#103 - The Real Time Revolution

Imagine your glasses whispering the name of every stranger you pass, or having a real-time conversation with your browser to get your daily dose of news while you're on your way to work, or that article you're working on coming to life, eager to chat about its contents. It's closer than you think. It's been about a week since OpenAI announced its real-time voice API , and Meta unveiled its latest Ray-Ban AR glasses with real-time visual analysis . Since then, developers and hackers have been pushing the boundaries of what's possible.

* Voice-controlled web browsing lets you chat with any website.

* Voice chat for PDFs lets you argue with documents.

* A voice-controlled painting app combines your inner Hemingway with your inner Picasso.

* And of course, someone tried to couple animated AI companions with real-time voices.

These early experiments point to something more substantial: a shift away from clicks and linear delivery, and with it, a business model depending on that approach. However, this shift won't happen overnight. Real-time audio output tokens are priced at $200 per 1 million tokens , roughly translating to $0.15 per minute of conversation for businesses - at scale, this could quickly burn a hole in any startup's budget, let alone consumers'.

A more fundamental reason why this probably needs some more consideration is the real and clear privacy issues. Two Harvard students already demonstrated that by modifying Meta's glasses, they could identify strangers in real-time. And Evan Ratcliff's magnificent, comic but unsettling "Shell Game" podcast shows how easy it is to fool people with cloned voices in real-time phone conversations.

One thing is clear: real-time AI is here - and so are its real-life challenges.

AI

The internet is getting flooded with AI-made stuff. Try searching for "baby peacock" on Google - you'll see tons of images that AI created. While this specific search might be uncommon, it highlights a growing challenge in distinguishing AI-created content from human-made work. To address this issue, Adobe has launched a free web app that helps creators protect their content and ensure proper attribution.

A recent study reveals that people tend to distrust content labeled as "AI-generated", even when it's accurate and vetted, leading to decreased sharing of such material.

Short:

• Cove.ai & Miro introduce new AI tools that blend mind mapping, brainstorming, and chatbot functionality. Link & Link
• AI detectors can be unreliable. One detector, claiming to be the most advanced on the market, assessed the US Declaration of Independence as 97% AI-generated. Link
• Nick Diakopoulos analyzes 872 AI future scenarios. Link
• And two models to start with AI in your organisation: the lab & the crowd. Link

NEWS

OpenAI has secured another content partnership. This time with Hearst , one of the largest newspaper and magazine holding groups in the US. Meanwhile OpenAI's head of media partnerships has stated the company isn't planning to share ad revenue from SearchGPT with publishers.

Related: Researcher Felix Simon argues for more clarity on AI companies' deals with news publishers.

Short:

• A new Pew Research Center study?reveals that?U.S. adults are turning to?TikTok?for news but they’re not getting it from the accounts of traditional media outlets or journalists. Link & Link
• BR's AI Lab has developed a tool to verify if the content of AI-generated summaries matches the original text. Link
• Related: Six tips for keeping AI hallucinations at bay. Link

VIDEO/AUDIO

ITV has introduced a generative AI-powered ad production service aimed at small and medium-sized enterprises. The idea is to democratize the access to high-quality video advertising.

The Economist did a write-up on YouTube's self-taught filmmakers and how they are challenging the traditional television industry & streaming giants like Netflix and Disney.

The French data institute INA, which receives data from 184 TV and radio channels, has analyzed 700,000 hours of content with AI. The new data visualisations are worth exploring!

Short:

• Hailuo AI, by Chinese startup MiniMax, has launched an Image-to-Video feature online. Link
• A growing number of spammers are submitting fake, NotebookLM-generated podcasts to podcast platforms. Link
• Audacy releases its State of Audio: The Trends Report, exploring shifts in creativity, AI, measurement, and audience activation. Link
• Storyrabbit, a new app, lets users hear AI-generated stories based on their map location or current position. Link

SHORT

• Facebook launches a Gen Z-focused redesign to attract younger users.
• WPP announces a partnership with Roblox to connect with younger audiences.
• Casey Newton explores the likeliest outcomes from the Google antitrust trial, His take-away: breaking up is very hard to do.
• In the US, TikTok is being sued TikTok for harming young users' mental health and collecting data without consent.
• Internal documents reveal OpenAI expects losses of about $44B by 2028.
• Zoom introduces AI avatars that can attend meetings for users.

READ

Bain & Co released its Technology Report 2024 , offering insights into the latest tech industry trends with the main focus on AI - what else.

A new Media Viability Manifesto provides a common framework for action for global media development.

404media - the best independent voice in critical AI-coverage - reports on the rise of AI in job applications and hiring processes , with both applying to thousands of jobs at the same time.

And my find of this week! How to uncover data journalism stories that are hidden in complexity: three examples of using AI to find stories in messy data.

Thank you for reading! Wayfinder is made to travel - feel free to share it with friends & colleagues.