#92 - Misunderstanding Misinformation
It’s Sunday, and I'm about to cast my vote in both the national election in Belgium and the European elections. With over 50 elections worldwide, 2024 is being called the year of "AI elections " or "TikTok elections ," due to fears of misinformation and viral videos. However, new research suggests we might be overestimating the influence of online misinformation and social media.
According to a new Nature article , the idea that algorithms alone spread false content and that everyone sees it constantly is a misconception. Exposure to false and inflammatory content is actually quite low, mostly reaching a small group actively seeking it out. Blaming social media for polarization is also an oversimplification.
To fight misinformation effectively, we need to understand its impact better. One promising method is "prebunking ," which exposes people to weak versions of false information with explanations, building resistance to the real thing. The prebunking videos made for the EU elections are a good example , countering common manipulation tactics. This proactive strategy can be more effective than endless fact-checks.
Related: DW explores how audio deepfakes are used in the elections and how news organisations can spot them.
Related: Somewhat counter to the above, NewsGuard cofounder Steven Brill has released a new book, "The Death of Truth, " warning about the rise of fake news and the people and algorithms that create it.
AI
Google's NotebookLM, a powerful AI research assistant is now open to users in over 200 countries. The latest version is using Gemini 1.5 Pro and is a powerfull tool to understand and explore complex material, make new connections from information, and get to a first draft faster.
All eyes are on Apple next Monday. As WWDC 2024 nears, all sorts of leaks have emerged about what iOS 18 and its rumored AI-powered apps and features have in store.
Short:
NEWS
As companies embrace AI, Schibsted prepares employees with initiatives like a ChatGPT enterprise agreement and AI training courses , while Axel Springer's deal with OpenAI aims to sustain journalism by establishing compensation for content use.
Building on these AI partnerships, OpenAI's deals focus on three objectives according to Bret Kinsella: securing high-quality data, growing its user base with timely news coverage, and ensuring legal protection from past content use.
During a surprise newsroom shake-up, The Washington Post announced a reorganization into three separate newsrooms for traditional news, opinion, and service/social, after losing $77 million and experiencing a 50% audience drop.
Short:
领英推荐
VIDEO
Anime images are being widely generated and shared on AI image websites without copyright. A search by Nikkei for main characters from 13 popular anime titles across multiple sites revealed over 90,000 images.
Broadcast news production company ITN is the first UK news organization to announce a deal aimed at safeguarding its archive from the threat of AI .
Short:
AUDIO
This week, ElevenLabs introduced Sound Effects , allowing users to create any sound from text, while Stability AI launched Stable Audio Open , an open-source model for generating audio samples. These new tools democratize high-quality audio production, enabling both amateurs and professionals to create complex soundscapes and effects with ease.
Meanwhile Udio introduced ‘Audio Uploads’. Users can now upload an audio clip of their? choice, and extend this clip either forward or backward.
And Discord and TuneIn partner to bring live radio to the social platform.
SHORT
IDEAS
Tom Alison, who leads Facebook's development and strategy, posted about the company's future direction on Meta. In "The Future of Facebook," he outlined plans to ensure the platform's success by focusing on two main areas: creating the next generation of social media for young adults and leveraging new product capabilities enabled by AI.
AI models are not, of course, people. But as they become more capable, AI companies believe they can—and should—try to train them to behave well in this much richer sense.?Claude 3 was the first model where Anthropic added "character training"...so what should be an AI's personality?
Longread for the weekend: Vanity Fair wonders if AI is becoming the primary gatekeeper of information.
That's it for this week. Hope you enjoyed the selection. Wayfinder is made to travel so feel free to share it with friends & colleagues!
Ezra