Google details improvements to AI search answers / OpenAI disrupts propaganda campaigns / Apple to upgrade Siri with advanced AI

Plus: Most Americans expect AI abuses to impact the presidential election, according to a survey.

Inside AI

For May 30, 2024

Here’s a look at today's AI briefing:

Google defends, makes improvements to AI search answers following errors.
OpenAI disrupts propaganda campaigns that used its AI tools.
Apple to upgrade Siri with advanced AI to control apps.
Survey: Most Americans expect AI abuses to impact presidential election.
Salesforce stock dives amid poor growth outlook, AI competition concerns.

Google says it made "technical improvements" to its AI-generated answers in Search following user reports of misinformation. In a blog post today, Google head of search Liz Reid detailed what went into creating the "AI Overviews" and what steps the company has taken to improve them.

More:

The AI Overviews, launched in the U.S. this month, show AI-generated summaries at the top of search results.
The overviews are powered by a custom LLM and integrated into Google's core web ranking to perform standard searches.
According to Reid, user feedback showed the AI Overviews boost satisfaction and help with more complex questions, comparing their accuracy to Search's "featured snippets."

Misinformation:

Reid acknowledged that over the past week, social media users have seen and shared some "odd and erroneous overviews" alongside fake screenshots
For example, the overviews advised using glue to keep cheese on pizza and recommended users eat at least one rock per day.
In some cases, the search queries were "aimed at producing erroneous results," Reid wrote.
Sometimes the overviews linked to the only site that included an answer to the question, such as the case in "How many rocks should I eat?"

Improvements:

Reid said Google has worked on updates to fix broad sets of queries.
Google improved its systems to better detect "nonsensical queries" and has limited satire and humor content.
The company also restricted user-generated content in responses that could mislead, and added triggers to limit less helpful AI Overviews for certain queries, she said.

OpenAI said it thwarted five foreign influence operations that used its AI models to generate fake and deceptive content. The campaigns, involving threat actors from Russia, China, Iran, and Israel, used AI to generate articles, comments, and fake names and bios for social media.

More:

The operations, disrupted over the past 3 months, were an "attempt to manipulate public opinion or influence political outcomes," according to OpenAI.
Two campaigns had Russian origins, including the Doppelganger campaign and the previously unreported "Bad Grammar" operation.
The Chinese network "Spamouflage" used its tools for code debugging, media research, and generating multilingual posts for social media.
Others involved the Iranian International Union of Virtual Media and an Israeli political campaign firm called Stoic.

Zoom out:

OpenAI reported AI-generated content about Ukraine's invasion, Indian elections, Western politics, and Chinese government criticism.
The attacks were labeled low in severity as they didn't gain wide traction, and some users recognized the content as AI-generated.

Apple plans to improve Siri with advanced AI, allowing users to control app functions through voice commands, according to Bloomberg. The update, planned for 2025, will analyze users' activity on their devices to automatically activate Siri features.

More:

Apple engineers reportedly revamped Siri's software with large language models.
Siri will be able to perform tasks like relocating notes, deleting emails, opening documents, summarizing articles, and emailing links.
At launch, the new Siri will control "hundreds" of commands within Apple's own apps. The assistant will only be able to perform one command at a time.
Apple eventually wants Siri to execute multiple commands, such as taking and sending a photo.

Zoom out:

The new Siri will include code to determine if tasks should be handled on-device or on Apple's servers.
The on-device AI features may require an iPhone 15 Pro or newer.
Apple plans to announce the new Siri features at next month's WWDC. However, they're not scheduled to launch until 2025.

Nearly 80% of Americans expect AI abuses to impact the outcome of this year's presidential election, according to a survey by Elon University Poll and the Imagining the Digital Future Center.

What the numbers show: The survey reveals that 73% of Americans expect AI to be used to manipulate social media, 70% foresee AI-generated fake media, and 62% believe AI will be used to try and suppress voter turnout. Together, a majority expects the AI abuses to impact who wins the November presidential election.

What else: The survey found that nearly 70% of Americans doubt voters can detect deepfaked media. Furthermore, 93% believe candidates who maliciously fake media should face penalties, with 46% supporting removal from office.

Quotable: Voters expect "new kinds of misinformation, faked material, and voter-manipulation tactics are being enabled by AI," said Lee Rainie, director of Elon University's Imagining the Digital Future Center. "What's worse, many aren't sure they can sort through the garbage they know will be polluting campaign-related content."

Salesforce shares fell 20% on Thursday after forecasting record-low sales growth, sparking fears that AI competition is impacting demand. The CRM leader reported that its AI data cloud business made up a quarter of its million-dollar-plus deals in Q1, the same as the previous quarter.

More:

Thursday marked Salesforce's steepest trading decline since July 2004.
Salesforce reported weaker-than-expected Q1 results and issued disappointing guidance amid macroeconomic challenges, marking its first revenue miss since 2006.
Amid the slowing growth, analysts said they are awaiting signs of momentum from Salesforce's data cloud and generative AI.
Morgan Stanley analysts noted that weak bookings and unmet GenAI revenue expectations raise competitive concerns.
Goldman Sachs analysts still called Salesforce "an under-appreciated Gen-AI winner."

Zoom out:

In April, Salesforce launched Einstein Copilot, its business-oriented suite of AI tools.
CEO Marc Benioff emphasized the company's potential in the AI era.

Quick Hits:

Thinking about growing your business? We'll work as hard as it takes to make Ohio your competitive advantage.*
The U.S. Army awarded Palantir a $480M contract for the AI-driven Maven Smart System. The system is part of the military's larger AI-powered Project Maven, which pulls different data sources into a single interface and uses the data to auto-detect potential targets.
U.S. officials are delaying Nvidia and AMD's AI accelerator licenses to the Middle East, pending a security review of AI development in the region.
Perplexity AI's new Pages feature pulls real-time web data to generate article drafts tailored to audiences.
Top tech, business, and government leaders are expected to discuss AI and many more topics at this week's secretive annual Bilderberg Meeting in Madrid.
Join industry's leading AI conference - free passes available! Register Now! passes are selling quickly.*

*This is sponsored content

Beth Duckett

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.