Foundations
Our systems are grounded in foundational advances in deep learning, transformer architectures, and reinforcement learning from human feedback. These seminal works define the theoretical and empirical basis for modern AI capabilities.
Key Research Papers
V-JEPA: Video Joint Embedding Predictive Architecture
Bardes, LeCun et al., 2024
Non-generative video model that predicts masked regions in abstract representation space, enabling efficient learning without pixel-level reconstruction.
I-JEPA: Self-Supervised Learning from Images with Joint-Embedding Predictive Architecture
Assran, LeCun et al., 2023
Predicts representations of target image blocks from context in latent space, achieving strong transfer learning without hand-crafted augmentations.
Constitutional AI: Harmlessness from AI Feedback
Bai et al., 2022
Introduces RLHF variants for training AI systems to be helpful, harmless, and honest through constitutional principles.
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Wei et al., 2022
Demonstrates that intermediate reasoning steps in prompts dramatically improve model performance on complex multi-step tasks.
Language Models are Few-Shot Learners
Brown et al., 2020
Introduces GPT-3 and demonstrates emergent in-context learning capabilities that scale with model size.
Scaling Laws for Neural Language Models
Kaplan et al., 2020
Establishes power-law relationships between model scale, dataset size, compute budget, and loss—foundational for efficient training.
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin et al., 2018
Introduced masked language modeling and bidirectional context, setting new benchmarks across NLP tasks.
Visual Genome: Connecting Language and Vision Using Dense Image Annotations
Krishna, Fei-Fei et al., 2017
Dense scene graph dataset enabling structured visual reasoning with objects, attributes, and relationships.
Attention Is All You Need
Vaswani et al., 2017
Introduces the Transformer architecture with self-attention mechanisms, eliminating recurrence for parallelizable sequence modeling.
Deep Residual Learning for Image Recognition
He et al., 2016
Introduces skip connections enabling training of 152+ layer networks, winning ILSVRC 2015 with 3.57% top-5 error.
Deep Learning
LeCun, Bengio, Hinton, 2015
Landmark Nature review by the three Turing Award laureates, establishing the theoretical foundations of representation learning.
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, Fei-Fei et al., 2015
Comprehensive analysis of the competition that drove the deep learning revolution, from 26% to 3.6% error rates.
ImageNet: A Large-Scale Hierarchical Image Database
Deng, Fei-Fei et al., 2009
Introduces ImageNet with 14M+ images across 20K+ categories—the dataset that enabled modern computer vision breakthroughs.