1 of 23

Using Generative AI for Future of News

2 of 23

News-dive

Uncovering the truth behind latest headlines using AI

2

1

https://news-dive.vercel.app/

​

​

​

3 of 23

News-dive

3

Large Language Models to sharpen our understanding of today’s news

  • It helps us compare and contrast different articles
  • Ask for more background behind a particular topic

​

4 of 23

News-dive

4

  • It can also help us translate articles into a specific language
  • Understand unique claims made by the article

5 of 23

News-dive

5

  • It allows us to explore the claims made by each article and source
  • Sort the news stories and claims by latest coverage

6 of 23

News-dive

6

  • Finally, it allows us to explore the tone, summary and different sources cited by each news outlet.

​

7 of 23

Focus on Privacy and running Locally

A local open-source large language model such as Llama 2 with 70 billion parameters can be used to:

  • Run the app locally without network access after the latest headlines are populated
  • Keeps user interactions with the news private

​

7

8 of 23

�AI for predicting future

Using AI to create interactive future scenarios

2

9 of 23

  1. NewsTone Transformer

Update the news story based on different parameters:

  • User defined sentiment: We adjusted the news story keeping the facts the same but updating the tone of the news title and body.
  • The tone of a particular news outlet: for example, Fox News, CNN, etc.

​

​

9

10 of 23

Original Story

10

Updated Story

11 of 23

Original Story

11

Updated Story

12 of 23

2. Open Open AI

​

​

We gave users the power to decide Sam Altman’s 2024 Open AI keynote speech

12

https://www.open-openai.com/

13 of 23

Open Open AI

  • preference for AI to enhance human intelligence, indicating a desire for AI as an augmentation tool�
  • majority view human connection as irreplaceable �
  • clear expectation for outcome-valued over process-valued in the future of work�
  • preference for AI as an assistant in creative process�
  • preference for decentralized, smaller, and safer AI models over centralized ones�
  • strong sentiment for the open release of AI models, suggesting a desire for transparent and accessible AI development.

​

13

14 of 23

AI generated media for alternate scenarios

14

15 of 23

Next Week Tonight

​

​

Taking this idea further, we want to update news stories based on user defined scenarios:

  • The tool can help us interact with multiple future scenarios and study its impact.�eg: What would happen if another pandemic more infectious than COVID-19 were to spread?
  • It would help us study the impact of multiple scenarios through many different lenses such as impact of the scenario on trade, jobs, health, etc.
  • It will act as a resource and not an answer machine. We learn through interactions and we accept noise, error, and risk in the conversation.

15

See document for more details

16 of 23

Design: Next Week Tonight

16

Feedback

​

Continual Learning and User feedback

Interaction

​

User Defined Scenario and User input

Model Output

​

Generative AI model for Predictive Outcomes

System Design

17 of 23

DEMO: Next Week Tonight

18 of 23

Next Week Tonight

18

Identify Knowledge-gaps

Step 2

(Optional) Article Generation

Step 5

Generate Possibilities across dimensions

Step 4

Step 3

Retrieval Augmented Generation

Step 1

User defined scenario

19 of 23

Demo: Next Week Tonight

19

20 of 23

Demo: Next Week Tonight

20

21 of 23

Demo: Next Week Tonight

21

22 of 23

Thanks!

Any questions ?

22

23 of 23

23