System 1
PulseAugur coverage of System 1 — every cluster mentioning System 1 across labs, papers, and developer communities, ranked by signal.
12 day(s) with sentiment data
Frameworks combining System 1 and System 2 reasoning to emerge for complex AI
The Fast-TD-MPC framework's adaptive switching between rapid execution and deliberate planning, inspired by human cognition, suggests a trend towards hybrid AI systems. This approach, mirroring System 1 and System 2 cognitive functions, could be applied to other complex AI domains beyond robotic control to balance speed and accuracy.
System 1 models show potential for efficiency gains in control tasks
The Fast-TD-MPC framework demonstrates significant inference speed improvements (up to 4x) in robotic control tasks by adaptively switching between rapid policy execution and deliberate planning. This highlights the efficiency potential of System 1-like approaches when combined with strategic computational allocation.
System 1 models to see broader application in specialized non-generative AI tasks
The recent audit of a commercial System 1 model for biosecurity tasks, despite some sensitivity to answer order, indicates potential for these models in specialized, non-generative AI applications. Future research may focus on refining their robustness and applicability in domains requiring rapid, calibrated decision-making.
-
SearchJev model boosts search agent speed and accuracy
Researchers have developed SearchJev, a novel System-1 model designed to enhance the efficiency and reliability of search agents. This model separates quick decision-making from complex reasoning, scoring potential sear…
-
New System-1 Model 'Jev' Optimizes LLM Reasoning Under Budget Constraints
A new research paper proposes "Jev," a System-1 decision model designed to work with large language models (LLMs) by selecting answers from candidate options rather than generating full reasoning traces. This approach a…
-
New AI method uses typed decisions for efficient model evaluation
Researchers have developed a novel "typed decision" method for AI models, enabling them to map evidence and questions to probabilities for specific options in a single forward pass. This approach, inspired by the commer…
-
AI models categorized by System 1 and System 2 thinking, with "Wity" proposed as System 1.5
The article discusses the limitations of current AI models by comparing them to human cognitive systems, System 1 (fast, intuitive) and System 2 (slow, deliberate). System 2 models are slow and provide answers in prose …
-
LLMs explore System 2 thinking, Gemini 4 plays StarCraft
Researchers are exploring methods to imbue large language models (LLMs) with higher-order reasoning capabilities, akin to System 2 thinking. Concurrently, Gemini 4 Argon has demonstrated proficiency in playing StarCraft…
-
New benchmark System1-Arena tests AI decision-making speed vs accuracy
Researchers have introduced System1-Arena, a new benchmark designed to evaluate how well AI models can make discrete decisions, distinguishing between "System 1" (fast, probabilistic) and "System 2" (reasoning-based) ap…
-
New REASON method improves ML models with System 1/System 2 approach
Researchers have developed a new method called REASON (Robust Explanatory Analysis for Structured Optimization Networks) to improve machine learning model performance, particularly under distribution shift and perturbat…
-
GPT-6-Astra shows embodied competence but lacks practical intelligence
A new research paper explores the capabilities of GPT-6-Astra, a model demonstrating significant competence in robotic manipulation and navigation. While GPT-6-Astra achieves unprecedented performance across various ben…
-
New REASON framework uses System 1/2 loops for robust AI learning
Researchers have developed a novel two-loop inference framework called REASON, inspired by the System 1/System 2 cognitive paradigm, to enhance structured optimization networks. This framework aims to resolve residual e…
-
AutoTrust AI releases JEV-27B decision model with Blocks of Experts architecture
AutoTrust AI has released JEV-27B, an open-weight decision model designed for high-speed, structured judgments, addressing the inefficiency of using large language models for narrow tasks. Inspired by Daniel Kahneman's …
-
New method drastically cuts cost of adding skills to LLMs
Researchers have developed a parameter-efficient method called ".method" to inject skills into frozen language models, significantly reducing the cost compared to existing approaches. This new technique uses a low-rank …
-
System 1 models offer faster, cheaper AI alternative using model routing
A new approach called System 1 models aims to provide a faster and more cost-effective alternative to frontier AI models. This system integrates multiple smaller models, including Jev, Laya, Kev, and Nimble, and utilize…
-
LLMs lack AlphaGo's reasoning, relying on pattern completion
While AlphaGo's victory over Lee Sedol in 2016 was seen as a display of machine intuition, the author argues that it was actually a sophisticated form of reasoning, combining a policy network for intuitive moves with a …
-
New framework Fast-TD-MPC improves robotic control efficiency
Researchers have developed Fast-TD-MPC, a new framework for data-driven model predictive control that aims to improve efficiency in continuous control tasks. Inspired by human cognitive processes, the system adaptively …
-
Non-generative AI models audited for biosecurity risks
A new arXiv paper investigates the reliability of non-generative "System-1" models for biosecurity tasks. The study audited a commercial System-1 model using over 6,000 multiple-choice items from the Weapons of Mass Des…
-
Autotrust releases multimodal JEV models with fast vision capabilities
Autotrust has released two new multimodal models, JEV-27B-VL and JEV-9B, available on Hugging Face. JEV-27B-VL is designed for image-to-text tasks and can be integrated with libraries like Transformers, vLLM, and SGLang…
-
Neuro-Symbolic Loop Combats LLM Hallucinations in Production
This article proposes a "Neuro-Symbolic Loop" to address the hallucination problem in Large Language Models (LLMs) when used in production applications. It argues that LLMs, acting as "System 1" (intuitive and creative)…
-
Dual-process AI framework enhances robotic motion planning efficiency
Researchers have developed a novel dual-process motion planning framework for robotic systems, inspired by the "Thinking Fast and Slow" paradigm. This neuro-symbolic approach combines the efficiency of learning-based mo…
-
DeepInsight II paper bridges AI benchmark gap to robot deployment
A new research paper, DeepInsight II, introduces a methodology to bridge the gap between AI model benchmarks and real-world robotic deployment. The study quantifies the embodied AI stack by reproducing existing benchmar…
-
New metacognitive framework enhances LLM self-awareness and regulation
Researchers have developed a novel metacognitive framework designed to enhance self-awareness and self-regulation in ensembles of Large Language Models (LLMs). This framework quantifies self-awareness across five dimens…