A recent poll shows K-12 students’ familiarity with ChatGPT rose from 37% to 75% in just over a year. The survey, by Impact Research for the Walton Family Foundation, also found that teachers’ familiarity with ChatGPT jumped from 55% to 79% from February 2023 to May 2024. — Read More
Recent Updates Page 108
Deception abilities emerged in large language models
Large language models (LLMs) are currently at the forefront of intertwining AI systems with human communication and everyday life. Thus, aligning them with human values is of great importance. However, given the steady increase in reasoning abilities, future LLMs are under suspicion of becoming able to deceive human operators and utilizing this ability to bypass monitoring efforts. As a prerequisite to this, LLMs need to possess a conceptual understanding of deception strategies. This study reveals that such strategies emerged in state-of-the-art LLMs, but were nonexistent in earlier LLMs. We conduct a series of experiments showing that state-of-the-art LLMs are able to understand and induce false beliefs in other agents, that their performance in complex deception scenarios can be amplified utilizing chain-of-thought reasoning, and that eliciting Machiavellianism in LLMs can trigger misaligned deceptive behavior. GPT-4, for instance, exhibits deceptive behavior in simple test scenarios 99.16% of the time (P < 0.001). In complex second-order deception test scenarios where the aim is to mislead someone who expects to be deceived, GPT-4 resorts to deceptive behavior 71.46% of the time (P < 0.001) when augmented with chain-of-thought reasoning. In sum, revealing hitherto unknown machine behavior in LLMs, our study contributes to the nascent field of machine psychology. — Read More
Apple Intelligence is Right On Time
Apple’s annual Worldwide Developer Conference keynote kicks off in a few hours, and Mark Gurman has extensive details of what will be announced in Bloomberg, including the name: “Apple Intelligence”. As John Gruber noted on Daring Fireball:
His report reads as though he’s gotten the notes from someone who’s already watched Monday’s keynote. I sort of think that’s what happened, given how much of this no one had reported before today.
… The irony of the leak being so huge is that nothing is particularly surprising: Apple is announcing and incorporating generative AI features throughout its operating systems and making them available to developers. Finally, the commentariat exclaims! Apple is in danger of falling dangerously behind! The fact they are partnering with OpenAI is evidence of how desperate they are! In fact, I would argue the opposite: Apple is not too late, they are taking the correct approach up-and-down the stack, and are well-positioned to be one of AI’s big winners. — Read More
NVIDIA Unveils “NIMS” Digital Humans, Robots, Earth 2.0, and AI Factories
NewsBreak: Most downloaded US news app has Chinese roots and ‘writes fiction’ using AI
Last Christmas Eve, NewsBreak, opens new tab, a free app with roots in China that is the most downloaded news app in the United States, published an alarming piece about a small town shooting. It was headlined “Christmas Day Tragedy Strikes Bridgeton, New Jersey Amid Rising Gun Violence in Small Towns.”
The problem was, no such shooting took place. The Bridgeton, New Jersey police department posted a statement on Facebook on December 27 dismissing the article – produced using AI technology – as “entirely false”.
“Nothing even similar to this story occurred on or around Christmas, or even in recent memory for the area they described,” the post said. “It seems this ‘news’ outlet’s AI writes fiction they have no problem publishing to readers.” — Read More
Securing Research Infrastructure for Advanced AI
We’re sharing some high-level details on the security architecture of our research supercomputers.
OpenAI operates some of the largest AI training supercomputers, enabling us to deliver models that are industry-leading in both capabilities and safety while advancing the frontiers of AI. Our mission is to ensure that advanced AI benefits everyone, and the foundation of this work is the infrastructure that powers our research.
To achieve this mission safely, we prioritize the security of these systems. Here, we outline our current architecture and operations that support the secure training of frontier models at scale. This includes measures designed to protect sensitive model weights within a secure environment for AI innovation. While these security features will evolve over time, we think it’s valuable to provide a current snapshot of how we think about security of our research infrastructure. We hope this insight will assist other AI research labs and security professionals as they approach securing their own systems (and we’re hiring). — Read More
The CEO of Zoom wants AI clones in meetings
Today, I’m talking with Zoom CEO Eric Yuan — and let me tell you: this conversation is nothing like what I expected. Eric started Zoom after working at Cisco and realizing there was an opportunity to make videoconferencing simpler and easier to use. And he was right: Zoom is now a household name — especially after usage exploded during the pandemic.
But usage has since come down, and Zoom faces a number of business challenges he and I talked about. Yet, it turns out, Eric wants Zoom to be much, much more than just a video chat platform. He wants to take on Microsoft and Google in the enterprise software market by making docs and email and other productivity tools like chat. And like virtually every other company, Zoom now has a big investment in AI — and Eric’s visions for what that AI will do are pretty wild.
See, Eric really wants you to stop having to attend Zoom meetings yourself. — Read More
Why Google’s AI Overviews gets things wrong
When Google announced it was rolling out its artificial-intelligence-powered search feature earlier this month, the company promised that “Google will do the googling for you.” The new feature, called AI Overviews, provides brief, AI-generated summaries highlighting key information and links on top of search results.
Unfortunately, AI systems are inherently unreliable. Within days of AI Overviews’ release in the US, users were sharing examples of responses that were strange at best. It suggested that users add glue to pizza or eat at least one small rock a day, and that former US president Andrew Johnson earned university degrees between 1947 and 2012, despite dying in 1875.
On Thursday, Liz Reid, head of Google Search, announced that the company has been making technical improvements to the system to make it less likely to generate incorrect answers, including better detection mechanisms for nonsensical queries. It is also limiting the inclusion of satirical, humorous, and user-generated content in responses, since such material could result in misleading advice.
But why is AI Overviews returning unreliable, potentially dangerous information? And what, if anything, can be done to fix it? — Read More
Hollywood Nightmare? New Streaming Service Lets Viewers Create Their Own Shows Using AI
Generative artificial intelligence is coming for streaming, with the release of a platform dedicated to AI content that allows users to create episodes with a prompt of just a couple of words.
Fable Studio, an Emmy-winning San Francisco startup, on Thursday announced Showrunner, a platform the company says can write, voice and animate episodes of shows it carries. Under the initial release, users will be able to watch AI-generated series and create their own content — complete with the ability to control dialogue, characters and shot types, among other controls. — Read More
Tribeca Festival to Debut Short Films Made Using OpenAI
The Tribeca Festival will feature five short films made using technology from OpenAI.
The films will use OpenAI’s Sora, which is a text-to-video model that accepts textual descriptions and generates video clips based on them. This is the first time films using this technology will be showcased at the festival.
… The films will be screened June 15, with a conversation afterwards with the filmmakers. — Read More
Rick's Cafe AI 