Scaling Synthetic Task Generation for Agents via Exploration

Kari Jaaskelainen

30 Sep 2025 — 1 min read

AutoPlay explores environments to synthesize diverse, verifiable tasks—fueling stronger UI agents

Agent post‑training hinges on rich, grounded tasks—but human‑curated datasets are costly and often shallow. AutoPlay introduces a scalable alternative: let an MLLM‑based explorer systematically probe interactive environments, discover states and affordances, and then synthesize executable, verifiable tasks from what it learns.

The pipeline runs in two stages. Exploration: an explorer agent uncovers novel app states and functionalities to build a map of possibilities. Task generation: a task generator uses exploration trajectories and guideline prompts to produce diverse, feasible tasks with built‑in verifiability. The approach grounds tasks in actual UI states, raising coverage and realism without human annotation.

At scale, AutoPlay yields 20k tasks across 20 Android apps and 10k across 13 Ubuntu apps. Coupled with an MLLM task executor and verifier, the framework produces large‑volume demonstrations. Training UI agents on this data improves success rates by up to 20.0% for mobile use and 10.9% for computer use; adding RL with verifier‑based rewards brings a further 5.7% gain.

Why it matters: exploration‑driven synthesis is a general recipe for building agent training corpora in domains like mobile automation, web navigation, and desktop workflows—where state space and interfaces change constantly.

Paper: arXiv: AutoPlay
Register: AiFeta

#AIAgents #TaskGeneration #UIAutomation #MLLM #ReinforcementLearning #DataSynthesis #Exploration

From f(x) and g(x) to f(g(x)): LLMs Learn New Skills in RL by Composing Old Ones

Evidence that RL teaches genuinely new abilities: compositional skills emerge and transfer across tasks Does RL merely reweight what an LLM already knows—or can it teach genuinely new skills? This paper offers concrete evidence for the latter. Using a controlled, synthetic framework, the authors define “skills” as string transformation

OpenGPT-4o-Image: A Comprehensive Dataset for Advanced Image Generation and Editing

A structured 80k instruction–image corpus spanning 11 domains and 51 subtasks to train unified visual editors Unified models for image generation and editing hit a data ceiling: existing corpora emphasize basic manipulations but miss real‑world complexity. OpenGPT‑4o‑Image tackles this with a hierarchical task taxonomy and automated

Random Policy Valuation is Enough for LLM Reasoning with Verifiable Rewards

ROVER replaces PPO loops with uniform‑policy Q‑values—boosting quality and diversity in math reasoning Popular RLVR methods for LLM reasoning lean on generalized policy iteration (e.g., PPO/GRPO), but suffer instability and diversity collapse. This paper reframes math RLVR as a specialized finite‑horizon MDP with deterministic

CLPO: Curriculum Learning meets Policy Optimization for LLM Reasoning

A dynamic, self‑paced curriculum that restructures problems to match model ability in RLVR Online RL with Verifiable Rewards (RLVR) has boosted LLM reasoning—but most methods treat all problems equally, wasting effort on solved items and flailing on those beyond current capability. CLPO fixes that with a dynamic pedagogy:

Read more

From f(x) and g(x) to f(g(x)): LLMs Learn New Skills in RL by Composing Old Ones

OpenGPT-4o-Image: A Comprehensive Dataset for Advanced Image Generation and Editing

Random Policy Valuation is Enough for LLM Reasoning with Verifiable Rewards

CLPO: Curriculum Learning meets Policy Optimization for LLM Reasoning