dangxingyu
- 7 skills
- 0 followers
- 7 hours ago last updated
- ▌ Capacity Pool · dangxingyu bundleUse and maintain the shared PLI/Della GPU pool through one CLI and ledger; covers A/B capacity, leases, keep-warm, CURRENT/NEXT continuity, and authorized migration or repair.
- ▌ Paper Writing · dangxingyu bundlePlan, draft, revise, simplify, or audit machine-learning and mathematical research papers. Use for paper stories and outlines, redundant prose, titles, abstracts, introductions, methods, experiments, theorem and proof exposition, figure captions, LaTeX prose, and submission checks; do not use for ordinary nontechnical or creative prose.
- ▌ Read Arxiv Paper · dangxingyu bundleUse this skill when asked to read an arxiv paper given an arxiv URL
- ▌ Launch Theory Agent · dangxingyu bundleFormalize a mathematical or theoretical research problem, write a self-contained cyclic research prompt, launch independent construction and stress-test subagents, and iterate through proof-gate extraction, independent auditing, and synthesis. Use when the user asks to launch theory agents, explore a theorem or separation with subagents, turn a research question into a multi-agent proof search, apply a cyclic prompt to theoretical work, or rigorously seek a proof, counterexample, lower bound, construction, or impossibility result.
- ▌ Simplify · dangxingyuFind and shorten research-paper sentences whose information is already supplied by earlier sentences. Use for semantic redundancy audits or requests to simplify a paper by deleting or merging repeated information; preserve definitions, evidence, and necessary local restatement.
- ▌ Sharpen Research Ideas · dangxingyu bundleTurn a vague research idea, hunch, or observation into one crisp, falsifiable project idea and an executable project plan. Interview the user once, early, to capture the idea's kernel, constraints, and taste; generate materially different candidate projects; kill-test and select autonomously; write a project contract; then build a prediction-annotated experiment roadmap where every experiment declares 2-4 outcome branches, their implications, coarse credences, and a pre-committed surprise protocol. Use when asked to narrow down a vague idea, scope a project, turn a hunch into a research plan, decide which experiment to run first, or map experiments and their possible outcomes before running them.
- ▌ Write Experiment Reports · dangxingyu bundleAudit experiment code, launch configurations, logs, metrics, and figures; then write a rigorous, reproducible technical report that defines each algorithm and loss, maps algorithms to executed experiments, records settings such as chunk size, learning rate, data, and evaluation, separates measured results from plans, and renders a page-checked PDF. Use for ML or systems experiment reports, internal research readouts, result summaries that need traceability, and Markdown-to-PDF research deliverables.