Search 280M+ Papers with AI
The smartest way to find, prioritize, read, and cite research — powered by AI and matched to your context.
Trusted by researchers from leading institutions
Product
Discover and Cite
A workspace built with AI agents to enhance your research workflow.
Literature Discovery
Conversational Agent
Upload files or ask complex questions. The AI understands your specific scope to find papers keyword search misses.
Search Across Databases
Search across 280M+ from hundreds of academic databases to surface the most relevant literature.
Relevance Ranking
AI reads, ranks, and filters papers based on how well they match your research context.
Literature Discovery Workflow
Discovered Papers
Citation Discovery
The @cite Tool
Import your draft and type @cite anywhere in your editor. The agent analyzes your paragraph to find the perfect supporting evidence.
Smart Context Analysis
The agent reads your draft—not just keywords—to suggest citations that actually support your arguments.
Verification Ready
Zero hallucinations. Every suggestion comes with complete citation, metadata, and source links ready for export.
Integrated Editor
Upload .docx or write directly in Chirpz. Manage your draft and discovered supporting papers in one view.
Citation Discovery Workflow
Your Editor
|
Discovered Papers
How it Works
Chirpz in Action
Zero friction. Total focus.
STEP 01
Ask or Upload
Type your research question or upload a file for analysis. The AI understands your context — no complex search syntax needed.
STEP 02
AI Scopes Research
The AI instantly extracts key research scopes and crafts smart search strategies — so your research starts targeted and fast.
Generating
STEP 03
Discover Papers
Relevant papers are discovered in real-time and ranked by relevance to your research scope.
STEP 04
Review & Cite
Browse, review, filter, and dive deep into discovered papers — all with rich, accurate metadata. Export citations with a single click.
Learning Conformal Abstention Policies for Adaptive Risk Management in Large Language and Vision-Language Models
Sina Tayebati, Divake Kumar, Nastaran Darabi, Dinithi Jayasuriya, Ranganath Krishnan, Amit Ranjan Trivedi
2025-02-08
arXiv (Cornell University)
Cited by 8
Type preprint
Open Access
Large Language and Vision-Language Models (LLMs/VLMs) are increasingly used in safety-critical applications, yet their opaque decision-making complicates risk assessment and reliability. Uncertainty quantification (UQ) helps assess prediction confidence and enables abstention when uncertainty is high. Conformal prediction (CP), a leading UQ method, provides statistical guarantees but relies on static thresholds, which fail to adapt to task complexity and evolving data distributions, leading to suboptimal trade-offs in accuracy, coverage, and informativeness. To address this, we propose learnable conformal abstention, integrating reinforcement learning (RL) with CP to optimize abstention thresholds dynamically. By treating CP thresholds as adaptive actions, our approach balances multiple objectives, minimizing prediction set size while maintaining reliable coverage. Extensive evaluations across diverse LLM/VLM benchmarks show our method outperforms Least Ambiguous Classifiers (LAC) and Adaptive Prediction Sets (APS), improving accuracy by up to 3.2%, boosting AUROC for hallucination detection by 22.19%, enhancing uncertainty-guided selective generation (AUARC) by 21.17%, and reducing calibration error by 70%-85%. These improvements hold across multiple models and datasets while consistently meeting the 90% coverage target, establishing our approach as a more effective and flexible solution for reliable decision-making in safety-critical applications. The code is available at: https://github.com/sinatayebati/vlm-uncertainty.
Make progress on your research, today
Start for free
No credit card required
Cancel anytime
FAQ
Frequently Asked Questions
Everything you need to know about Chirpz
Chirpz offers two solutions: the literature discovery agent finds papers based on your question in natural language or by analyzing your uploaded file, while the citation discovery agent locates sources for your drafted .docx or text.
Chirpz uses AI agents to search 280M+ papers across PubMed, arXiv, journals, and conferences all at once. Ask your research question naturally — no need for complex search syntax or multiple database searches.
We use the latest AI models. Different models perform better at different tasks. We constantly test and evaluate new models to ensure the highest quality results.
Chirpz reads each paper, then filters and ranks them by true relevance to your research. It understands the details of each paper and helps you cut through the noise.
Complete metadata (title, abstract, authors, date, journal, citations), AI-generated summaries, ready-to-export BibTeX citations, and PDF previews when available.
Absolutely. Every result is real and verified — zero hallucinations. All papers come from established databases with authentic, accurate metadata you can cite with confidence.
Yes! Start with 200 free credits per month - no credit card required. Pro plan unlocks unlimited requests and advanced features for power users and research teams.
We are committed to protecting your privacy. Your data is never used to train our AI models or shared with anyone else — your information always remains yours.