox, a generative model for multiple voice synthesis tasks
Meta Platforms' AI research arm has unveiled Voicebox, a generative model that can generate speech from text. What sets Voicebox apart is its ability to perform tasks such as editing, noise removal, and style transfer, even without being specifically trained for them. The model has been trained on a general task of mapping voice audio samples to their transcripts and can synthesize speech in six languages. It uses Meta's "Flow Matching" technique, which allows it to learn from varied speech data without manual labeling. Voicebox has potential applications in speech customization, voice sampling, and synthetic data generation for training speech processing models.
Our Take: Voicebox is an impressive generative model that showcases the potential of AI in speech synthesis. Its ability to perform tasks beyond its training is a significant advancement, enabling applications such as speech customization and voice sampling. However, the ethical concerns raised by Meta about the misuse of AI-generated content are valid. As AI technology becomes more powerful, there is a need for responsible development and deployment to prevent malicious use. Striking a balance between innovation and safeguarding against potential harm is crucial. The development of classifier models to detect speech and audio generated by Voicebox is a step in the right direction, but ongoing efforts to address limitations and mitigate risks are necessary.
Other News
Research
I-JEPA: The first AI model based on Yann LeCun’s vision for more human-like AI - "Researchers have developed the I-JEPA model, based on Yann LeCun's vision, which aims to capture common sense background knowledge about the world through self-supervised learning from unlabeled data, avoiding biases and limitations associated with other methods like invariance-based pretraining and generative approaches."
GPT-4 + Stable-Diffusion = ?: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models - "Advancements in text-to-image generation with diffusion models have led to impressive results, but these models often struggle with prompt understanding in scenarios requiring spatial or common sense reasoning; a solution proposed is to use off-the-shelf frozen large language models (LLMs) in a two-stage generation process to enhance prompt understanding and generate images conditioned on layouts."
Controlling Text-to-Image Diffusion by Orthogonal Finetuning - "Large text-to-image diffusion models have impressive capabilities in generating photorealistic images from text prompts, and this article introduces a principled finetuning method called Orthogonal Finetuning (OFT) to effectively guide or control these models for different downstream tasks, such as subject-driven generation and controllable generation."
Rerender A Video: Zero-Shot Text-Guided Video-to-Video Translation - "The article discusses a novel zero-shot text-guided video-to-video translation framework that aims to ensure temporal consistency across video frames by adapting image models to videos, achieving global style and local texture temporal consistency at a low cost."
Matting Anything - "The Matting Anything Model (MAM) is a versatile framework for estimating alpha matte in images, capable of handling various types of image matting with user prompt guidance and achieving comparable performance to specialized models."
Four-legged robot traverses tricky terrains thanks to improved 3D vision - "Researchers have developed a new model that improves the 3D vision of four-legged robots, enabling them to autonomously navigate challenging terrains and clear obstacles."
Multi-Modal Classifiers for Open-Vocabulary Object Detection - "The article discusses a method for open-vocabulary object detection using multi-modal classifiers, combining language descriptions and image exemplars to improve performance."
Applications
New A.I. Chatbot Tutors Could Upend Student Learning - "A.I. chatbot tutors developed by Khan Academy are being pilot-tested in schools and aim to democratize student access to individualized tutoring while also assisting teachers with tasks like lesson planning."
Amazon is using generative A.I. to summarize product reviews - "Amazon is using AI to summarize customer reviews on its shopping app, providing a brief overview of positive and negative feedback on products."
Meta releases 'human-like' AI image creation model - "Meta Platforms has released a new "human-like" AI model called I-JEPA that can analyze and complete unfinished images more accurately than existing models by using background knowledge about the world to fill in missing pieces of images."
Meta open sources an AI-powered music generator - "Meta has released an open-source AI-powered music generator called MusicGen, which can turn text descriptions into short audio clips and can be steered with reference audio, though it still falls short of human musicians."
I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients. - "An ER doctor explains how he is using ChatGPT to help treat patients by using it to empathically explain medical scenarios to patients and their loved ones, freeing up time for him and his staff to focus on patient care."
AI-powered church service in Germany draws a large crowd - "An experimental AI-powered church service in Germany, featuring a sermon generated by OpenAI's ChatGPT chatbot and delivered by avatars, draws a large crowd and mixed reactions from attendees."
AI-powered church service in Germany draws a large crowd - "An experimental AI-powered church service in Germany, featuring a sermon generated by OpenAI's ChatGPT chatbot and delivered by avatars, draws a large crowd and mixed reactions from attendees."
Business
The economic potential of generative AI: The next productivity frontier - "Generative AI has the potential to transform roles and boost performance across various business functions, adding trillions of dollars in value to the global economy, according to a report by McKinsey. The report estimates that generative AI could add $2.6 trillion to $4.4 trillion annually across 63 use cases, with customer operations, marketing and sales, software engineering, and R&D accounting for 75% of the value. The technology could also significantly increase labor productivity, but will require investments"
Salesforce pledges to invest $500M in generative AI startups - "Salesforce is increasing its investment in generative AI startups, growing its Generative AI Fund from $250 million to $500 million, in order to accelerate the development of transformative AI solutions for the enterprise."
