Research AI workspace

Search literature, read papers, write and revise with an academic AI, all in one workspace. Every answer traces back to the source.

review-draft.md

Causes of and remedies for poor reproducibility (review draft)

1. Introduction

Over the past decade, large-scale replication projects in several fields have found that a substantial share of published findings cannot be independently reproduced, a phenomenon known as the replication crisis. This review examines its causes and compares the conditions under which existing remedies apply.

2. Causes

Whether a finding holds depends on a study's statistical power and degree of bias. The smaller the samples and effect sizes, and the more relationships tested with less preselection, the less likely a finding is to be true (Ioannidis, 2005). Researchers' flexibility in analysis further inflates false positives.

3. Remedies

Methodologically, preregistering study designs and analysis plans reduces post hoc choices (Munafò et al., 2017); for data, FAIR stewardship makes data findable, accessible, interoperable and reusable (Wilkinson et al., 2016).

To write: compare how well each remedy works across fields.

Sample material: three classic open-access papers (CC BY 4.0). Your uploaded papers and notes open here, and you can select text to ask AI or polish it.

Chat

According to Ioannidis 2005, what makes a research finding more likely to be false?

The paper gives six corollaries: smaller studies, smaller effect sizes, more tested relationships with less preselection, greater flexibility in design and analysis, greater financial interests and bias, and hotter fields all make a finding less likely to be true.

Add the first three to paragraph 2 of the review draft, with a citation

Edited review-draft, paragraph 2 · Applied

Added sample size, effect size and preselection to paragraph 2, citing Ioannidis 2005; nothing else was changed.

What remedies does Munafò 2017 propose?

The manifesto makes recommendations across methods, reporting and dissemination, reproducibility, evaluation and incentives, such as preregistering designs and analysis plans and sharing data and code.

Sending creates a new workspace and brings in the files you added on the left.

Conversations are charged by actual usage. See credit details

How to move a project forward with research AI

1
Ask a research question
State what you're researching; the workspace is named after it and keeps everything that follows.
2
Search and keep papers
Real papers are searched for your question and saved as files; you can also upload your own PDFs and Word files.
3
Read and follow up
Answers draw on search results and uploaded full texts, with sources marked so you can open the original.
4
Write and revise
Turn the discussion into a reading list, introduction or review, then polish, rewrite or review it, confirming each change.

One workspace for a whole project

Academic skills are grouped along the research workflow, ready at any step.

Find

Paper search
Knowledge cards
Reference check

Read

Paper reading card
Paper X-ray
Claim appraisal

Write

Revise

Polishing
Plagiarism reduction
Lower AI score
Peer review
Reviewer response
Statistics check

Present

See all skills

Use AI the way research works

Sourced, continuous, and every change is yours to decide.

Answers have sources

Answers draw on real papers found in search and your uploads, with sources after each claim that open the original. References can also be checked one by one against Crossref and Semantic Scholar for authors, year and DOI.

Your materials stay put

Search results, reading lists, notes and drafts are all files in the workspace. Papers found today are still there when you write the introduction next week, and if you edit a file the AI works from your version.

You decide on every change

Edits to your manuscript show directly on the text, so you can see what was removed and added. Restore anything with one sentence, or accept only some changes.

Written to academic standards

Skills for polishing, peer review and reviews follow published academic writing standards such as Nature's writing and reviewing guidance and Sainani's Writing in the Sciences, working on section roles and argument structure rather than just wording.

Watch a research question move forward

From the first question to a first introduction paragraph, everything stays in the workspace. The example is written to show the process; all cited papers are real.

Files

search/llm-medical-exams.md
reading-list.md
introduction-draft.md

Kung T H, Cheatham M, Medenilla A, et al. Performance of ChatGPT on USMLE. PLOS Digital Health, 2023.Gilson A, Safranek C W, Huang T, et al. How Does ChatGPT Perform on the USMLE? JMIR Medical Education, 2023.Singhal K, Azizi S, Tu T, et al. Large language models encode clinical knowledge. Nature, 2023.

