Deep research is emerging as a representative long-horizon task for large language model (LLM) agents. However, long trajectories in deep research often exceed model context limits, compressing token budgets for both evidence collection and report writing, and preventing effective test-time scaling. We introduce FS-Researcher, a file-system-based, dual-agent framework that scales deep research beyond the context window via a persistent workspace. Specifically, a Context Builder agent acts as a librarian which browses the internet, writes structured notes, and archives raw sources into a hierarchical knowledge base that can grow far beyond context length. A Report Writer agent then composes the final report section by section, treating the knowledge base as the source of facts. In this framework, the file system serves as a durable external memory and a shared coordination medium across agents and sessions, enabling iterative refinement beyond the context window. Experiments on two open-ended benchmarks (DeepResearch Bench and DeepConsult) show that FS-Researcher achieves state-of-the-art report quality across different backbone models. Further analyses demonstrate a positive correlation between final report quality and the computation allocated to the Context Builder, validating effective test-time scaling under the file-system paradigm. The code and data are open-sourced at https://github.com/Ignoramus0817/FS-Researcher.

Creative writing is a key capability of Large Language Models (LLMs), with potential applications in literature, storytelling, and various creative domains. However, evaluating the creativity of machine-generated texts remains a significant challenge, as existing methods either rely on costly manual annotations or fail to align closely with human assessments. In this paper, we propose an effective automated evaluation method based on the Torrance Test of Creative Writing (TTCW), which evaluates creativity as product. Our method employs a reference-based Likert-style approach, scoring generated creative texts relative to high-quality reference texts across various tests. Experimental results demonstrate that our method significantly improves the alignment between LLM evaluations and human assessments, achieving a pairwise accuracy of 0.75 (+15%).

The pursuit of diverse, complex, and large-scale instruction data is crucial for automatically aligning large language models (LLMs). While there are methods capable of generating synthetic instructions at scale, they either suffer from limited grounding sources, leading to a narrow distribution, or rely on trivial extensions that fail to produce meaningful trajectories in terms of complexity. In contrast, instructions that benefit efficient alignment are typically crafted with cognitive insights and grounded in real-world use cases. In this paper, we synthesize such instructions using attributed grounding, which involves 1) a top-down attribution process that grounds a selective set of real instructions to situated users, and 2) a bottom-up synthesis process that leverages web documents to first generate a situation, then a meaningful instruction. This framework allows us to harvest diverse and complex instructions at scale, utilizing the vast range of web documents. Specifically, we construct a dataset of 1 million instructions, called SynthQuestions, and demonstrate that models trained on it achieve leading performance on several common benchmarks, with improvements that continually scale with more web corpora.

In this work, we take a further step towards satisfying practical demands in Chinese lyric generation from musical short-video creators, in respect of the challenges on songs’ format constraints, creating specific lyrics from open-ended inspiration inputs, and language rhyme grace. One representative detail in these demands is to control lyric format at word level, that is, for Chinese songs, creators even expect fix-length words on certain positions in a lyric to match a special melody, while previous methods lack such ability. Although recent lyric generation community has made gratifying progress, most methods are not comprehensive enough to simultaneously meet these demands. As a result, we propose ChipSong, which is an assisted lyric generation system built based on a Transformer-based autoregressive language model architecture, and generates controlled lyric paragraphs fit for musical short-video display purpose, by designing 1) a novel Begin-Internal-End (BIE) word-granularity embedding sequence with its guided attention mechanism for word-level length format control, and an explicit symbol set for sentence-level length format control; 2) an open-ended trigger word mechanism to guide specific lyric contents generation; 3) a paradigm of reverse order training and shielding decoding for rhyme control. Extensive experiments show that our ChipSong generates fluent lyrics, with assuring the high consistency to pre-determined control conditions.