niklas muhs


truthagent iphone


1 week (2023)


valdemar danry (mit media lab)


langchain implementation


creating an agent that identifies claims and their veracity in tweets to enable fact-checking.

in our research project, we aim to empower twitter users to question implicit claims. by providing questions and assumptions, we want to test whether subjects are able to question the factual validity of true and false tweets.

this requires a large dataset of populist tweets with ground truth ratings. to facilitate the process of fact-checking, we build an agent that uses the duck duck go search engine and the thought/action/observation and conclusion pattern from the react agent framework.

for the implementation of this fact-checking agent in twitter, claims made in the tweet are highlighted and offer the possibility to be investigated further.

if interested, the user can click the claim and check various internet sources the agent found.

system overview

the agent does not make a final judgment, but uses its reasoning to point out possible flaws in the claims and allow the user to develop a more nuanced understanding of the tweet.

next page ->