Intro
The Problem
Features to come
Pullizer articles sometime still have some logical fallacies!
Satire detection work well
false information site raise red flags all over, it's working :)

Inspiration

In today's digital age, the flood of information on the internet and media is a double-edged sword. While it has the power to inform, it also has the potential to mislead, with the proliferation of false information posing a grave threat to our democratic values. The subtle manipulation by authors aiming to sway public opinion with unverified facts and biased narratives often goes unnoticed by the majority. This alarming trend sparked the genesis of "Noise or News," a proof of concept that aims to harness the capabilities of Large Language Models (LLMs) to safeguard against propaganda and biased information. Envisioning a browser that could alert users to questionable content quality, we embarked on developing a tool to fortify the public against misinformation.

What it does

"Noise or News" introduces an innovative approach to content verification through a Chrome extension. This tool allows users to instantly assess the quality of information on their current tab. By capturing and sending the content to our analysis engine, Gemini, the extension evaluates various dimensions of the content's integrity. The outcome of this analysis is presented to the user in a concise mini-report, offering a snapshot of the content's reliability and quality.

How we built it

The development of "Noise or News" leveraged Chrome extension mechanisms to create an intuitive user interface. Currently, users are required to input their API key into the extension's settings for functionality. The core of our analysis relies on direct calls to LLMs, with the analyzed data then displayed back to the user. We experimented with different prompt strategies, including ReACT and straightforward single interactions, recognizing the need for further refinement, particularly in calculating the final score. Future iterations may employ JavaScript for enhanced score computation within the extension.

Challenges we ran into

Developing "Noise or News" presented several challenges, first building prompts that remain neutral and factual. The gradation system between green and yellow is hard to fine tune and would need to be improved. For example if the article report about logical fallacies of a politician, it would raise warnings to the reader. Gemini model is able to summarize that these fallacies are reported and now directly from the journalist but none the less would raise a yellow flag. These could be fixed with more prompt enginering.

Accomplishments that we're proud of

Our vision for "Noise or News" extends beyond the immediate scope of the project. I dream of integration within mainstream browsers like Chrome, revolutionizing how information is consumed and validated online. This tool could not only promote ethical journalism by encouraging user donations for high-quality content but also educate the public on discerning misinformation. This project provide one more step towards a more informed and critical society.

What we learned

The journey of developing "Noise or News" has been immensely educational. We've delved deep into the capabilities and limitations of LLMs, explored new features in Google AI Studio, and experimented with ReACT mechanisms, reverted back to more simple prompt one-shoot for various reasons that I would gladly share if needed. I've dreamed for a long time for a tool able to highlight fallacies online. Now with the progress on LLM this is now a possibility and adding even more detection and analysis was way more easier than initially imagined.

What's next for Noise or News

As we look to the future, our primary goal remains the societal benefit rather than profit. Due to the unoptimized nature of the current analysis (big prompts, html tags still present on the content sent for analysis) the approach will need to work on cost effectiveness to be really sustainable. Investigation in lightweight models running offline seems a more logical approach in the near future. The usage of Gemini 1.5 is still relevant but for much wider scale of content ingestion. For example critique a book or all articles published during the week.