ChatGPT and most other generative AI relies on the power of the cloud and a strong connection to the internet, but now some big names in tech are pushing hard to run generative AI directly on smartphones and laptops.

Why it matters: Running AI models on device could make them cheaper, faster and more private.

Cell phone chip specialist Qualcomm showed off a demo Tuesday of a reference smartphone running Stable Diffusion and producing an image in less than a second.

• In a demo earlier this year, a similar task took 15 seconds.
• At its Snapdragon Summit in Maui yesterday, Qualcomm also showed versions of Meta's Llama 2 model running on both smartphones and a prototype Snapdragon-powered laptop.

What's next: Apple was caught a bit flat footed as generative AI took off, but is said to be hard at work on models of its own and is aiming to incorporate broader uses of the technology into upcoming versions of iOS and Siri, per a weekend report from Bloomberg's Mark Gurman.

• Running its models on-device would be in line with the company's privacy efforts which have focused on minimizing the amount of sensitive information collected by Apple servers.
• Google is making widespread use of on-device AI in its Pixel devices, especially with photos. It said Tuesday it is also working with Qualcomm to allow a number of its AI models to run on Android devices.

The big picture: Generative AI shows huge potential for a range of tasks from drafting documents and presentations to sifting through large amounts of data. However, creating software that has to constantly ping servers over the internet adds significant costs.

• "Because cloud costs are going to skyrocket, these guys are going to say 'OK, I'm going to try to figure out how to do more and more on device,'" Qualcomm senior vice president Alex Katouzian told Axios.
• Qualcomm focused on the power of on-device AI in touting both its latest Snapdragon 8 Gen 3 mobile chip and a long awaited PC chip it hopes to use to finally grab a significant chunk of the PC market.
• And even if some queries do require the power of the cloud, running others locally can save money and improve performance. Microsoft and others envision such a hybrid approach.

Hardware and chip makers are also hoping that generative AI could help spur sales of computers and smartphones, both of which have flagged after seeing a boom during the pandemic.

• Consumers who once replaced their smartphones every couple of years are now waiting as long as five or six years before purchasing a new one, Katouzian said.

Between the lines: Companies ranging from startups to giants like Microsoft and Meta also see the potential of AI to help unlock the hold that Google and Apple have had over smartphones, due in large part to their control over the main app stores for their respective platforms.

• There are new uses for generative AI that can run proactively, rather than requiring a person to actively query a chatbot. Qualcomm showed demos on Tuesday of an always-running AI agent that can offer suggested meeting places and times, or photos to share.
• "There is zero chance of that running in the cloud," Qualcomm senior vice president Durga Malladi said in an interview. "It's simply not possible."

Yes, but: Today's generative AI consumes a tremendous amount of computing power. Running similar models on devices still requires a lot of horsepower and its impact on battery life is yet to be seen.

• Even if today's AI services move on device, tomorrow's will keep taxing the hardware.

Go deeper: New Qualcomm chip looks to break Intel's PC dominance

Research from IBM shows ChatGPT is already pretty good at writing believable phishing emails, despite efforts to limit its ability to do harm, as Axios' Sam Sabin reports.

Why it matters: Cybersecurity officials and industry leaders have long warned that hackers could weaponize ChatGPT and similar AI tools to quickly write phishing emails that the average person would think are authentic.

• IBM's research is some of the first that provides concrete details of how close AI-enabled tools are to perfecting phishing.

Driving the news: A team of IBM researchers released the results of an A/B testing experiment they ran with an unspecified global healthcare company's roughly 1,600 employees.

• In the experiment, half of the employees got a phishing email written fully by IBM's X-Force team.
• The other half got an email written using ChatGPT.

By the numbers: 14% of employees who received the human-written phishing email fell for it and clicked on a malicious link, according to the IBM report released Tuesday.

• But the ChatGPT-written email was pretty close, with 11% of its targets falling for the note it wrote.

Between the lines: It only took five minutes for Stephanie "Snow" Carruthers, IBM's chief people hacker who led the experiment, and her team to get ChatGPT to spit out the email her team ended up using.

• Meanwhile, her team usually needs about 16 hours to write a believable phishing email, since they closely study the organization they're targeting to determine what issues employees are interested in.

What they're saying: "It makes me kind of fearful for the future," Carruthers told Axios.

How it works: ChatGPT developer OpenAI has put in safeguards that prevent the generative AI chatbot from responding to direct requests for a phishing email, malware or other malicious cyber tools.

• However, social engineers like Carruthers have been able to work around those safeguards by asking ChatGPT a series of questions about the target company and its industry.

The intrigue: Initially, three of IBM's clients were signed up to participate in the study. But once they saw the email ChatGPT was able to write, two companies backed out because they feared too many of their employees would fall for it.

Yes, but: Carruthers told Axios the ChatGPT-written email lacked the emotional intelligence needed to trick more employees.

• "That human element is so important to social engineering," she said. "The AI one, it still kind of felt cold and robotic to me."

Threat level: IBM's X-Force has yet to see wide use of generative AI in current campaigns, but hackers are already selling AI tools that could soon help expedite attacks.

• A group of 33 state attorneys general on Tuesday sued Meta, alleging the Facebook and Instagram parent knowingly issued products and features via its apps that pose psychological risks to children and teen users. (Axios)
• California suspended the driverless car permit for GM's Cruise unit, putting its San Francisco robotaxi service on hold. (Axios)
• Apple has scheduled an online event for 5pm on Oct. 30, where it is expected to debut new Macs. (CNBC)
• Pebble (formerly called T2), a would-be successor to Twitter, said Tuesday it will shut down. (TechCrunch)
• Microsoft reported increased profit and revenue in first quarter results, and Satya Nadella says the company is "rapidly infusing AI across every layer of the tech stack and for every role and business process."
• Alphabet released decent third quarter earnings yesterday, assuring advertisers that they would soon be able to make money from generative AI. (The Verge)

I had some time to snorkel off Molokini and off the Maui coast, where I used GoPro's latest Hero 12 action camera to capture this shot of a sea turtle.