. . AI: Riddle time: What is highly personalized, in your pocket, and more persuasive than a human? (3.22.24)

The answer is AI. Below you'll find a new study finding that personalization in AI-powered debates with humans leads to more persuasive outcomes. While this isn't surprising, the fact that GPT-4 beats most humans is a phenomenal feat and highlights a comment Sam Altman, CEO of OpenAI, made a few months ago: "I expect AI to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes." 

It's Friday, the top links clicked by readers of this newsletter this week have been: VC Yohei Nakajima's open source MindGraph, Glowbom's sketch-to-app AI, and Contextual.ai's RAG 2.0.

Thanks for reading and sharing,

Marshall Kirkpatrick, Editor

Study Reveals Personalization Increases AI's Persuasiveness

First impacted: Social media users, Social Media Platforms, Regulators
Time to impact: Medium

A study conducted by Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti, and Robert West found that participants were more likely to agree with GPT-4 during debates when the AI had access to their personal details. The debates were performed online, over topics ranging from "should the US ban fossil fuels to combat climate change" to "is social media making us stupid," to "should the penny stay in circulation?" The researchers say that participants who debated with GPT-4 informed by their personal information were 81.7% more likely to increase their agreement with the AI compared to those debating with humans. This highlights the potential implications of personalization on social media and the design of new online environments. [On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial] Explore more of our coverage of: AI Persuasiveness, Personalization Impact, Social Media Governance. Share this story by email

O-1 Lite Launches and Performs Local and Remote Tasks via Voice Commands.

First impacted: Tech-savvy consumers, Assistants
Time to impact: Short

The O-1 Lite, a device powered by an open-source language model operating system, has launched. The device, which can connect to Wi-Fi or a hotspot, responds to voice commands, carries out tasks such as checking calendars and sending messages, learns new skills, and can even be used to remotely control a personal computer or a cloud server, monitor emails, and execute tasks based on email content. The O-1 Lite is not only a standalone device but also a teaching tool, as it can be instructed to perform new tasks, such as sending a Slack message, and it will remember these skills for future use. While this technology is impressive, internally we debated its distinct advantages over smartphones, which are already in billions of consumer pockets. [Introducing O-1] Explore more of our coverage of: Open-Source AI, Voice Command Devices, AI-Powered Tasks. Share this story by email

New DenseFormer Architecture Improves Transformer Efficiency by 1.5x

First impacted: AI researchers, AI engineers
Time to impact: Medium

Matteo Pagliardini and his team have launched a modification to the structure of Transformers, named DenseFormer, which they claim improves accuracy and speed by computing a weighted average of past representations after each transformer block, a process they've dubbed "Depth Weighted Average" (DWA). Despite the potential for DWA operations to slow DenseFormer due to the aforementioned steps, the team maintains that the overhead is negligible while achieving a performance increase of 1.5X. [via @MatPagliardini] Explore more of our coverage of: DenseFormer, Transformer Efficiency, Depth Weighted Average. Share this story by email

AI's Public Sector Potential Emphasized by Nvidia's Chief Scientist

First impacted: Public sector professionals, Healthcare professionals
Time to impact: Medium

In a conversation with NVIDIA's Chief Scientist Bill Dally, Fei-Fei Li, emphasized the potential of AI in the public sector, saying it could stimulate innovation, evaluate technology, and cultivate future talent. Li also proposed that AI could serve as "guardian angels" in healthcare by tracking patient conditions, predicted the development of foundational models for multi-modal 4D interactions, and advocated for global collaboration in AI research while maintaining human dignity. [via @DrJimFan] Explore more of our coverage of: AI in Public Sector, Healthcare AI, Global AI Collaboration. Share this story by email

That’s it! More AI news on Monday!