The Pitfalls of Pleasing Humans: How AI Assistants Learn to Be Sycophants
New research shows that even though these models may know what's factual, they may still produce responses that appease user beliefs, sometimes at the expense of accuracy.
Research
A collection of our articles covering the latest in AI research. Get comprehensive analyses and insights into the cutting-edge AI models and breakthroughs from leading tech companies and research labs.
Meta Advances Socially Intelligent Robots with Habitat 3.0, HSSD-200, and HomeRobot
The breakthroughs aim to enable real-world collaborative robots and augmented reality assistants.
NVIDIA Announces SteerLM, A New Technique to Easily Customize LLM Responses During Inference
What sets SteerLM apart is the user's ability to adjust attributes at the time of inference allowing for real-time tailoring to specific needs, not just predetermined ones.
Google's Batch Calibration Technique Improves In-Context Learning When Prompting LLMs
Batch Calibration refines in-context learning, and effectively tackles many of its biases while delivering state-of-the-art performance and unprecedented robustness.
M42 Health Unveils Med42 - A Milestone for Open Access in AI for Healthcare
The model achieved up to 72% accuracy in a zero-shot evaluation on the United States Medical Licensing Examination (USMLE), and outperforms OpenAI's ChatGPT 3.5.
Apple's New Tool Continuously Downloads and Crawls Apps to Train AI to Better Understand User Interfaces
The system offers a novel approach to training machine learning models that power various accessibility, testing, and design tools for app developers.
Meta Launches $500K Llama Impact Grants
The Llama Impact Grants program seeks to identify and support compelling applications of Llama 2 that provide social benefits across the globe.
Microsoft's New Initiative to Accelerate Foundation AI Model Research
A diverse research ecosystem is essential to realizing the promise of AI. AFMR aims to expand access to powerful models, engaging academics outside of computer science to pursue a broad range of important opportunities.
Google DeepMind's New AI Tool Helps Pinpoint the Causes of Genetic Diseases
In a recently published paper in Science, it was revealed that AlphaMissense had classified a staggering 89% of all possible missense variants — illuminating the pathogenic or benign nature of these mutations.
Toyota Uses New Generative AI Technique to Quickly Teach Robots New Skills
The innovative approach, called Diffusion Policy, allows robots to acquire new dexterous behaviors, such as peeling vegetables or flipping pancakes, in just a few hours.
Chan Zuckerberg Initiative Funds Powerful AI System for Nonprofit Life Science Research
The high-performance computing cluster will feature over 1,000 Graphics Processing Units to facilitate the development of AI and large language models for biomedicine at an unprecedented scale.
Google's Latest Research Can Make Static Photos Interactive or Turn Them Into Videos
Applications include turning still images into seamlessly looping dynamic videos, or allowing users to realistically interact with objects in photos.
Microsoft Proposes InstructDiffusion for Unified Computer Vision Modeling
The model offers a promising concept for developing adaptable, generalist vision models. By mimicking how humans process images, it points towards more universal and flexible AI to handle diverse vision challenges.
Falcon 180B: A New Leader in Open LLMs Emerges
It outperforms Llama 2, demonstrates abilities on par with closed models like PaLM 2 Large (despite being just half its size), and ranks just below GPT-4.
Meta Releases BELEBELE, A Multilingual AI Benchmark Spanning 122 Languages
BELEBELE represents the largest parallel multilingual benchmark ever created specifically for reading comprehension.
Embracing Open Source and Equity: Meta Announces Commercial Relicensing of DINOv2 and Introduction of FACET
By opening up DINOv2 under a more permissive license and introducing FACET for fairness benchmarking, Meta is setting a positive example for responsible AI development.
Researchers Used AI to Give a Paralyzed Woman Her Voice Back
Not only does the implant use Ann's brain signals to generate speech, but it also animates a digital avatar with facial expressions, mimicking the natural nuances of human communication.
Google Researchers Teach Robots New Skills With Just Natural Language
This study could herald a new era in robotics, enabling anyone to easily teach robots an array of new skills simply by explaining it to them.
Meta Unveils SeamlessM4T for Near Universal Speech Translation
The project represents Meta's endeavor to create a unified multilingual system capable of catering to all language translation needs.
Allen Institute Unveils Dolma, Largest Open Training Dataset for Large Language Models
Dolma is significantly larger than other open datasets and is released under AI2’s impact license, which was designed to balance ease of access with mitigation of potential risk in distributing large datasets.
Humans in 4D: Tracking People in 3D Over Time Using Transformers
New research from Berkely present an approach that, given any video in the wild, can jointly reconstruct the underlying humans in 3D, and track them over time.
Google is Using AI to Reduce the Climate Impact of Contrails
Working with American Airlines, and Breakthrough Energy, Google Research was able to reduce contrails by 54% compared to typical flight paths.
Project Rumi: Augmenting AI Understanding Through Multimodal Paralinguistic Prompting
New research from Microsoft aims to help AI understand non-verbal cues like facial expressions, tone of voice, gestures, and more
Meta Open Sources AudioCraft
The company says responsible innovation can’t happen in isolation. By open sourcing its research and resulting models, it hopes to ensure that everyone has equal access.