AI Roundup 085: You win some, you Newsom
AI Roundup 085: You win some, you Newsom
September 20, 2024.
NotebookLM
If you haven’t seen it yet, Google’s NotebookLM tool now lets you generate podcast-style summaries of uploaded sources. I’ve used today’s roundup as a test with no adjustments or editing, and Google’s AI podcast is below. Let me know what you think!
You win some, you Newsom
California Governor Gavin Newsom signed eight different laws regulating AI deepfakes this week, with implications for both Hollywood and Silicon Valley tech companies.
Breaking down the bills:
AB 2655, AB 2839, and AB 2355 are meant to combat deepfake election content. AB 2655 in particular would require tech companies to remove or label election-related deepfakes within certain periods.
AB 2602 and AB 1836 place more stipulations on when companies can use AI-generated likenesses of celebrities, aiming to curb unauthorized uses of AI replicas.
SB 942, SB 926, and SB 981 prohibit creating and sharing deepfake porn, and require social platforms to handle user reports of unauthorized deepfakes.
Meanwhile, SB 1047 - California's hotly contested AI regulation - is still sitting on the Governor's desk, as he weighs the bill's potential "chilling effect."
Elsewhere in AI geopolitics:
Nearly a year after the UK's AI Safety Summit and President Biden's AI Executive Order, the White House is planning the first meeting of the International Network of AI Safety Institutes.
A UN advisory body made seven recommendations for governing AI, including establishing a panel to provide impartial scientific knowledge about AI.
And in an open letter coordinated by Meta, more than two dozen companies warned that the EU might reap fewer rewards of AI because of the bloc's tech regulations.
Speaking of Hollywood
More and more AI startups are making inroads with Tinseltown, as they look to partner with existing IP holders rather than seek forgiveness later (and risk potential lawsuits).
Between the lines:
Lionsgate is working with Runway to use its content library in exchange for a custom model that the studio can use to edit and produce new movies and TV shows.
The estates of long-dead celebrities are finding a new cash cow in the form of AI voices, as more partner with companies like ElevenLabs to license iconic audio.
But perhaps the biggest AI entertainment is yet to come next year, when Elvis Evolution - a two-hour immersive AI performance -premieres in London.
Elsewhere in AI anxiety:
LinkedIn confirms it's training its AI models on user data by default, while allowing opt-outs for those with privacy concerns.
A Starling Bank survey reveals 28% of UK adults have encountered an AI voice cloning scam in the past year, while 46% were unaware such scams existed.
And Wordfreq, a project analyzing human language usage, shuts down due to data pollution from generative AI.
Elsewhere in the FAANG free-for-all:
Amazon launched an AI-powered video generator for advertisers and Amelia, an AI assistant for third-party sellers.
Google debuted DataGemma models to reduce hallucinations in statistical data queries.
YouTube integrated Google DeepMind's AI video generation model Veo into Shorts for creating high-quality backgrounds and clips.
And Microsoft’s new Copilot Pages give 365 customers the ability to share and collaborate on AI-generated content.
OpenAI is oversubscribed
Despite the drama surrounding OpenAI in recent months, the company appears to be on track to complete its latest $6.5 billion funding round - with investors clamoring to put in billions more.
Why it matters:
While overall sentiment around AI's potential ROI is cooling, clearly there's still enormous demand to invest in its frontrunners.
The new round values OpenAI at an eye-watering $150B, nearly double its previous $85B valuation.
But the new price tag may hinge on whether the company successfully transitions to a more traditional, for-profit structure.
Elsewhere in AI investment:
BlackRock and Microsoft are raising a new $30B AI fund to invest in AI infrastructure.
Groq and Aramco plan to build the world's largest AI inferencing center in Saudi Arabia by the end of 2024.
And ByteDance, owner of TikTok, aims to mass produce two AI chips designed in collaboration with TSMC by 2026.
Things happen
SocialAI is a social network for you and AIs. Snap new tool generates AI videos from text prompts. AI is most useful as a feature, not as a standalone product. T-Mobile partners with OpenAI on an AI-powered customer service system. Alibaba releases 100+ open-source models under Qwen 2.5. Google plans to flag AI-made images with C2PA metadata in Search. Runway announces an API for its video-generating AI model. Researchers calculate the water usage of GPT-4 generating a 100-word email. Larry Ellison's AI-powered surveillance dystopia is already here. Slack unveils Agents for Slack, allowing access to various AI agents. Microsoft pitches AI to help oil companies find new reserves. BHP warns that AI demand could exacerbate a copper shortage. A profile of Fei-Fei Li's World Labs, which aims to create "spatial intelligence" in AI. Homemade porn site promises not to train AI on performers. Google serves AI slop as the top result for Hieronymous Bosch painting. Researchers release Llama 3.1 Omni Model. Void, an open-source alternative to Cursor and GitHub Copilot. Stephen Fry on AI: A means to an end or a means to our end. Don't ask if AI can make art - ask how AI can be art. The 'deep doubt era' is here. A look at Google's 7-year mission to give AI a robot body. OpenAI launches 'independent' safety board that can stop model releases. Salesforce upgrades all customers from Einstein Copilot to Agentforce. Intel and AWS plan to coinvest in making a new AI chip.