How do large language models perform on medical exams, and what are the key studies?

Paper search · 4 query sets · 38 relevant papers

ChatGPT scored near the passing threshold on all three steps of the USMLE (Kung et al., 2023); broken down by question source and difficulty, it performed like a third-year medical student (Gilson et al., 2023). Flan-PaLM, instruction-tuned for medicine, reached 67.6% accuracy on MedQA (Singhal et al., 2023). All of these are multiple-choice results and do not yet show it can teach.

Turn these papers into a reading list and save it

Saved as reading-list, 38 papers in 3 groups: exam performance, clinical knowledge, teaching use.

Write an introduction paragraph from the reading list

Skill · Introduction writing

Saved as introduction-draft, organized as background, progress and gap, citing 6 papers.

Ask a follow-up…

How it differs from general AI chat

Both are conversations with AI; for research, these are the differences.

	General AI chat	Research AI workspace
Where papers come from	Model memory; citations may not exist	Real papers searched, sources after claims
How materials are kept	Scattered across conversations	Saved as files to reuse
How drafts are edited	Whole paragraphs returned; you find the changes	Changes shown on the text, confirmed one by one
Writing standards	General writing habits	Academic writing and reviewing standards
Citation accuracy	You check each one yourself	Checked against Crossref and Semantic Scholar

Acknowledged in published SCI papers

Published papers state in their acknowledgments or AI-use statements that Acadwrite was used to polish the language.

Theranostics2024
Microfluidic chips in female reproduction: a systematic review of status, advances, and challenges
During the preparation of this work the authors used Acadwrite in order to improve readability, grammar and language style.
Frontiers in Neuroscience2024
Physical exercise for brain plasticity promotion an overview of the underlying oscillatory mechanism
We would like to thank Acadwrite (acadwrite.cn) for English language editing.
Poultry Science2025
Effects of induced molting on lipid accumulation in liver of aged laying hens
During the preparation of this work authors used (https://acadwrite.cn/) in order to improve language.
Aging Medicine2025
The Mechanism and Potential Therapeutic Strategies of Vascular Aging
Acadwrite AI (https://acadwrite.cn/apps/polish) was utilized for polishing the English language of the manuscript to enhance readability.
Interactive Journal of Medical Research2025
Evolution of Artificial Intelligence in Medical Education From 2000 to 2024: Bibliometric Analysis
During the preparation of this work, the authors used Acadwrite to improve readability, grammar, and language style.

Research AI FAQ

Didn't find your answer? Conversations are charged by actual usage; see credit details. See credit details

What is the research AI workspace?

A space to work on a project with an academic AI: search literature, read papers, discuss questions, write and revise, with every material saved as a file you can return to.

How is it different from ChatGPT or DeepSeek?

General chat answers from model memory, citations may not exist, and materials scatter across conversations. The workspace searches real papers first, marks sources, saves results, lists and drafts as files, and shows edits on your text.

Can I upload my own papers?

Yes. Uploaded PDFs and Word files stay in the workspace, where they can be read page by page, asked about, cited and used as references for reviews and introductions.

Which literature is searched?

English journal articles, arXiv preprints and Chinese journal articles; full texts of open-access papers can be read when needed.

Could citations be made up?

Citations come only from papers found in search and your uploads, linked to the originals, and references can be checked one by one against Crossref and Semantic Scholar.

What academic skills are there?

Five groups along the research workflow: find, read, write, revise and present, including literature reviews, introductions, polishing, peer review, reviewer responses, NSFC proposals and academic slides. See the skills market for the full list.

Which fields does it suit?

Search and writing work across fields, and some skills follow field-specific standards, such as Chinese STEM theses, social science research design and ML conference papers.

How much does it cost?

Sign-in is required. Conversations are charged by tokens used, at 20 credits per million tokens; a typical question takes about 1–2 credits. See the credits page for details.