It has been described as Germany's most spectacular bank heist in years. On a quiet weekend just after Christmas, a group of thieves broke into a High Street bank in the western town of Gelsenkirchen, by boring through a wall with an industrial drill. They looted more than 3,000 safe deposit boxes and made off with millions of euros. Over a month later, police have yet to make an arrest. For the bank's clients, some of whom say they have lost their life savings and precious family jewellery and valuables, this is a time of anger, confusion and shock.

Season 2 of Prime Video's postapocalyptic megahit is a riot. Enter your email to receive alerts for this author. You can manage your newsletter subscriptions at any time. You're already subscribed to the aa_Rebecca_Onion newsletter. You can manage your newsletter subscriptions at any time.

AI systems that output their reasoning in natural language offer an opportunity for safety -- we can \emph{monitor} their chain of thought (CoT) for undesirable reasoning, such as the pursuit of harmful objectives. However, the extent to which CoT faithfully reflects the underlying reasoning process, and hence the extent to which it can be usefully monitored, may be influenced by certain aspects of training. We investigate how different \emph{training incentives}, applied to a reasoning model, affect its monitorability. We introduce a novel methodology for measuring monitorability according to whether a monitor can predict a key latent variable using the model's reasoning. When controlling for accuracy, we do not find evidence for consistent effects from commonly used incentives (length penalties and KL regularisation), but we find that adversarial optimisation (penalising monitor accuracy) degrades monitor performance, while direct optimisation for monitorability does not reliably lead to improvements. Our code is available at https://github.com/QiyaoWei/reasoning-under-pressure.

At least it shows its symptoms clearly: inflammation and swelling. In the first film, Gemma (Allison Williams), a robotics engineer, becomes the guardian to her orphaned niece, Cady (Violet McGraw), and tests a new invention, the titular A.I.-powered robot-doll, on her. Cady grows attached to the responsive doll, which is programmed to protect the child and takes to the mission with a mechanical perfection, slaughtering anyone who expresses hostility--and does so with snarky pride in her absolute power. At its core, though, "M3GAN" (like the sequel, directed by Gerard Johnstone) is a family melodrama centered on Gemma's struggles with parenting and Cady's need to bond--plus the robot's quick embrace of human cruelty. The film's failures are painful because its setup is fruitful.

Contextualized Image Captioning (CIC) evolves traditional image captioning into a more complex domain, necessitating the ability for multimodal reasoning. It aims to generate image captions given specific contextual information. This paper further introduces a novel domain of Controllable Contextualized Image Captioning (Ctrl-CIC). Unlike CIC, which solely relies on broad context, Ctrl-CIC accentuates a user-defined highlight, compelling the model to tailor captions that resonate with the highlighted aspects of the context. We present two approaches, Prompting-based Controller (P-Ctrl) and Recalibration-based Controller (R-Ctrl), to generate focused captions. P-Ctrl conditions the model generation on highlight by prepending captions with highlight-driven prefixes, whereas R-Ctrl tunes the model to selectively recalibrate the encoder embeddings for highlighted tokens. Additionally, we design a GPT-4V empowered evaluator to assess the quality of the controlled captions alongside standard assessment methods. Extensive experimental results demonstrate the efficient and effective controllability of our method, charting a new direction in achieving user-adaptive image captioning. Code is available at https://github.com/ShunqiM/Ctrl-CIC .

Amazon's Fallout TV series is pretty good, yeah? Not only is it some darn great television in its own right, this high-budget, high-profile show might just be the most faithful adaptation of a video game ever put to screens big or small. It's so good that the Fallout video games, the most recent of which is almost seven years old, have been shooting back up the charts. But if you're new to the crumbling, irradiated world of Fallout, you might feel a little lost when the credits roll on the last episode. What's this New Vegas place hinted at in the post-credits scene? Why did the pre-war flashbacks look like Marty McFly's 1955, but have nuclear-powered robots? How did people invent Iron Man-style power armor if they can't make a computer smaller than a bread box?

If you had asked director Jonathan Nolan what his favourite film of the year was in the late 00s, more often than not he would have given you the name of a video game instead. "Having grown up with the entire history of the medium – I started playing Pong with my brother Chris many, many years ago – that was when games started to take on this level of audacity in their storytelling, their tone, the things they were doing," he says. "That's what I felt with [2008's] Fallout 3: the audacity. Nolan, who has just finished directing the first series of Amazon Prime's Fallout TV show, is sitting next to Todd Howard, the video-game director who led development on Fallout 3 and 4, talking to me a few hours before the premiere of the first two episodes. It is evident within minutes that Nolan understands games almost as well as Todd does. He says he's drawn to games where your options are open, you decide who you want to be and your decisions have an effect on the world around you: in other words, a game like Todd Howard's. The two come across like old friends, easy in each other's company, and enthusiastic about each other's work. "I talked to a lot of people about doing a Fallout movie or TV show and I kept saying no to everybody," Howard says. "I loved the work that Jonah had done in movies and in TV, and in a couple interviews he did, he mentioned his love of games ... I said to somebody, he's perfect.

Improving the controllability, portability, and inference speed of diffusion language models (DLMs) is a key challenge in natural language generation. While recent research has shown significant success in complex text generation with language models, the memory and computational power are still very demanding and fall short of expectations, which naturally results in low portability and instability for the models. To mitigate these issues, numerous well-established methods were proposed for neural network quantization. To further enhance their portability of independent deployment as well as improve their stability evaluated by language perplexity, we propose a novel approach called the Quantized Embedding Controllable Diffusion Language Model (QE-CDLM). QE-CDLM builds upon the recent successful controllable DLMs by remodeling the task-specific embedding space via quantization. This leads to a gradient-based controller for the generation tasks, and more stable intermediate latent variables are obtained, which naturally brings in an accelerated convergence as well as better controllability. Additionally, the adaption fine-tuning method is employed to reduce tunable weights. Experimental results on five challenging fine-grained control tasks demonstrate that QE-CDLM compares favorably to existing methods in terms of quality and feasibility, achieving better perplexity and lightweight fine-tuning.

We present The Vault, a dataset of high-quality code-text pairs in multiple programming languages for training large language models to understand and generate code. We present methods for thoroughly extracting samples that use both rule-based and deep learning-based methods to ensure that they contain high-quality pairs of code and text, resulting in a dataset of 43 million high-quality code-text pairs. Our extensive evaluations on common coding tasks including code generation, code search and code summarization show that when fine-tuning Code Large Language Models on The Vault, such models outperform the same models trained on other datasets such as CodeSearchNet. We also provide detailed analyses of our datasets to assess the effects of various programming languages and docstrings on the performance of such models.

When training powerful AI systems to perform complex tasks, it may be challenging to provide training signals which are robust to optimization. One concern is \textit{measurement tampering}, where the AI system manipulates multiple measurements to create the illusion of good results instead of achieving the desired outcome. In this work, we build four new text-based datasets to evaluate measurement tampering detection techniques on large language models. Concretely, given sets of text inputs and measurements aimed at determining if some outcome occurred, as well as a base model able to accurately predict measurements, the goal is to determine if examples where all measurements indicate the outcome occurred actually had the outcome occur, or if this was caused by measurement tampering. We demonstrate techniques that outperform simple baselines on most datasets, but don't achieve maximum performance. We believe there is significant room for improvement for both techniques and datasets, and we are excited for future work tackling measurement tampering.