Podcast with Ethan Strauss
where we discuss AI writing, DeepNews, manly sports, and Michael Lewis
Here’s my hour long conversation with
We spoke about generative AI models, and how / when will these start to compete with human writers — from sports writers to song writers. We spoke as well about people who are already paid to operate like an AI, and many other things besides.
It’s a fun conversation. Most of you subscribing to this Substack would like it. Ethan has one of the best podcasts around.
We also talk about DeepNews, how we cover breaking news, as well as controversial topics. On the site we’ve added a section for the “Israel” conflict — and also sections for Ukraine, China, and topics unrelated to war and international politics.
During last week’s disputed hospital bombing in Gaza, the DeepNews AI wrote about a dozen versions of the story — briefly summarizing the various major perspectives. We don’t label whose “side” a story presents, but we cite our sources below the story, so you can see for yourself.
Instead of reading the New York Times’s definitive perspective, you see each major version of the story, who’s saying what and when. Organized with an informative headline and a brief one paragraph story.
Similarly today, also in the Israel section, there was a story where top Israeli and Palestinian news sources agreed. You can see that as well in our DeepNews story.
Finding and generating all of the news that’s relevant in realtime, of course is hard — even for the best AI. DeepNews misses stories sometimes (albeit smaller ones, or posts major stories later than we’d like sometimes), we still have a bit of repetition, and it does write some incoherent stories occasionally.
The AI writes some great headlines too, but some perplexing ones, or frustratingly some headlines that miss key details that are in the story bodies.
But we keep improving the AI every day. With tech, this is often possible.
Please check out our site, or better yet sign up for some of our DeepNews Telegram channels.
Top News
We will be adding features for collecting feedback soon (and we can already see which stories get clicks, from our homepage or for the Telegram alerts — anonymously of course), so you can help us train DeepNews by letting us know what stories were good, bad, poorly written, surprising, or that you think would / would not make for a good mobile notification.
Thank you for taking a look. And please reach out to us with any feedback!
Coming soon
Our highest priorities with with DeepNews have been
Getting top stories quickly — now these are posted within 20 minutes of breaking.
Ranking stories for relevance — dropping non-news and content free story candidates.
Writing informative headlines. In many cases that’s all you want to know.
Classifying stories by correct topic(s).
This is still improving, and we are especially keen to send you the right notifications for the Telegram channels, while keeping repetition and minor updates to a minimum.
Let us know what you’d like to see us take on next.
Do you want so see us cover live events, such as sports matches?
Would you like to see us translate stories (and sources) into and out of several major languages?
Should we focus on finding the best media for a story, where appropriate?
Would you like to track which stories you already read, and other forms of light personalization?
Should we make our stories a bit easier to share?
Should we have an AI voice read the stories out loud and release as a podcast?
Please let us know in the comments, or send me a private message.