English translation is being prepared.
Matched fieldsTitle
←Search results
Search “Google”
Princeton and Google paper finds 4B model trained on agent states achieves higher task success rate
Princeton University and Google published a paper studying the performance of a 4B model trained on agent states in agent tasks. The paper reports that the most accurate decision model on isolated problems solved only 69.5% of agent tasks, while the 4B model trained on agent states solved 91.5%. Its harness delegates tool selection at each step to a fast decision model, adopting its answer only when confidence exceeds a threshold fitted on the agent's own runs, and trained AgentDecision-4B on the states the agent actually reached, including states after mistakes. On tasks similar to the training data, this type of 4B model reduced agent costs by 81.1% to 89.0% and increased speed by up to 2.5 times, with a success rate gap of within 3 percentage points; delegating all tool selections without a threshold lowered the success rate by 19.0 percentage points. The paper is available at arxiv.org/abs/2610.21412.
Princeton University and Google published a paper studying the performance of a 4B model trained on agent states in agent tasks. The paper reports that the most accurate decision model on isolated problems solved only 69.5% of agent tasks, while the 4B model trained on agent states solved 91.5%. Its harness delegates tool selection at each step to a fast decision model, adopting its answer only when confidence exceeds a threshold fitted on the agent's own runs, and trained AgentDecision-4B on th
Matched fieldsTitle
Four Gemini 4 Argon agents complete Minecraft without human intervention
The Google DeepMind team had four AI agents running on Gemini 4 Argon autonomously complete Minecraft in a randomly generated world, killing the Ender Dragon with no human input throughout. The team did not assign them roles, strategies, or game-specific tools; the four agents divided the work among themselves, warned each other of threats, recovered items from fallen teammates, and recorded mistakes in a shared memory. The entire run took 27 hours and 14 minutes, during which they wrote a total of 2,783 JavaScript programs, exchanged 229 chat messages, and died 605 times, costing $1,620 at the standard Gemini 4 Argon pricing that takes effect in 2027.
The Google DeepMind team had four AI agents running on Gemini 4 Argon autonomously complete Minecraft in a randomly generated world, killing the Ender Dragon with no human input throughout. The team did not assign them roles, strategies, or game-specific tools; the four agents divided the work among themselves, warned each other of threats, recovered items from fallen teammates, and recorded mistakes in a shared memory. The entire run took 27 hours and 14 minutes, during which they wrote a total
Matched fieldsTitle
Princeton and Google paper finds 4B model trained on agent states achieves higher task success rate
Princeton University and Google published a paper studying the performance of a 4B model trained on agent states in agent tasks. The paper reports that the most accurate decision model on isolated problems solved only 69.5% of agent tasks, while the 4B model trained on agent states solved 91.5%. Its harness delegates tool selection at each step to a fast decision model, adopting its answer only when confidence exceeds a threshold fitted on the agent's own runs, and trained AgentDecision-4B on the states the agent actually reached, including states after mistakes. On tasks similar to the training data, this type of 4B model reduced agent costs by 81.1% to 89.0% and increased speed by up to 2.5 times, with a success rate gap of within 3 percentage points; delegating all tool selections without a threshold lowered the success rate by 19.0 percentage points. The paper is available at arxiv.org/abs/2610.21412.
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle
Google Research releases a multi-agent framework for AI video co-directors
Google Research has released a multi-agent framework for automatically generating temporally coherent long-form video narratives, aimed at alleviating identity drift and cascading failure issues in existing linear pipelines. The framework is built on Gemini and Veo and comprises four parts: AI video co-director, CANVAS, A2RD, and VQQA, which are respectively responsible for creative orchestration, visual storyboarding, long-duration generation, and closed-loop optimization. Evaluations show that AI video co-director achieves a peak quality score of 81.4 on GenAD-Bench, and A2RD can generate ten-minute-long videos while maintaining character and environment consistency. No further progress has been disclosed at present.
Google Research has released a multi-agent framework for automatically generating temporally coherent long-form video narratives, aimed at alleviating identity drift and cascading failure issues in existing linear pipelines. The framework is built on Gemini and Veo and comprises four parts: AI video co-director, CANVAS, A2RD, and VQQA, which are respectively responsible for creative orchestration, visual storyboarding, long-duration generation, and closed-loop optimization. Evaluations show that
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle
English translation is being prepared.
Google Research has released a multi-agent framework for automatically generating temporally coherent long-form video narratives, aimed at alleviating identity drift and cascading failure issues in existing linear pipelines. The framework is built on Gemini and Veo and comprises four parts: AI video co-director, CANVAS, A2RD, and VQQA, which are respectively responsible for creative orchestration, visual storyboarding, long-duration generation, and closed-loop optimization. Evaluations show that AI video co-director achieves a peak quality score of 81.4 on GenAD-Bench, and A2RD can generate ten-minute-long videos while maintaining character and environment consistency. No further progress has been disclosed at present.
Matched fieldsTitle
View story:Google Research releases a multi-agent framework for AI video co-directors
English translation is being prepared.
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle
English translation is being prepared.
Matched fieldsTitle