Key Points:
- Audible invites narrators to create AI voice models, offering royalties and control.
- Meta admits to using public data from all Australian adults to train its AI as privacy laws tighten.
- OpenAI’s o1 model shines in math and science, though it lacks some GPT-4o features.
This week’s AI news covers SAG-AFTRA’s new AI deal with game studios, Elon Musk’s denial of Tesla-xAI licensing rumors, and Adobe’s upcoming AI video tools for Premiere Pro. Here’s a breakdown of the top stories:
Video Game Actors Secure AI Agreement with Major Studios
SAG-AFTRA has reached agreements with 80 game studios on AI protections and wage improvements for performers following a strike. However, strikes continue against major players like Disney and Warner Bros. as the union seeks broader protections from AI exploitation.
Nvidia’s AI Chips Cheaper to Rent in China than the US
Despite export bans, Nvidia’s AI chips are being rented in China at $6 per hour, compared to $10 in the U.S.. The chips are reportedly smuggled into China, and Nvidia is now under a U.S. antitrust investigation due to its AI market dominance.
Elon Musk Denies Tesla-xAI Licensing Deal
Elon Musk refuted rumors that Tesla would license xAI’s technology, stating there’s “no need to license anything.” However, he did mention plans to discuss a potential $5 billion investment in xAI with Tesla’s board.
Audible Invites Narrators to Create AI Voice Models
Amazon’s Audible is inviting select narrators to create AI voice models, allowing them to earn royalties and control which titles feature their voice. This beta trial is limited to the U.S. and focuses on improving AI-generated audiobooks while compensating narrators per title.
Amazon to Invest £8 Billion in UK for Cloud and AI Expansion
Amazon Web Services (AWS) is set to invest £8 billion in UK data centers over the next four years, generating 14,000 jobs annually. This investment is expected to contribute £14 billion to the UK economy by 2028, fueled by the growing demand for AI.
Taylor Swift Endorses Kamala Harris After AI Misinformation Incident
Taylor Swift publicly endorsed Kamala Harris for the 2024 U.S. presidential election after AI-generated images falsely linked her to Donald Trump’s campaign. Swift stressed the importance of transparency to combat AI-driven misinformation.
Study Uncovers AI Bias Against African American English
A study found that AI models like ChatGPT and T5 exhibit bias against speakers of African American English (AAE), often associating them with negative traits and lower-status jobs compared to speakers of Standard American English. This raises concerns about AI’s impact on hiring and law enforcement.
UBS Launches AI Tool for Rapid M&A Analysis
UBS has rolled out a new AI tool that can analyze 300,000 companies in seconds to facilitate M&A deals, generate buy-side ideas, and identify potential buyers. However, UBS faces compliance and data security challenges with this powerful tool.
Mistral Introduces Its First Multimodal AI Model
Mistral has unveiled its Pixtral 12B model for text and image processing. The model, which incorporates a vision adapter, is designed for tasks like image captioning and classification. It’s available for free on GitHub and Hugging Face for research purposes but requires a commercial license for business use.
Adobe’s AI-Generated Video Tools Coming to Premiere Pro
Adobe will add AI-driven video features to Premiere Pro by the end of 2024. These tools will allow users to extend and generate video clips with prompts and images. Adobe is also enhancing protections for stock content creators and expanding moderation efforts.
Google’s AI Notebook Now Creates Podcasts from Source Material
Google’s NotebookLM has introduced a new Audio Overview feature, where AI generates podcast summaries of source material. While offering a natural-sounding overview, the tool has accuracy limitations and currently supports only English.
Amazon Adds Sponsored Ads to AI Chatbot Rufus
Amazon is set to integrate sponsored ads into its AI shopping assistant, Rufus, for U.S. users. These ads will be contextually relevant, but advertisers won’t receive performance metrics, which could complicate campaign analysis.
Chinese AI Models Lag Six Months Behind US Counterparts
Chinese AI models are currently trailing their U.S. counterparts by six to nine months, but heavy investments by companies like Alibaba and Tencent are expected to help them catch up quickly.
OpenAI’s New O1 AI Model Tackles Complex Math Problems
OpenAI launched its o1 model, designed to handle complex math, coding, and science tasks with performance comparable to PhD students. While the model excels in these areas, it is still in preview mode and lacks certain features found in GPT-4o, such as file uploads and web browsing capabilities.