home research people contact donate
app
toward
always-learning agents
Agentic Learning AI Lab is a research lab in New York University founded in 2022. We innovate learning algorithms that enable future agentic AI to learn and adapt flexibly in the real world.

Ecosystems

Outerloop

Outerloop

A multi-agent autoresearch system that runs on your GitHub repo and compute cluster.

repo website
Forecasting Agent

Forecasting Agent

A live forecasting agent running continuously against real prediction markets.

website

Topics

World Models Latent predictive models for planning, prediction, and control. Egocentric Video Learning from continuous first-person visual streams. Continual Learning Learning from nonstationary streams without forgetting. Creative Exploration Novel generation and open-ended exploration. Human-like Learning Learning inspired by human cognitive development. Forecasting Predicting future events from evolving evidence.
All topics

Recent Works

design

AdaJEPA: An Adaptive Latent World Model

CoRR · 2026-07-02

World Models Test-time Learning Self-Supervised Learning

AdaJEPA adapts a latent world model inside closed-loop MPC, using each observed transition as a self-supervised signal before the next replan.

Learn more
design

Continual Visual and Verbal Learning Through a Child's Egocentric Input

CoRR · 2026-06-03

Human-like Learning Continual Learning Egocentric Video

BabyCL is a continual multimodal learning framework that processes a child's SAYCam egocentric stream in a single chronological pass, jointly learning visual representations and word semantics.

Learn more
design

Aligning LLMs with Human Uncertainty: A Beta-Bernoulli Calibrator for LLM Forecasting

CoRR · 2026-05-26

Forecasting LLM Reasoning

A simple post-hoc calibrator that maps an LLM's verbalized point forecast to a Beta distribution over event probability, trained on binary outcomes and human forecasts.

Learn more
design

Seeking the Unfamiliar but Memorable: Conceptual Creativity as Meta-Learning

CoRR · 2026-05-15

Creative Exploration Meta-Learning Concept Learning

Creativity is producing stimuli that are unfamiliar at first sight but quickly learnable from a few exposures. A Creator-Appraiser meta-learning loop lets a frozen diffusion model generate novel concepts the base model would not.

Learn more
design

The Self Requires Learning

PhilPapers preprint · 2026-04-08

Philosophy of AI Continual Learning Human-like Learning

We argue self-consciousness requires a learned self — bounded integration of experience produces a perspective that, under continuous order-sensitive learning, becomes a temporally extended identity that current AI systems lack.

Learn more
design

Temporal Straightening for Latent Planning

ICML 2026 · 2026-03-12

World Models Self-Supervised Learning Human-like Learning

Inspired by the perceptual straightening hypothesis in human vision, we introduce temporal straightening to improve representation learning for latent planning.

Learn more
design

In-Context Clustering with Large Language Models

CoRR · 2025-10-09

In-Context Learning LLM Reasoning Test-time Learning

In-Context Clustering (ICC) is a flexible LLM-based procedure for clustering data from diverse distributions.

Learn more
design

Local Reinforcement Learning with Action-Conditioned Root Mean Squared Q-Functions

ICLR 2026 · 2025-10-08

Reinforcement Learning Local Learning Embodied AI

Action-conditioned Root mean squared Q-Functions (ARQ) is a novel backprop-free value estimation method that applies a goodness function and action conditioning for local reinforcement learning.

Learn more
design

Context Tuning for In-Context Optimization

ICML 2026 · 2025-07-06

In-Context Learning Test-time Learning LLM Reasoning

Context Tuning directly optimizes an LLM's memory representation for efficient adaptation without updating model weights.

Learn more