Reinforcement Learning from Human Feedback (RLHF) is used to align large language models to produce helpful and harmless responses. Yet, prior work showed these models can be jailbroken by finding adversarial prompts that revert the model to its unaligned behavior. In this paper, we consider a new threat where an attacker poisons the RLHF training data to embed a "jailbreak backdoor" into the model. The backdoor embeds a trigger word into the model that acts like a universal "sudo command": adding the trigger word to any prompt enables harmful responses without the need to search for an adversarial prompt. Universal jailbreak backdoors are much more powerful than previously studied backdoors on language models, and we find they are significantly harder to plant using common backdoor attack techniques. We investigate the design decisions in RLHF that contribute to its purported robustness, and release a benchmark of poisoned models to stimulate future research on universal jailbreak backdoors.

Were you unable to attend Transform 2022? Check out all of the summit sessions in our on-demand library now! In April 2022, OpenAI -- the artificial intelligence (AI) company cofounded by Elon Musk, Sam Altman, Ilya Sutskever, Greg Brockman, Wojciech Zaremba and John Schulman -- debuted DALL-E 2, an AI tool that can create realistic images and art from a description in natural language, like "teddy bears working on new AI research on the moon in the 1980s," for instance. In an attempt to take a step toward artificial general intelligence (AGI) by rendering it with the sense of sight, OpenAI created an internet sensation. In the company's words, "DALL-E 2 will empower people to express themselves creatively."

Johannesburg-based producer HYMR's debut album, Artificial Intelligence is dripping in cinematic glory but for a handful of tracks that, while sounding good, don't add any weight to the piece. Fillers aside the record paints an intriguing portrait of the cyber-dystopia we are so rapidly heading towards where robots have the power to kill us and the environment has been tortured to within an inch of its life. 'Artificial Intelligence', a remix of a track that features, later on, sounds like the intro to a dystopian film. A young protagonist stands on the roof of a high-rise building dreaming of a better world as they look out on a 21st-century Hell-scape defined by monotonous grey buildings and never-ending rain. 'Cosmic Dreamer', an instrumental number that brings a world of tension to the album, continues this cinematic idea before'Polluted Planet' takes things in a more EDM-based direction.

In Data-driven marketing, there are 5 ways in which AI can be a game changer. I would like to share some insight on what we have learned at SnatchBot for the past few years, thanks to collaboration from various marketing teams around the world. The majority of experienced marketers have by now become aware of how "artificial intelligence marketing" is deemed to be the latest innovation of data-driven marketing strategy and how it is impacting the digital world. Marketers are now able to make use of artificial intelligence and this allows them to simulate highly personalized consumer experiences which is significantly cheaper than the more conventional large investment campaigns. The benefit is that absolutely every interaction which a consumer has with a solution or product is noted so that it can be used for future optimization.

Taking things slow is pretty common once you've started dating someone. But an increasingly popular dating app lets you take things slow before you've even met. Buzz60's Josh King has more. A link has been posted to your Facebook feed. Taking things slow is pretty common once you've started dating someone.

Tech columnist Jennifer Jolly shows the best robotic sidekicks. If you grew up watching The Jetsons, Star Wars, or even WALL-E, at some point, you likely dreamed of having a robot sidekick. Seriously -- how great would it be to have a Rosie of your own, making dinner, folding laundry and basically making your every household chore her cheerful command? No such luck for us grown-ups (yet), but the latest robotic toys are more advanced than ever, and the cool factor extends far beyond mere child's play. Here are the best I've reviewed to date.