AI Roundup 068: The ScarJo thing
AI Roundup 068: The ScarJo thing
May 24, 2024.
The ScarJo thing
OpenAI made headlines again after Scarlett Johansson's legal team requested details on how the latest GPT-4o voice was trained, claiming it sounds eerily similar to hers.
Between the lines:
If you want to understand the timeline, OpenAI has published the sequence of events from its point of view.
Some lawyers believe that Johansson may have a case to due California’s “right of publicity” laws, though the court of public opinion seems to already be on her side.
Records so far have shown that OpenAI worked with other voice actors months before contacting Johansson, but given Sam Altman's "her" tweet, it's still not a good look.
Ultimately, the voice has been pulled from ChatGPT until the legal concerns are ironed out.
Elsewhere in OpenAI:
The company also walked back aggressive employee exit contracts that made non-disparagement agreements a requirement for keeping vested equity.
News Corp and OpenAI have a new deal to bring news content to OpenAI, reportedly worth $250M over five years.
Apple is reportedly gearing up for a major announcement with OpenAI at this year's WWDC.
And multiple Silicon Valley rivals are uniting behind Triton, OpenAI's software designed to compete with Nvidia.
Copilot PCs
Microsoft held its Surface and Windows AI and Build 2024 events, which were jam-packed full of Copilot news.
All the announcements I could find:
A slew of "Copilot Plus PCs" from Acer, Asus, Dell, HP, Lenovo, and Samsung.
Recall, a Windows 11 tool that remembers user activity to make it searchable.
GitHub Copilot Extensions, for third-party software to add new skills to GitHub Copilot.
Team Copilot, to work across Microsoft Teams.
Custom Copilot AI agents for developers and businesses.
MS Paint is getting an AI image generator.
Azure AI Studio, now with support for GPT-4o and a multimodal Phi-3.
Real-time translation and dubbing for videos in Microsoft Edge.
Elsewhere in the FAANG free-for-all:
Google is rolling out new photo editing tools, as well as ads with AI-generated images that preview clothes on different body types.
Amazon is reportedly planning to launch an upgraded Alexa with generative AI features later this year.
And Meta may consider charging users for a more advanced version of Meta AI.
People are worried about AI deepfakes
The issues of political deepfakes, nonconsensual porn, and AI CSAM haven't gone away - if anything, we’ll see more issues as we get closer to major elections.
What to watch:
The White House called on AI companies to help curb the creation, spread, and monetization of explicit AI deepfakes.
The FBI arrested a man for generating CSAM using AI, while predators flock to AI-generated images of children elsewhere on social media.
FCC Chairwoman Rosenworcel proposed requiring disclosure of any AI content in political ads on radio and TV.
And Indian politicians are bombarding voters with deepfakes, as social platforms struggle to properly police problematic ads.
Elsewhere in AI anxiety:
Google's "AI Overview" search results are running into issues, like telling users to put glue on their pizza and eat one small rock a day.
Alphabet and Meta have talked to Hollywood studios about licensing content for AI video generation tools, with mixed results.
After giving students a $10,000 prize for their AI startup, Emory University then suspended them for cheating when they built it.
And leading AI companies commit to "safe AI development" at the second AI Safety Summit, hosted in Seoul.
Things happen
Anthropic researchers figure out a way to look inside the "black box" of LLMs. After tripling sales for three quarters, Nvidia faces new competition and a shifting AI market. Meta AI chief says large language models will not reach human intelligence. Zuckerberg creates Meta Advisory Group to advise on AI products. China’s latest answer to OpenAI is ‘Chat Xi PT’. Google's new Search AI Overview will soon have ads. Chameleon: Meta's new multimodal LLM. Humane is looking for a buyer after its disastrous Ai Pin launch. Amazon changes plans to order Nvidia ‘superchip’ to await updated model. Adobe adds generative AI tools to Lightroom. Alibaba slashes prices for its AI services by up to 97%. A look at Snap's shift towards AI and AR after revamping its ad business. VCs and the military are fueling self-driving, off-roading startups. Building an AI game studio: what we've learned so far. Hollywood agency CAA aims to help stars manage their own AI likenesses. EU countries vote to approve the EU AI Act, taking effect in June 2024. Inflection AI, after losing its founders and dozens of staff, debuts a new leadership team. TikTok turns to generative AI to boost its ads business. AI doppelganger experiment, Part 1. Training algorithms on stereotypical facial expressions is bound to mislead. A look at Meta's AI strategy: Give it away for free. Training AI on stereotypical facial expressions is bound to mislead. AI companies seek big profits from ‘small’ language models. $6M fine for robocaller who used AI to clone Biden’s voice. Google plans to spend €1 billion to expand its Finland data center. Palantir's first ever AI warfare conference. China’s ‘AI-in-a-box’ products threaten Big Tech’s cloud growth strategies. How cuddly robots could change dementia care. It's time to believe the AI hype.
