| |
| Here's a look at today's AI briefing: - Apple shareholders reject sharing of AI transparency report.
- Top AI models generate inaccurate responses to election questions, study finds.
- Tumblr owner to sell AI training data to OpenAI, Midjourney.
Beth
p/beth-duckett | |
| 1 | | Apple shareholders today rejected the AFL-CIO's proposal for an AI transparency report, fearing it would harm the company's competitive strategy. The pension trust of the labor union federation wanted Apple to disclose how it uses AI in its business practices and share ethical guidelines for the technology. More: - Norges Bank Investment Management and Legal & General, Apple's eighth- and 10th-largest shareholders, also wanted Apple to disclose its AI usage.
- Apple opposed the measure, saying it could reveal strategic advantages in the competitive AI field.
- The AFL-CIO Equity Index Funds has also pressed companies like Disney, Netflix, and Warner Bros. Discovery to publish AI transparency reports and ethical guidelines.
- A similar proposal is up for a vote at Disney's annual meeting in April.
Zoom out: - Also during Apple's shareholders meeting, CEO Tim Cook said the company plans to "break new ground" in generative AI this year.
- Apple sees "incredible breakthrough potential" in GenAI, which is why it's "currently investing significantly" in the technology, he said.
- Cook also praised the machine-learning features in Apple's new Vision Pro headset, citing "decades of innovation."
|     | |
| 2 | | Leading AI models from OpenAI, Google, Meta, Anthropic, and Mistral AI often produce inaccurate or misleading responses to election-related questions, according to a report by the AI Democracy Projects. For example, the study found that Meta's Llama 2 mistakenly claimed California voters could vote by text. 
Details: The study, involving over 40 election officials, researchers, and journalists, gauged the accuracy of OpenAI's GPT-4, Google's Gemini, Meta's Llama 2, Anthropic's Claude, and Mistral AI's Mixtral. They tested the LLMs on 26 common voting questions, rating a "small sample" of 130 answers for bias, accuracy, completeness, and harmfulness. What the numbers show: Researchers discovered that 51% of the chatbot responses were inaccurate, 40% harmful, 38% incomplete, and 13% biased. GPT-4 outperformed other models in accuracy, while Anthropic's Claude was inaccurate about half the time. Google's Gemini, Meta's Llama 2, and Mistral's Mixtral had over 60% inaccurate responses. More: The models often misdirected voters to nonexistent polling places and generated illogical answers based on outdated data. Meta's Llama 2 incorrectly stated California allows voting by text, and none correctly noted Texas' ban on campaign logo attire at polls. Response: Meta dismissed the study findings, arguing they don't reflect real chatbot interactions. The company clarified that Llama 2 is for developers, not consumers, adding that its public-facing chatbot often directs users to official election information. Anthropic said it will launch a new version of its AI tool with accurate voting data, while OpenAI intends to hone its tools based on user feedback. Google and Mistral had not yet responded. Why it matters: As U.S. presidential primaries get underway, more Americans are turning to chatbots for information, sparking concerns about AI spreading misinformation or deterring voter turnout. Recently, AI robocalls using a fake Joe Biden voice discouraged voters from casting ballots in the New Hampshire primary. "The chatbots are not ready for prime time when it comes to giving important nuanced information about elections," said Seth Bluestein, a Philadelphia city commissioner and a study participant. | | | |
| |
|
| 3 | | Automattic, the owner of Tumblr and WordPress.com, is expected to sign deals to share data with OpenAI and Midjourney for AI model training, 404 Media reports. While Automattic didn't confirm the report, the company said it would allow users to opt out of sharing their public blog content with AI platforms and other third parties. More: - Automattic said it currently blocks crawlers from major AI platforms, including those from leading tech companies, and updates its lists to include new ones as they are launched.
- However, the company said it is now "working directly with select AI companies" as long as their plans align with "attribution, opt-outs, and control."
- According to 404, internal messages from Automattic reveal that the company mistakenly scraped an "initial data dump" of all Tumblr's public posts from 2014 to 2023, including content not meant for public view.
