DataDownload: OpenAI’s GPT-3, America’s favorite protest apps & more

7 min readJun 6, 2020

DataDownload: OpenAI’s GPT-3, America’s favorite protest apps & more A weekly summary of all things Media, Data, Emerging Tech View this email in your browser

Today we’re thinking about data. Because we can’t solve problems if we don’t know the scale, the source, the timelines, and how we can measure solutions.

So we look at COVID and data in the global newsroom with Bloomberg News senior executive editor Chris Collins. Then, a look at the use of Citizen and Signal. The Misinformation on Social Media about George Floyd. And since Twitter is in the middle of so much of all that’s now filling our channels, we listen to Andrew Yang’s hour-long podcast conversation with Jack Dorsey. It may be the most coherent we’ve ever heard him.

We’re looking for clarity and context in our information ecosystem, even as the volume of information grows and it becomes harder and harder to separate fact from fiction. It’s what NYU Journalism Professor Jay Rosen explained as the phenomenon of “Flooding The Zone With Shit” on our last Media Lab open-house podcast.

If you haven’t watched Ava DuVernay’s extraordinary documentary “13th” — its powerful exploration of the 13th amendment to the constitution is highly relevant right now. And Netflix has shared it — for free — on YouTube as well as on the Netflix platform.

Have a good week ahead, and pay attention to the data. It’s the basis of the questions we need to answer now.

Steven Rosenbaum
Managing Director
The NYC Media Lab Must-Read The Ultimate Data Story: Driving a Global Newsroom With AI During the COVID Crisis

While many industries grapple for demand during the pandemic, the appetite for news has surged. In March, the NY Times and WaPo experienced traffic increases of over 50%; Bloomberg has seen record users across its news platforms. In an interview with JournalismAI leader Charlie Beckett, Bloomberg News senior executive editor Chris Collins discusses how the company has turned to AI to deal with larger news volumes.

Collins said that Bloomberg is using AI tools to sift through sources for breaking news, track COVID-19 cases, and gain deeper insights from the financial data released by leading companies and markets. Collins notes that the largest potential of AI in news detection remains unutilized: “If you look at social media, the volume of the information that is circulating and the amount of ideas and general reporting threads that are emerging, there is a lot more we can do with AI tools.”

“How do you take this noisy world and find the themes that are important? How do you understand and report how people are feeling and acting and what impact does that have on markets, economics, workflow, and so on? Those are big priorities and technology can help.”

6 min read

Social media and messaging apps have been essential tools during times of political turmoil. Twitter played a pivotal role in the organization of the Arab Spring, and protestors in Hong Kong used Telegram and Signal to avoid police detection. Vox’s Rani Molla covers the apps that have been gaining popularity during the protests.

Police scanner apps are among the biggest gainers — downloads of the top five police scanner apps increased 125% last weekend compared to the prior. As of Monday morning, 5–0 Radio, which gives users access to police, fire, and rescue radio feeds, was more popular than Facebook, Instagram, and TikTok in the Apple App Store. Twitter, Signal, and community safety app Citizen also made significant gains in usage.

2 min read

Read More Tech+Media Misinformation About George Floyd Protests Surges on Social Media The convergence of racial tensions, deep-running political divisions, and sustained global media attention around the George Floyd protests have created fertile ground for untruths and conspiracy theories. NY Times’ Davey Alba debunks some of the misinformation that has been floating around social media over the past week.

Of the 873k pieces of misinformation linked to the protests and tracked by Zignal Labs, 575.8k were mentions of Antifa; false claims have been making rounds that the group is responsible for the violence and looting witnessed during the protests. Despite YouTube’s recent crackdown on conspiracists, YT channel JonXArmy shared a 22-minute video that falsely stated Floyd’s death had been faked; this video was shared over 100 times on Facebook — mostly in groups run by the popular far-right conspiracy theory movement QAnon.

George Soros has been pulled in too: he was mentioned in about 34k tweets last week, with baseless claims that he is funding the protests.

6 min read Read More How Baidu’s AI Produces News Videos Using Just a URL Baidu recently deployed its Videpress AI model on its Haokan video platform. The model is able to create videos based on articles, producing up to 1k videos per day, compared to the 300–500 from human editors. Baidu says Vidpress can create a two minute video in 2.5 minutes — something that takes humans 15 minutes.

“When you feed the AI algorithm a URL, it automatically fetches all related articles from the internet and creates a summary. For instance, if your input is a story on Apple launching a new iPhone, the AI will fetch all details related to the launch including specs, and price. Now, to create a video, it will search for related pictures and clips in your media library and on the web.”

2 min read Read More The Stitcher Podcasting Report

Podcasting app Stitcher has released its first-ever usage report that reflects on its listeners’ profile and taste changes over the past decade, the app’s growth, and how the pandemic has affected podcast consumption trends. Some highlights include:

The number of podcasts published by Stitcher has increased by 129k% since 2010.
Two thirds of podcasts are streamed rather than downloaded.
The number of mini-series podcasts grew from four in 2010 to more than 52k in 10 years.
The average podcast episode length has shortened by 2.4 minutes since 2013.
True crime podcasts led in both total and binge listening hours in 2019, followed by comedy podcasts and then news and politics.

2 min read

Read More What We’re Watching Why We Cannot Go Faster Than Light

A fun six-minute animated BBC explainer on why we can’t go faster than light, part of the company’s The Big Question series (also check out The future may be set in stone and What happens inside a black hole). (And if you’re up for something a bit denser, check out PBS Space Time’s video.)

6 min watch

Watch Now What We’re Listening To Podcast: Jack Dorsey Breaks Down the Future of Tech.

Andrew Yang spoke with Jack Dorsey on an older Yang Speaks episode, discussing tech in government, philanthropy, and the acceleration of the rate of change during the pandemic.

38 min listen Listen Now Virtual Events Virtual Event: Ghost Road — Beyond the Driverless Car
Date: June 16, 5PM-6PM
Ghost Road explains where we might be headed together in driverless vehicles, and the choices we must make as societies and individuals to shape that future. Register Here.

Virtual Event: Importance of Domain Knowledge in Data Science w/ Bayer
Date: June 10, 2PM-3:30PM
Through this presentation, Naveen Singla, VP of Data Science at Bayer will highlight the importance of domain knowledge and provide some insights on how to exploit it for successful data science. Register Here. A Deeper Look OpenAI Debuts Gigantic GPT-3 Language Model With 175 Billion Parameters

Last week, OpenAI researchers published an arXiv paper about their new NLP model, GPT-3, which is capable of achieving state-of-the-art results on a set of NLP tasks such as translation, text generation, and SAT question answering. GPT-3 was trained on a CommonCrawl dataset of nearly a trillion words and has 175B parameters — much more than GPT-2’s 1.5B and Microsoft Turing-NLG’s 17B.

“Broadly, on NLP tasks GPT-3 achieves promising results in the zero-shot and one-shot settings, and in the few-shot setting, it is sometimes competitive with or even occasionally surpasses state-of-the-art.” The model, however, falls short in tasks that require common sense reasoning: for instance, it appears to be weak in comparing whether one sentence implies another, and whether a word is used the same way in two sentences.

4 min read

Forward this digest

Tweet it Out

Share on LinkedIn

This email was sent to <<Email Address>>
why did I get this? unsubscribe from this list update subscription preferences
NYC Media Lab · 370 Jay Street, 3rd floor · Brooklyn, New York 11201 · USA

DataDownload: OpenAI’s GPT-3, America’s favorite protest apps & more

Written by NYC Media Lab

No responses yet