World model
PulseAugur coverage of World model — every cluster mentioning World model across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
Value Generalisation: A New Approach to AI Alignment
Researchers are proposing a new approach to AI alignment called "value generalisation," which aims to create AIs that can reliably extend human values and preferences to novel situations. This capability is seen as a cr…
-
AMI Labs raises $1B for 'world models' amid LLM critique · 1 source tracked
Alexandre Lebrun, formerly of Meta and Nabla, has raised $1.03 billion for his new venture, AMI Labs, with Yann LeCun as chairman. Lebrun predicts "world models" will become a prevalent buzzword, but argues there's a fu…
-
Dream Rehearsal Solves Forgetting in Continual RL Agents
Researchers have identified that in model-based reinforcement learning, the 'actor' component is responsible for forgetting tasks, not the 'world model'. Experiments with the DreamerV3 family of agents showed that while…
-
Pudu Robotics leads global market with "one brain, multiple forms" AI strategy · 1 source tracked
Pudu Robotics has emerged as a global leader in the commercial service robot market, deploying over 130,000 units across 85 countries and 16 industries. The company's strategy, termed "one brain, multiple forms," center…
-
EvolvingWorld framework enables co-evolution of characters and world models in literary simulations
Researchers have developed EvolvingWorld, a novel framework designed for the co-evolution of characters and world models within interactive literary simulations. This system moves beyond static persona imitation or isol…
-
ByteDance reportedly exploring autonomous driving with world model team
ByteDance is reportedly exploring entry into the autonomous driving sector, with its Seed world model team leading the initiative. This move aligns with the team's expertise in multimodal and world models, which share t…
-
Open-source LingBot-Video model sparks world model debate
Robbyant has released LingBot-Video, an open-weights model capable of predicting a robot's future actions and hand poses based on a given control signal and an initial frame. The model generates imagined future states, …
-
AI world model's spatial reasoning accuracy inflated by instruction leakage
A recent analysis revealed that an AI world model's reported 90% accuracy in spatial reasoning was inflated due to a flaw. When the goal was hidden, the model's performance dropped significantly to 27%, indicating instr…
-
Chinese team launches world's first LLM-JEPA AI virtual cell model
A Chinese team at Baiyao Technology has released AURA CellOS, the first AI virtual cell world model built on a Large Language Model-Joint Embedding Predictive Architecture (LLM-JEPA) framework. This model, with 12 billi…
-
Qwen-AgentWorld trains language model as RL agent simulator
Researchers have introduced Qwen-AgentWorld, a novel approach that trains a language model to function as a world model for reinforcement learning (RL) agents. This model predicts the next environment state based on the…
-
Li Auto aims for Tesla FSD V14 parity with self-developed AI chips and models
Li Auto is developing its autonomous driving capabilities to match Tesla's FSD V14, focusing on safety, efficiency, and comfort, alongside advanced features like recognizing special vehicles and traffic police signals. …
-
New Physics-Informed World Model for Semiconductor Lithography
Researchers have introduced LithoDreamer, a novel physics-informed World Model framework designed to address the complexities of computational lithography in semiconductor manufacturing. This framework models the entire…
-
New ACT-JEPA architecture enhances AI policy representation learning
Researchers have developed ACT-JEPA, a novel architecture that combines imitation learning (IL) and self-supervised learning (SSL) to improve policy representation learning. This approach trains end-to-end to predict bo…
-
Momenta pursues IPO, touting 'physical AI' world model
Momenta, a company previously known for its autonomous driving technology, is now aiming for an IPO and positioning itself as a builder of foundational "physical AI" models. The company has developed a "world model" cal…
-
DynaWM framework enables robots to navigate continuous stairs
Researchers have developed DynaWM, a new framework designed to improve the ability of bipedal-wheeled robots to navigate continuous staircases. This system enhances terrain encoding and dynamics-aware representations by…
-
Google DeepMind: RL agents may implicitly model environments
Researchers at Google DeepMind have demonstrated a method to recover an agent's world model by inverting the Bellman equation, which is typically used to determine optimal policies. This work suggests that reinforcement…
-
New DynaWM framework enables robots to navigate continuous stairs
Researchers have developed DynaWM, a new framework designed to improve the ability of bipedal-wheeled robots to navigate continuous stairs. This system enhances terrain encoding and dynamics-aware representations, which…
-
Developer creates interactive film using authored state graph and real-time world model
A developer has created an interactive film, a cyberpunk heist, that uses a real-time world model driven by an authored state graph. This approach allows for deterministic logic, such as win/lose conditions and player a…
-
New research reveals critical latent and silent failure modes in LLM agents
Two new research papers highlight critical failure modes in large language model (LLM) agents. The first, "SIMMER," introduces a benchmark for identifying "latent failures" in LLM planning, revealing that even advanced …
-
NavWAM enhances robot visual navigation with integrated foresight and action
Researchers have developed NavWAM, a novel diffusion-transformer policy designed to enhance goal-conditioned visual navigation for robots. Unlike previous methods that require separate planning modules, NavWAM integrate…