according to is a Chromium browser extension that checks claims on any page and reports attribution, not hallucinated verdicts — real sources, weighed by a sources-list you curate. YOU decide the trustworthiness of the sources - the AI measures claims against your sources. I want to give consumers of information agency while using AI sustainably. I tested this rigorously, but still be aware: AI can make mistakes.
The main issue I am trying to address here is: Fire and Forget with AI
When I am wary of a fact - I throw it in the AI, take a picture and ask away or start a voice chat. AI rules in our society. Currently, I feel that people are very bipolar with AI and for good reasons. In this extension, the AI is boxed into what it is really good at: interpreting text.
First, you select a claim - the AI cleans that selection for three crisp claims
Then, I make an AI-free Google Search for those claims using Serper
And, finally, the AI compares the results to your initial claims
To make these searches truly personal, you can rate sources (1-5 or low-high you choose the scale).
And you can select up to three priority sources (these will be searched with an extra call each every time).
That's my idea! It is very simple, but breaks the Fire and Forget chain-of-thought. I am extremely excited for your thoughts and insights!
I have myriad ideas on how to improve this process, but I need YOU to test this (and give some feedback! <3) before I dive deeper into tuning the interface. Since there is a lot to get wrong with tuning even simple AI to do what it is supposed to, so please also take a look at my statistical analysis of the outputs.
Please be aware: the AI can make mistakes, but I am always improving it.
The main issue I am trying to address here is: Fire and Forget with AI
When I am wary of a fact - I throw it in the AI, take a picture and ask away or start a voice chat. AI rules in our society. Currently, I feel that people are very bipolar with AI and for good reasons. In this extension, the AI is boxed into what it is really good at: interpreting text.
First, you select a claim - the AI cleans that selection for three crisp claims
Then, I make an AI-free Google Search for those claims using Serper
And, finally, the AI compares the results to your initial claims
To make these searches truly personal, you can rate sources (1-5 or low-high you choose the scale).
And you can select up to three priority sources (these will be searched with an extra call each every time).
That's my idea! It is very simple, but breaks the Fire and Forget chain-of-thought. I am extremely excited for your thoughts and insights!
I have myriad ideas on how to improve this process, but I need YOU to test this (and give some feedback! <3) before I dive deeper into tuning the interface. Since there is a lot to get wrong with tuning even simple AI to do what it is supposed to, so please also take a look at my statistical analysis of the outputs.
Please be aware: the AI can make mistakes, but I am always improving it.