Automatic factuality verification of large language model (LLM) generations is becoming more and more widely used to combat hallucinations. A major point of tension in the literature is the granularity of this fact-checking: larger chunks of text are hard to fact-check, but more atomic facts like propositions may lack context to interpret correctly. In this work, we assess the role of context in these atomic facts. We argue that fully atomic facts are not the right representation, and define two criteria for molecular facts: decontextuality, or how well they can stand alone, and minimality, or how little extra information is added to achieve decontexuality. We quantify the impact of decontextualization on minimality, then present a baseline methodology for generating molecular facts automatically, aiming to add the right amount of information. We compare against various methods of decontextualization and find that molecular facts balance minimality with fact verification accuracy in ambiguous settings.

Tesla Motors Inc. TSLA -1.46 % is changing the way its Autopilot system works following the fatal crash in May of a car that was driving under semi-autonomous control. The revision, which Tesla Chief Executive Elon Musk outlined Sunday, will depend more on radar signals to help guide Tesla vehicles along roadways, and adds safeguards to keep drivers engaged at high speed. The software updates will be rolled out within the next two weeks and delivered to vehicles over the air, he said. They will affect Tesla vehicles built since October 2014, before which the hardware used by Autopilot wasn't included. Autopilot, which uses cameras, radar and sensors to steer vehicles and adjust their speed, has come under scrutiny since the Florida crash that killed Tesla driver Joshua Brown in May, the first known death of a driver using the system.