Zoom out: - Tumblr and WordPress join Reddit in choosing to strike deals with AI firms instead of filing lawsuits over their content usage.
- In December, The New York Times filed a lawsuit against OpenAI and Microsoft for allegedly training chatbots on millions of its articles without consent.
- OpenAI has now asked a judge to dismiss portions of that copyright suit, claiming the Times "hacked" ChatGPT and other systems to create deceptive evidence for its arguments.
| | | |
| 4 | | Adobe is developing a generative AI tool for creating music. Project Music GenAI Control, currently an "early-stage" experiment, generates music from text prompts and lets users directly edit the audio. More: - The tool generates music in specific styles based on text prompts, allowing users to customize tempo, intensity, and structure and create loops for backing tracks.
- According to Adobe, the tool can also revise, cut, and loop music sections, serving as a track editor.
- Project Music GenAI is a collaborative effort between Adobe, the University of California, and Carnegie Mellon University's School of Computer Science.
- A public release date is yet to be announced.
Zoom out: - Google's MusicLM and Meta's AudioCraft also generate audio from text prompts, although their music editing tools are limited.
| | | |
| 5 | | 🍋 FRESH FUNDING 🍋 - AI-powered enterprise search engine Glean Technologies secured $200M in Series D funding at a $2.2B valuation. Kleiner Perkins and Lightspeed Venture Partners led the round, with additional backing from venture capital divisions of Databricks, Workday, and Capital One Financial Corp.
- AI-powered healthcare technology data and analytics company Reveleer raked in $65M in fresh funding led by Hercules Capital.
- Lightspeed led a $64M Series B funding round into Intenseye, an AI startup that prevents workplace accidents. Existing backers Insight Partners, Point Nine, and Air Street Capital participated in the round.
- French photo-editing generative AI startup Photoroom raised a Series B round worth $43M led by Balderton Capital and Aglaé Ventures. Y Combinator backed the round, which valued the firm at around $500M.
- Data curation startup DatologyAI raised $11.65M in seed funding. The seed round was led by Amplify Partners, with support from Radical Ventures, Conviction Capital, Outset Capital, and Quiet Capital.
For the latest on AI startup funding and more in the VC community, subscribe to our Inside Venture Capital newsletter. | | | |
| 6 | | QUICK HITS: - Discover the trends shaping 2024 and learn how trust will be the key to business success. Shape your strategy and get ahead with this latest report.*
- Google CEO Sundar Pichai said the company is working to fix its Gemini AI bot after it produced racially inaccurate images and text, deeming the outputs "completely unacceptable."
- A French finance ministry official refuted claims that France was aware of Microsoft's $16M Mistral AI investment or lobbied for more relaxed EU AI regulations to favor the tech giant.
- Police are investigating AI-created nude images of students made by peers at a Beverly Hills middle school.
- Intel plans to supply processors for 100 million AI-equipped PCs by 2025, targeting over 20% of the global PC market, according to executives.
- Visa and Mastercard have come to a historic $5B settlement, marking a pivotal moment for small business owners. If your business accepts Visa/MC, claim your share today*
*This is a sponsored post. | | | |
| Upcoming Events | | MAR
7 | | Portfolio Diversification - Learn from knowledgable experts on the latest trends in alternative investment assets like private equity, crypto, real estate, and the art market. | | | | | | | MAR
27 | | Discover how generative AI can speed up your enterprise, expand your business model, and improve your value proposition. Network with hundreds of AI leaders.* | | | | | | * This is a sponsored event | | | |
 | AI and technology writer | | Beth is a contributing editor and writer of Inside's AI and Tech newsletters. She has written for publications including USA Today, the Arizona Business Gazette, and The Arizona Republic, where she received recognition with a Pulitzer Prize nomination and a First Amendment Award for collaborative reporting on state pension cost increases. You can reach her at Beth.Duckett@yahoo.com. | | This newsletter was edited by Aaron Crutchfield | |
| |
 | Fellow is the all-in-one AI meeting transcription and management software for remote and hybrid teams. | |
|
| |
Comments
Post a Comment