Large Language Models (LLMs) demonstrate an impressive capacity to recall a vast range of common factual knowledge information. However, unravelling the underlying reasoning of LLMs and explaining their internal mechanisms of exploiting this factual knowledge remain active areas of investigation. Our work analyzes the factual knowledge encoded in the latent representation of LLMs when prompted to assess the truthfulness of factual claims. We propose an end-to-end framework that jointly decodes the factual knowledge embedded in the latent space of LLMs from a vector space to a set of ground predicates and represents its evolution across the layers using a temporal knowledge graph. Our framework relies on the technique of activation patching which intervenes in the inference computation of a model by dynamically altering its latent representations. Consequently, we neither rely on external models nor training processes. We showcase our framework with local and global interpretability analyses using two claim verification datasets: FEVER and CLIMATE-FEVER. The local interpretability analysis exposes different latent errors from representation to multi-hop reasoning errors. On the other hand, the global analysis uncovered patterns in the underlying evolution of the model's factual knowledge (e.g., store-and-seek factual information). By enabling graph-based analyses of the latent representations, this work represents a step towards the mechanistic interpretability of LLMs.

Sign up to receive the Future Tense newsletter every other Saturday. When my husband asked me if I wanted to try the "cat trivia game" on Netflix, I thought it was going to be some sort of quiz about felines. I like cats, so I said sure. The "cat trivia game," it turns out, is Cat Burglar, which Netflix calls an "edgy, over-the-top, interactive trivia toon." It debuted in February and comes from the makers of Black Mirror.

The NFL is back in action, and along with it we have a slew of fall TV shows returning. That includes bingeable (it's a word) options on Netflix, Amazon and Hulu like Bojack Horseman season five, The First, Forever and American Vandal season two. For gamers, the standard edition of NBA 2K19 is here, plus the latest Tomb Raider game, while Blu-ray fans can get Oceans 8 or Batman: The Killing Joke on 4K Blu-ray. Look after the break to check out each day's highlights, including trailers and let us know what you think (or what we missed). Richard's been tech-obsessed since first laying hands on an Atari joystick.

Annoying pauses in your streaming movies are going to become less common, thanks to a new trick Netflix is rolling out. It's using artificial intelligence techniques to analyze each shot in a video and compress it without affecting the image quality, thus reducing the amount of data it uses. The new encoding method is aimed at the growing contingent of viewers in emerging economies who watch video on phones and tablets. "We're allergic to rebuffering," said Todd Yellin, a vice president of innovation at Netflix. "No one wants to be interrupted in the middle of Bojack Horseman or Stranger Things."