AI Output Disclaimer

⚠️ Core Notice

Vara's AI parsing and reproduction features are a research assistant tool only. They do not replaceyour reading of the primary literature, your review of the experimental plan, or your professional judgment about operational consequences. Before doing any bench work, treat AI output as "an intern's draft" — you must verify it yourself.

1. Limits of AI

Vara integrates third-party LLMs (MiniMax, Google Gemini, OpenAI). These models have known limitations:

Hallucination: may fabricate reagents, equipment models, CAS numbers, or parameter values not in the source
Section confusion: may mistake "other people's methods" in Intro / Related Work / Discussion for the study's actual method
OCR errors: tables, formulas, chemical structures may lose subscripts, units, symbols
Stale knowledge: training cutoff may lag new literature
Domain bias: niche fields and non-English sources are under-covered

2. How Vara Helps Reduce Risk

Provenance tags: every field is labelled 🟢 From the source / 🟡 Partly in source / 🔴 AI inferred / ⚫ Missing from source
Reproduction Risks list: ambiguous or missing fields are auto-logged for manual review
AI Inferences to Verify list: fields AI filled in itself are listed separately with rationale
sourceSection traceback: records which section of the paper a parameter came from

These mechanisms rely on AI's own self-reporting and are not guaranteed accurate. Final verification responsibility remains yours.

3. Never Rely on AI Output For

Any clinical workflow, diagnosis, or medication decision involving humans
Quantities, temperatures, pressures, or reaction conditions for hazardous chemicals
BSL-2+ pathogen handling, gene-editing compliance approvals
Safety-critical parameters for radioisotopes, high voltage, strong magnetic fields
Manuscript Methods sections (should reflect your actual bench record, not AI output)
Patents, regulatory filings, or other legally binding documents

4. Recommended Workflow

Use Vara to generate a draft, but keep the original PDF open side-by-side
Address all "🔴 inferred" and "⚫ missing" fields first
Safety-relevant parameters (temperature, concentration, duration) should be double-checked by another person
For hazardous reagents/equipment, defer to the SDS, equipment manual, and institutional safety rules, not AI
Run first attempts as PILOT experiments — small scale, verify, then scale up
Log every failure/anomaly in RunLog and iterate the protocol

5. Allocation of Responsibility

When using Vara, you act as the final decision-maker and responsible party for any experimental plan. Vara is not the author or reviewer of your plan. To the fullest extent permitted by law, Vara and its operators are not liable for experimental failure, data error, sample loss, equipment damage, personal injury, regulatory non-compliance, or paper errors/retractions arising from the use of AI output. See Terms of Service Sections 5 and 7.

6. Report Errors

If you spot a serious AI error (especially safety-related), please let us know so we can improve prompts / model strategy: [email protected]