Mistral AI secures €105M in Europe’s largest-ever seed round - "Paris-based startup Mistral AI has raised €105 million in what is reportedly Europe's largest-ever seed round, with plans to launch a large language model (LLM) similar to OpenAI's ChatGPT in early 2024, targeting enterprise clients to improve processes and build new products with AI."
AI Video Creation Pioneer Synthesia Raises $90 Million Series C Led by Accel - "Synthesia, an AI video creation platform, has raised $90 million in Series C funding to continue developing its technology that simplifies video production without the need for cameras or studios, with a focus on making video creation easy for everyone."
Exclusive: Amazon's cloud unit is considering AMD's new AI chips - "Amazon Web Services (AWS) is considering using new artificial intelligence chips from Advanced Micro Devices Inc (AMD) for its cloud services, according to an AWS executive, potentially diversifying the AI development hardware market dominated by Nvidia."
Google launches AI-powered advertiser features in push for automation - "Google is launching two new AI-powered features for advertisers that will automatically find the best ad placements across its services, removing the need for advertisers to think about ad placement and maximizing views of video ads."
A leaked document of Amazon's ideas for using ChatGPT and AI at work lists 67 ways to take advantage of the ChatGPT boom - "Amazon employees are eager to capitalize on the growing popularity of ChatGPT and similar AI technology, as evidenced by a leaked internal document that outlines 67 potential use cases for these chatbots across various teams at Amazon."
Backed by Google’s Gradient, Versed wants to help storytellers create video games using generative AI - "Backed by Google's Gradient Ventures, Versed is a European startup that aims to allow anyone to create their own role-playing game (RPG) simply by writing text-based stories and instructions, using generative AI to interpret the narrative and assign characters and locations to create immersive worlds."
GitHub Survey Finds 92% of Programmers Are Using AI Tools - "A survey conducted by GitHub found that 92% of programmers at large companies are using AI tools in their workflow, with 70% of respondents reporting benefits from using these tools, although code volume may not be the best metric for measuring productivity."
Big banks are talking up generative A.I. — but the risks mean they're not diving in headfirst - "Major banks and fintech companies are cautiously exploring the use of generative artificial intelligence, praising its potential for innovation but expressing concerns about risks and pitfalls, with many banks currently focusing on internal use cases rather than customer-facing applications."
Microsoft's stock hits record after executives predict $10 billion in annual A.I. revenue - "Microsoft's stock reached a record high after analysts predicted that the company's annual revenue from artificial intelligence could reach $10 billion, with the success of OpenAI's ChatGPT chatbot contributing to the growth."
Investors shut out of traditional funding rounds are scouring secondary markets to snap up shares in buzzy AI startups like Dataminr, Hugging Face, and Anthropic - "Investors are turning to secondary markets to buy shares in AI startups, as the demand for AI and machine learning companies continues to grow and venture capitalists seek to get in early on the transformative industry."
Nvidia GPUs are so hard to get that rich venture capitalists are buying them for the startups they invest in - "Rich venture capitalists are purchasing Nvidia GPUs to help AI startups overcome the high prices and shortages of these essential chips, with some VCs even setting up their own AI cloud service to support startups."
Concerns
The harm from AI is already here. What can the US do to protect us? - "The article discusses the need for regulation of artificial intelligence (AI) in the US, highlighting the potential harms already being caused by AI and the lack of effective regulation thus far, while also examining the European Union's efforts to regulate AI."
A.I. makes workers feel so isolated and conflicted that it’s driving them to drink and suffer from insomnia, study finds - "Using artificial intelligence at work can lead to increased feelings of loneliness, insomnia, and alcohol consumption among employees, according to a study, highlighting the psychological impact of AI in the workplace that companies need to consider as they integrate the technology into their operations."
Google, one of AI’s biggest backers, warns own staff about chatbots - "Google is cautioning its employees about the use of chatbots, including its own Bard, due to concerns about the potential leakage of confidential information and the reproduction of data by AI programs."
Policy
UN chief backs idea of global AI watchdog like nuclear agency - "UN Secretary-General Antonio Guterres backs the idea of creating an international AI watchdog body, similar to the International Atomic Energy Agency, to address concerns over the misuse of AI technology."
SEC to Weigh New Artificial-Intelligence Rules for Brokerages - "The SEC is considering implementing new rules to address conflicts of interest related to the use of artificial intelligence in brokerages."
Google challenges OpenAI's calls for government A.I. czar - "Google and OpenAI have differing opinions on how artificial intelligence (AI) should be regulated by the government, with Google preferring a multi-layered, multi-stakeholder approach and OpenAI advocating for a centralized regulatory model."
Copyright © 2023 Skynet Today, All rights reserved.
Last Week in AI #224: EU's landmark AI regulation, AI used to voice John Lennon in last Beatles record, Meta's new voice synthesis AI, and more!
I ask this question a lot, but: in this firehose environment, what would you rank as the top, 2nd, and 3rd most important stories this week? It's pretty subjective overall. I generally tend to gravitate toward where the money is going, but there are lots of ways to take this question.