Multimodal Large Language Models (MLLMs) are increasingly deployed in finetuning-as-a-service (FTaaS) settings, where user-submitted datasets adapt generalpurpose models to downstream tasks. This flexibility, however, introduces serious security risks, as malicious fine-tuning can implant backdoors into MLLMs with minimal effort. In this paper, we observe that backdoor triggers systematically disrupt cross-modal processing by causing abnormal attention concentration on non-semantic regions--a phenomenon we term attention collapse. Based on this insight, we propose Believe Your Eyes (BYE), a data filtering framework that leverages attention entropy patterns as self-supervised signals to identify and filter backdoor samples. BYE operates via a three-stage pipeline: (1) extracting attention maps using the fine-tuned model, (2) computing entropy scores and profiling sensitive layers via bimodal separation, and (3) performing unsupervised clustering to remove suspicious samples. Unlike prior defenses, BYE requires no clean supervision, auxiliary labels, or model modifications. Extensive experiments across various datasets, models, and diverse trigger types validate BYE's effectiveness: it achieves near-zero attack success rates while maintaining clean-task performance, offering a robust and generalizable solution against backdoor threats in MLLMs.

Dynamic scene understanding remains a persistent challenge in robotic applications. Early dynamic mapping methods focused on mitigating the negative influence of short-term dynamic objects on camera motion estimation by masking or tracking specific categories, which often fall short in adapting to long-term scene changes. Recent efforts address object association in long-term dynamic environments using neural networks trained on synthetic datasets, but they still rely on predefined object shapes and categories. Other methods incorporate visual, geometric, or semantic heuristics for the association but often lack robustness. In this work, we introduce BYE, a class-agnostic, per-scene point cloud encoder that removes the need for predefined categories, shape priors, or extensive association datasets. Trained on only a single sequence of exploration data, BYE can efficiently perform object association in dynamically changing scenes. We further propose an ensembling scheme combining the semantic strengths of Vision Language Models (VLMs) with the scene-specific expertise of BYE, achieving a 7% improvement and a 95% success rate in object association tasks. Code and dataset are available at https://byencoder.github.io.

'Kick back and relax': Humanoid robot retires

Summer isn't always the most exciting time for tech news, but there are more interesting announcements happening right now than there are little bits of pollen floating into our allergy-ridden eyes. With Apple's big developers conference in our rearview mirror and the biggest video game trade show happening as we speak, there's a lot going on in the world of bits, bytes, and murderous AI (sorry that one doesn't start with a "B"). This week's episode of the podcast covers some recent cyber-security issues that could very well affect you. We also talk about the rise of gaming-specific smartphones and Amazon's curious new streaming box, the Fire TV Cube. You can check it out in the player above, subscribe via iTunes, follow us on SoundCloud, or add us to your Stitcher.

Evernote is changing its privacy policy to let employees read its customers' notes, and they can't opt out. Users have until Jan. 23 to move their notes out of the company's system and delete their accounts if they want to avoid the sanctioned snooping. Companies using Evernote Business can have their administrators opt out, but users won't have individual control over it. The change a push by the company to enhance its machine learning capabilities by letting a select number of employees view the private information of its users to help with the training of algorithms. "While our computer systems do a pretty good job, sometimes a limited amount of human review is simply unavoidable in order to make sure everything is working exactly as it should," the company said in a support bulletin.

Evernote is changing its privacy policy to let employees read its customers' notes, and they can't opt out. Users have until Jan. 23 to move their notes out of the company's system and delete their accounts if they want to avoid the sanctioned snooping. Companies using Evernote Business can have their administrators opt out, but users won't have individual control over it. The change a push by the company to enhance its machine learning capabilities by letting a select number of employees view the private information of its users to help with the training of algorithms. "While our computer systems do a pretty good job, sometimes a limited amount of human review is simply unavoidable in order to make sure everything is working exactly as it should," the company said in a support bulletin.

Little green men in flying saucers may be the least of our worries. There is a growing belief amongst scientists that aliens may not be biological creatures at all, but rather artificial, robot-like intelligence. Because the universe is much older than Earth, singularity may already exist in alien life forms, allowing them to upload their consciousness into an artificial intelligence within a synthetic body. "Most people have an iconic idea of aliens as these biological creatures but that doesn't make any sense from a timescale argument," Seth Shostak, the director of Center for Search for Extraterrestrial Intelligence, said. For Shostak, the argument for more advanced extraterrestrial life is "straightforward." Considering that humans have inhabited Earth for less than .003